30th TABU Dag 2009: Accepted abstracts (90)

(view abstracts of keynote speakers; view complete schedule; return to the programme)

Typological Variation in Sentential Negation: Can it occur with one negative marker?
Nasser Al-Horais (Newcastle University)

In this paper, I examine the question of parametric variation in the syntactic expression of sentential negation. It is argued that typological variation in sentential negation is not only subject to a broad cross-linguistic variation (mainly via a single negative marker with a particular syntactic status), but it can also be found within a language, and even with one and the same negative marker. This negative marker is the Arabic negation marker laysa which has a rich variety of different patterns to express sentential negation. Three different syntactic functions can be found with laysa:

(i)	a negative copular verb, negating only present tense copular constructions with a zero-form copula:

(1)	a.	ʔal-ʔawlaad-u	*lays-uu*	sighaar-an
		the-boys-Nom	neg.m3p	little.Acc
		‘The boys are not little.’

(ii)	a negative auxiliary verb followed by a main verb in an imperfective form, or in perfective form obligatorily preceded by the preverbal marker QAD:

(2)	a.	*laysa*	Khalid-un	ya-ktubu	Ŝ-Ŝ iʕr-a
		neg.3ms	xalid-Nom	3m-write-Present	the-poetry-Acc
		“Khalid does not write poetry.”

(2)	b.	*laysa*	kul-u l-	madʕuwe-ina	*(QAD)	jaʔuu
		neg.3m	all-Nom	the-invited-people-Gen	perf marker	came.3mp
		“All of the invited people have not come.”

(iii)

a negative particle negating a structure derived by so-called Bare Argument Ellipsis:

(3)	a.	aljamiʕat-u	l-jadidat-u	sa-tubna	fii l-Gahirat-i	wa	*laysa*	fii	l-askaadariat-i
		the-university-Nom	the-new-Nom	will-be-built	in the-Cairo-Gen	and	neg	in	Alexandria-Gen
		“The new university will be built in Cairo and not in Alexandria.”

(3)	b.	Gabal-tu	Khalid-an	wa	*laysa*	zaid-an
		met-I.3s	xalid-Acc	and	neg	Zaid-Acc
		“I met Khalid and not Zaid.”

(3)	c.	kaana	fii d-dar-i	xalid-un	wa	*laysa*	zaid-un
		was-past.3ms	in the-house-Gen	Khalid-Nom	and	neg	Zaid-Nom
		“Khalid was in the house and not Zaid.”

Besides shedding new light on the variation in the expression of sentential negation, it is worth noting that laysa, with these three different syntactic functions, is a negation head or just a particle. In (1) and (2), where the negation is inflected and requires the presence of a particular tense (present) or modal particle (QAD), the negation is a head, whereas in the remaining example (3), where laysa negates Bare Argument Ellipsis constructions, the negation is not a head as it lacks inflection, and does not require the presence of a particular tense. This strongly imposes a tight relationship between negative heads and tense as proposed by Fassi Fehri (1993), Haegeman (1995), Zanuttini (1997) and Collberg & Håkansson (1999), among others.

Developmental Dyslexia in Bahasa Indonesia: Developing a Screening Test
Harwintha Anjarningsih¹ and Ria de Bleser²
(¹University of Groningen; ²Potsdam University)

Aimed at filling the vacuum in reading research in Bahasa Indonesia, this study has three objectives: to find out whether phonological awareness correlates with reading success in Bahasa Indonesia and if it does, to identify the level of phonological awareness that correlates best with normal reading in Bahasa Indonesia: whether it is at the phoneme, rime, or syllable level; to point out how the lexical and sub-lexical route is employed in normal reading in Bahasa Indonesia; and to identify reading performance that may suggest the occurrence of impaired reading in Bahasa Indonesia which is possibly displayed by the children with the lowest reading performance. Ten typically developing third-grade students participated in the study by doing tasks divided into two parts: phonological awareness part and reading part. Error patterns of the ten students are analysed for the error analysis and for the analysis on the relationship between psycholinguistic variables and reading performance. Five of those students were recorded during the reading part and the data were used in the latencies analysis.

Syllable level phonological awareness was found to be correlated well with reading success and followed by phoneme level phonological awareness. Beginning reading in Bahasa Indonesia seems to be characterized by the utilization of the sub-lexical route and skilled reading by lexical route. The development of the lexical route depends on the development of the sub-lexical route which is influenced by the complexity of the Grapheme Phoneme Correspondence. The lexical route for words with simple graphemes is developed first, the lexical route for words with consonant clusters developed later, and the last to develop is the lexical route for words with digraphs. Furthermore, Late AoA, imageability, syllable number, and resemblance to words influence the lexical processing of the stimuli in the test. Impaired reading seems to be characterized by less developed sub-lexical route as evidenced by difficulties with words containing consonant clusters and digraphs and longer latencies for reading aloud words and pseudowords.

Although further study with a bigger sample and improvements to the test are needed, the result of this study is a good start in developing a developmental dyslexia screening-test and enhancing reading research in Bahasa Indonesia

The Locational Predicative Possession Construction in Russian
Aysa Arylova (University of Groningen)

The topic of this talk is the Russian predicative possession construction illustrated in (1). The construction involves a PP possessor, a Nominative possessum, and an existential verb, mainly byt’ ‘be’:

(1)	U menja	est’	sobak-a.
	at I.GEN	is	dog.F-NOM.SG
	‘I have a dog.’

The sentence in (1) is an instance of the locational strategy of encoding predicative possession, according to the typological classification of Stassen (2005). The predominant analysis of the structure of the Russian locational predicative possession construction (LPPC) is a PP/Small Clause analysis in the style of Freeze (1992), Den Dikken (1998), Harley (2002). The small clause containing the possessor and the possessum is headed either by the preposition u ‘at’, or by some kind of abstract head X. The verb byt’ ‘be’ in a small clause analysis can be viewed either as a functional element in the IP domain or a one-place predicate taking the small clause as its argument. One of the problems for the PP/Small Clause analysis is that it cannot account for the differences between the LPPC and unambiguously small clause constructions, such as constructions with AP/NP predicates. The differences include the distribution of the Genitive of Negation, the behavior of the verb byt’ ‘be’, and the possibility of embedding under control and raising predicates.

In this talk I argue for an alternative view of the structure of the Russian LPPC: a double unaccusative analysis, where the existential byt’ ‘be’ is a lexical predicate taking the possessor and the possessum as its two internal arguments (cf. Chvany 1975). I discuss the details of the double unaccusative analysis, such as the ordering of the arguments in the LPPC, which can be established with the help of Harley’s (2002) quantifier binding diagnostic. Finally, I suggest that a double unaccusative analysis of the Russian LPPC can more easily incorporate the properties differentiating the LPPC from AP/NP predicate constructions.

Explaining knowledge change through writing
Veerle Baaijen¹, David Galbraith² and Kees de Glopper¹
(¹University of Groningen; ²Staffordshire University)

There is general agreement within research on writing that writing involves more than translating preconceived ideas into written text. Instead, it is assumed that writing also involves the development and refinement of ideas in the course of writing. This has led to theories of writing as a problem solving activity (Flower & Hayes, 1980; Galbraith & Torrance, 1999a). Such theories also assume that writing should be a valuable tool for learning (Ackerman, 1993). However, until now, there is no conclusive evidence whether and how writing contributes to learning (Klein, 1999).

Generally speaking there are two approaches within writing research to the question of how writing facilitates learning. Some researchers focus on the effect of writing on knowledge change and others focus exclusively on the processes involved in writing. The aim of this research is to bring those two approaches together and to test how processes of writing relate to knowledge change. Therefore, this first experiment focuses on different writing conditions (Galbraith, 1999) in which knowledge change appears to be happening and simultaneously investigates the underlying writing processes.

In this study 84 students, half of whom were low self-monitors and half of whom were high self-monitors, were asked to plan and write an article for the university newspaper about a current affairs topic. In order to assess the extent to which writers develop their ideas during writing participants were asked to list their ideas, to rate the importance of these ideas and to indicate how much they felt they knew about the topic, both before and after writing. Then, the writing task was divided into two phases. First, half the participants were asked to plan their essay by making an organized outline and the other half of the participants was asked to write down a single sentence summing up their overall opinion. During the second phase, participants had 30 minutes to write their article for the university newspaper. During this phase Inputlog was used to record the writing processes while writing.

Analysis of the results is currently in progress. It will include analysis of differences between high and low self- monitors on the following dependent variables: (i) the extent to which knowledge is transformed during writing (ii) the extent to which the different planning tasks influence the knowledge change and (iii) the extent to which writers pause within and between different text units and before and after introducing new content.

Information Structure and Indirect Anaphora Resolution
Joaquim Barbosa, Idalina Ferreira and Fátima Silva (Universidade do Porto)

Nominal indirect anaphora is commonly described as a text/discourse phenomenon deeply related with reference processing, which contributes simultaneously to the introduction of new discourse referents and to the maintenance of information flow. Indirect anaphora is, for this reason, clearly connected to information structure and to the distinction between topic and focus, in the way that is a process of dependent interpretation between an anaphoric expression and its anchor.

Erkü & Gundel (1987:543) have already pointed out the correlation between anaphora resolution and the sentence topic, as shown in (1):

(1)	We stopped for drinks at the New York Hilton before going to the Thai restaurant. The waitress was from Bangkok.

These authors claim that, in sentences like (1), the anaphoric expression is more likely to be interpreted anaphorically referring to the Hilton than to the Thai restaurant, in spite of the fact that world knowledge and the immediacy between the NP Thai restaurant and the anaphoric expression might suggest otherwise.

Based on this problem (see also Hajičová, Partee & Sgall 1998; Komagata 2003), several tests were applied to a corpus of European Portuguese (i) to deal with the correlation between nominal indirect anaphora resolution in complex sentences and the information structure of those sentences, and (ii) to discuss whether modified anaphoric expressions occurring in such contexts, as in (2), may contribute to its better understanding.

(2)	We stopped for drinks at the New York Hilton before going to the Thai restaurant. The nice waitress was from Bangkok.

Based on some authors who claim that nominal indirect anaphora has a hybrid discursive status as a thematic/rhematic structure (Schwarz 2000), we consider that sentences like (1) and (2) may have tripartite structures in which the antecedent of an anaphoric expression is always in the main clause, consisting of the focus of the previous sentence. Therefore, we claim that the resolution of this type of anaphora is both related to topic and focus.

References

Erkü, Feride and Gundel, Jeanette (1987) – The pragmatics of indirect anaphors. In Verschuren, J. & Bertuccelli-Papi, M. (eds.) – The Pragmatic Perspective: selected papers from the 1985 International Pragmatics Conference. Amsterdam: John Benjamin’s.

Hajičová, E., Partee, B., and Sgall, P. (1998) – Topic-Focus Articulation, Tripartite Structures, and Semantic Content. Dordrecht: Kluwer.

Komagata, N. (2003) – Information Structure in Subordinate and Subordinate-Like Clauses. Journal of Logic, Language and Information, 12.

Schwarz, M. (2000) – Indireckte Anaphern in Text. Tubingen: Verlag.

The grammaticalization of Mainland Scandinavian MAYBE
Karin Beijering (University of Groningen)

This paper is concerned with epistemic adverbs of the MAYBE type (deriving from a modal auxiliary ‘can’ or ‘may’ + a verb meaning ‘happen’ or ‘be’) in Mainland Scandinavian. Swedish kanske is an epistemic adverb with special syntactic status (cf. Andréasson 2002), because it does not always adhere to standard Swedish word order in declarative main clauses. That is, kanske may violate the Verb Second (V2) principle, according to which the finite verb must always be the second constituent in a declarative main clause. As the examples in (1) show, both V2 and non V2 word orders are possible in the declarative main clause "Maybe Olle did not sleep last night" (cf. Andréasson 2002:1):

(1)	a.	Olle	har	*kanske*	inte	sovit	inatt
		Olle	has	maybe	not	slept	last.night

(1)	b.	Olle	*kanske*	inte	har	sovit	inatt
		Olle	maybe	not	has	slept	last.night

(1)	c.	*Kanske*	Olle	inte	har	sovit	inatt
		Maybe	Olle	not	has	slept	last.night

There are three different clause types containing kanske. The first type is a V2 clause in which MAYBE can be any constituent except the second one ((1a)); the second type is a non V2 clause which has MAYBE as its second constituent, that is, it occupies the position of the finite verb ((1b)); and the third type is a non V2 clause with MAYBE as its first constituent followed by a clause which has subordinate word order ((1c)).

The number of syntactic positions for a linguistic item is a useful criterion in determining the degree of grammaticalization. Since grammaticalization generally involves syntactic fixation, a decrease in syntactic freedom would imply advanced grammaticalization of MAYBE. Corpus data of MAYBE is analyzed and compared for Danish, Norwegian and Swedish in order to determine the degree of grammaticalization for MAYBE in the Mainland Scandinavian languages. When comparing the degree of multipositionality of MAYBE, one might find different stages of grammaticalization. That is, MAYBE can be more grammaticalized (=less syntactic freedom) in one language than in another (=more syntactic freedom).

I will outline the etymology, development and grammaticalization path of Mainland Scandinavian MAYBE and show that the coexistence of three different clause types is due to different stages within one and the same grammaticalization process instead of separate grammaticalization paths (cf. Andréasson 2002: 40-46).

Reference

Andréasson, Maia. 2002. Kanske- en vilde i satsschemat. (=Meddelanden från Institutionen för svenska språket 41). Göteborg: Institutionen för Svenska Språket.

The structure of lexical cohesion
Ildikó Berzlánovich and Gisela Redeker (University of Groningen)

The structure of lexical cohesion in texts can be described in terms of lexical chains, i.e., sequences of related words (Barzilay & Elhadad 1999). Alternatively, it can be conceptualized as a graph structure (Erkan & Radev 2004), allowing multiple relations among the lexical items and ignoring directionality. The graph-based model provides a much richer representation of lexical cohesion than approaches that compute lexical chains, and thus offers a better chance of capturing the total impact of lexical cohesive links on the textual organization. In this talk, I will illustrate such a rich representation and show how it can be related to discourse structure. The lexical cohesion in a text is conceptualized as an undirected weighted multi-graph, where the vertices are the elementary discourse units and the edges are the lexical cohesive links. Discourse structure is described in terms of genre-specific 'moves' expressed in (usually complex) discourse units that are located high in the rhetorical structure (RST; Mann & Thompson 1988) tree. By comparing the centrality of a discourse unit in the cohesion network with its centrality in the discourse structure tree, we can assess the alignment between cohesion and coherence. In information-oriented texts, e.g. encyclopedia entries, this alignment should be very close, whereas cohesion should not or only loosely align with discourse structure in texts with a dominant intentional structure, e.g. fundraising letters. This has been shown in Author et al. (2008).

In that earlier study, we used a simple version of a cohesion analysis yielding a graph structure, calculating centrality in terms of the number of lexical links of discourse units with other units in the text. I will now present a refined measure, where the lexical cohesive links are weighted by (i) the textual distance and (ii) the type of relation between the lexical items. The centrality of an EDU is then measured as the weighted sum of the edges connecting it to other EDUs in the text.

With these refinements in our measurement, we expect cohesion-based centrality to align even more closely with the discourse structure in encyclopedia texts, while no increased alignment is expected for fundraising letters. Initial tests show that this seems to be the case (further tests will be conducted in the next months).

Corpus Analysis and Text Types in Literary Journals. Observations and Explorations of the AAC-Austrian Academy Corpus
Hanno Biber (Austrian Academy of Sciences)

The "AAC-Austrian Academy Corpus" is the corpus research unit of the Austrian Academy of Sciences in Vienna and concerned with establishing and exploring large electronic text corpora and conducting scholarly research in the fields of digital text corpora. In this paper the problems and methodological questions concerning the annotation of different text types within the text corpus will be addressed. The relations between various approaches will be discussed and examples of the annotation procedures and their analytical concepts will be given.

The texts integrated into the collections of the AAC stemming from the last 150 years are predominantly German language texts of historical and cultural significance. The AAC has collected thousands of texts representing an astonishing range of different text types. Among the sources, which cover manifold domains and genres, there are literary journals, newspapers, novels, dramas, poems, advertisements, essays, travel accounts, cookbooks, pamphlets, political speeches as well as plenty of scientific, legal, and religious texts, to name just a few.

In this paper particular emphasis will be given to the corpus based methods for the analysis of text types within literary journals and similar text type containers. The AAC can provide a great number of reliable resources and interesting corpus based approaches for investigations into the linguistic and textual properties of these texts. Well beyond 400 million running words of text have already been scanned, converted into machine-readable text and carefully annotated. The annotation and mark-up schemes of the AAC are based upon XML related standards. Most of the projects in the first phase of corpus build-up were dealing with issues related to the application of basic structural mark-up and selected thematic features of the texts. In the next phase of application development the AAC will further intensify its efforts towards deeper thematic annotations, thus exploring problems of linguistics and textual scholarship by means of experimental as well as exploratory analyses.

Reference

http://www.aac.ac.at

Implicit Linguistics
Antal van den Bosch (Tilburg University)

Natural language processing (NLP) models and systems typically employ abstract linguistic representations (syntactic, semantic, or pragmatic) as intermediate languages. It is scientifically mandatory to ask ourselves the question whether we could do without them. We know that any invented intermediate structure is always implicitly encoded somehow in the words at the surface, and the way they are ordered. Can we somehow make use of this implicit encoding when performing NLP tasks?

Classes of NLP tasks in which this question can be investigated in extremo are processes in which form is mapped to form, i.e., in which neither the input nor the output contains abstract elements to begin with. Examples are spelling correction (converting a corrupted text to a clean, intended form of the same text); machine translation (converting a text in one language to a text in another language carrying approximately the same meaning); paraphrasing and summarization (which are like translation, except the language remains the same). The text-to-text scheme also fits communicative turn-based language use where one utterance is followed by another, such as question answering and dialogue. The scheme even generalizes to information retrieval, where short queries are mapped onto ranked lists of documents.

Interestingly, this list of examples covers a major part of NLP's current application areas. Even more interesting is the fact that many successful applications in all of these areas (in fact nearly all of the most successful applications) make no use of explicit abstract linguistic representations.

In this contribution I propose to investigate the possibility of NLP with implicit linguistics by way of memory-based learning (MBL). MBL is a class of machine learning algorithms that implements De Saussure's analogy principle: it makes use of analogical proportions to map unseen input to output, by making use of similarity-based reasoning on memorized examples of input-output mappings. The simplicity of the approach makes it amenable to efficient implementation in computer programs. Most of the successful implicit approaches to the aforementioned text-to-text task in fact make use of some variant of this memory-based or similarity-based approach, often in a stochastic variant.

I argue that MBL approaches do not operate without linguistics, because they successfully accomplish tasks that involve morphosyntactic or semantic knowledge. I discuss the question where the linguistics is in these models, as it is not explicitly present. I argue that the linguistics happens in the similarity-based reasoning, i.e., in the processing itself.

The interpretation of Prepositional Phrases as Arguments and Adjuncts in L2 acquisition
Rita Calabrese (University of Salerno)

Early research on second language (L2) acquisition pointed out that semantically transparent properties of a given target language (TL) are easier to be acquired/learnt than more abstract syntactic properties which do not directly correspond to semantic representations (Kellerman 1987). This assumption suggests that aspects of grammar that require not only syntactic knowledge, but also the ability to integrate syntactic knowledge and knowledge from other domains of language is hardly to be acquired by L2 learners (Sorace 2005). In particular, features that belong to the interface between syntax and lexicon may be vulnerable to variability and even deviation with respect to the TL constraints.

Starting from the assumption that L2 learners employ qualitatively different parsing strategies from native speakers at the semantic-syntactic level, I have carried out a study on a sample of L2 data to investigate learners' difficulty to automatically integrate phrase structure and lexical-semantic information by focusing on argument structure and complementation in their interlanguage. Given the assumption that the notion of argument is a notion at the interface of the syntactic and semantic levels determining the valency of a verb and its subcategorization frame, so presumably fuzzy categories like `argument' and `adjunct' may lead to frequent misinterpretation and misuse of prepositional phrase (PP) attachments and verb complementation by L2 learners.

To verify this hypothesis a small corpus of written productions from Italian university students has been automatically parsed by using the VISL applications/language tools which can provide both syntactic and semantic information on a given constituent structure. Every PPs occurring in the annotated corpus have been then searched for and classified by: 1. head word and 2. syntactic function (argument and adjunct). Following the procedure adopted in a recent study on the automatic detection and extraction of arguments and adjuncts from a parsed corpus of English native speakers (Merlo-Ferrer 2006), I have matched corpus-based evidence and the linguistic diagnostics (e.g. head dependency and optionality) generally used to decide whether a PP is an argument or an adjunct to find out similarities and\or differences in L1\L2 syntactic constructions.

The receptive-productive gap in L2 vocabulary development: a dynamic analysis
Tal Caspi (University of Groningen)

Vocabulary knowledge is often described as a continuum of receptive-to-productive abilities (Read, 2000), yet several studies have noted a discrepancy between receptive and productive knowledge levels, which remains consistent despite overall increases in proficiency (Melka, 1997; Meara & Schmitt, 1997, Laufer & Paribakht 1998; Fitzpatrick et al., 2008). The receptive-productive vocabulary knowledge gap is the focus of the current study, which investigates it from a Dynamic Systems Theory (DST) perspective.

DST is a theory of change and development which is broadly applied in the natural, social, behavioral and cognitive sciences. It is a framework for analyzing complex systems, defined as networks of interacting components that are embedded in environments consisting of similar structures, as functions of the interactions between their components and with the environment. The dynamic approach to language extends this perspective to linguistic development in both communities and individual users (Komarowa & Nowak 2001; de Bot 2007).

The main methods of investigating development applied by the dynamic approach are variability analysis and mathematical modeling (van Geert, 1994; van Dijk, 2003). The two methods are complementary: variability analysis supplements the general trends obtained by traditional statistical techniques with temporal patterns of development and interaction, while modeling configures findings as parameters in functions that simulate development. So far, variability analysis of L1 and L2 data has revealed that a typical dynamic pattern of interaction known as the precursor model (van Geert, 2003) has explanatory power with regard to the differential development of various components of linguistic proficiency (Robinson & Mervis, 1998; Verspoor, et al 2008).

The current study assessed growth along a receptive-productive continuum of word knowledge levels that ranges from least to most productive: recognition, recall, controlled production and free production (Laufer & Nation 1995; Laufer et al 2004). The study’s main questions are whether the receptive-productive gap is stable, and whether it can be considered as resulting from a dynamic precursor interaction between the continuum levels. The study collected longitudinal L2 data from four case studies, analyzed its growth trends and variability, and configured the findings in a model of dynamic interaction between knowledge levels. The findings indicate that the gap is robust, and that the precursor model can account for its presence. The presentation will include an introduction to DST in the context of L2 acquisition, an explanation of the methodology and main findings, and a discussion on the implications of applying DST to L2 vocabulary knowledge.

References

de Bot, C. L. J. (2007). Dynamic systems theory, life-span development and language attrition. In B. Köpke, M.S. Schmid, M. Keijzer, & S. Dostert (Eds.), Language attrition: theoretical perspectives (pp. 53-68). Amsterdam: John Benjamins.

Fitzpatrick, T., Al-Qarni, I., and Meara, P. (2008). Intensive vocabulary testing: a case study. Language learning journal, 36, 239-248.

Komarova, N. L. and Nowak, M. A. (2001). Towards an evolutionary theory of language. Trends in cognitive sciences, 5, 288-295.

Laufer, B., Elder, C., Hill, K., and Congdon, P. (2004). Size and strength: do we need both to measure vocabulary knowledge? Language testing, 21, 202-226.

Laufer, B. and Nation, P. (1995). Vocabulary size and use: lexical richness in L2 written production. applied linguistics, 16, 307-322.

Laufer, B. and Paribakht, T. S. (1998). The relationship between passive and active vocabularies: Effects of language learning context. Language learning, 48, 365-391.

Meara, P. & Schmitt, N (1997). Towards a new approach to modeling vocabulary acquisition. In N. Schmitt & M. McCarthy (eds.), Vocabulary: description, acquisition and pedagogy (pp. 109-121). Cambridge: Cambridge Univ. Press.

Melka, F. (1997). Receptive vs. productive aspects of vocabulary. In N. Schmitt & M. McCarthy (Eds.), Vocabulary: description, acquisition and pedagogy (pp. 84-102). Cambridge: Cambridge University Press.

Read, J. (2000). Assessing vocabulary. Cambridge: Cambridge Univ. Press.

Robinson, B. F and Mervis, C. B (1998). Disentangling early language development: modeling lexical and grammatical acquisition using an extension of case-study methodology. Developmental psychology, 34, 363-375.

van Geert, P. (2003). Dynamic systems approaches and modeling of developmental processes. In J.Valsiner & K. J. Conolly (Eds.), Handbook of developmental psychology (pp. 640-672). London: Sage.

van Geert, P. (2008). A dynamic systems approach to the study of L1 and L2 acquisition: an introduction. The modern language journal, 92, 179-199.

Verspoor, M. H., Lowie, W. M., and van Dijk, M. (2008). Variability in second language development from a dynamic systems perspective. The modern language journal, 92, 214-231.

Language Acquisition and Learning Theory: Some Misconceptions about Learnability
Çağrı Çöltekin (University of Groningen)

Children's apparent ease of acquiring languages in the face of the complexity of the human languages and arguably inadequate input that they are exposed to in the course of acquisition has been one of the puzzles for the theories of language. Since we can not (yet) study the process of language acquisition directly, researchers propose theories or models of language acquisition and test their predictions in the light of empirical data about how children learn languages. As well as accounting for the empirical data, the models of human language acquisition have to account for certain formal theoretical results. In this talk, I will argue that some of the widely accepted theories of language acquisition are based on misguided---or, at best, highly restrictive--- interpretations of formal learning theory results. Besides, the developments in machine learning provided many theoretical and practical results on learnability, which are generally overlooked by researchers working on language acquisition.

This talk will focus on two of these misconceptions. First, I will discuss the use of Gold's theorem (Gold, 1967) for claims about language acquisition, and argue against arriving at strong conclusions on language acquisition based on it. Gold's work has been one of the most influential studies in learning theory, however, one should consider its application to the human language acquisition problem cautiously. Among other assumptions, Gold's theorem is based on the assumption that the learning succeeds when the target grammar is `identified'. I argue that, a different learning criterion such as `PAC learning' (Valiant, 1984) is more suitable for modeling language acquisition.

Second, some theories of language acquisition are formed with the assumption that the size of hypothesis space is the determining factor for learnability. However, relatively recent research in the learning theory showed that the learnability problem is more complex than it was assumed. Rather than the size of hypothesis space, other metrics, such as `VC dimension' (Vapnik and Chervonenkis, 1971), determine the learnability of the problem. In other words, learning problems formulated as a search through infinite number of hypotheses are not necessarily `unlearnable'.

In this talk, I will summarize the above arguments and provide a simple linguistic example, where a supposedly unlearnable problem becomes learnable with an alternative parametrization, using techniques from machine learning.

References

Gold, M. E. (1967) "Language Identification in the Limit." Information and Control, 10:447-–474

Vapnik, V. & Chervonenkis, A. (1971) "On the uniform convergence of relative frequencies of events to their probabilities." Theory of Probability and its Applications, 16(2):264--280

Valiant, L. (1984) "A theory of the learnable." Communications of the ACM, 27(11):1134--1142

Light Verbs as Predicates
Inês Duarte (University of Lisbon)

Since Jespersen’s remark on the peculiar behaviour of ‘light verbs’, several attempts to characterize them have been proposed in the literature. Such attempts fall under three kinds of approaches. The functional-like head approach claims light verbs are devoided of lexical meaning and behave like functional heads bearing tense and agreement features (Gross 1981, Cattell 1984, Grimshaw & Mester 1988, a o.). The auxiliary-like approach argues that light verbs are auxiliaries with aspectual features (Hook 1974, 1991, Abeillé, Godard & Sag 1998, a.o.). The predicate-like approach claims that light verbs are a specific subclass of verbs that play a relevant role in the predication (see Rosen 1990, Alsina 1996, Butt & Geuder 2001, Butt 2003, Samek-Lodovici 2003, a.o.)

In this paper, we will argue for the third approach on the basis of evidence from European Portuguese. We will concentrate on two properties of light verbs when they combine with deverbal nouns.

1. Light verbs always select for complements with the proto θ-role THEME (Samek-Lodovici 2003) and for an external argument. In most cases, the external argument bears the proto θ-role selected for by the corresponding main verb (1).

(1)	a.	A Maria deu um livro ao pai.	(main verb dar)
		the Maria-CAUSE gave a book to-the father
	b.	A Maria deu uma contribuição importante ao debate.	(light verb dar)
		the Maria-CAUSE gave a contribution important to-the debate
	c.	A Maria deu um salto.	(light verb dar)
		the Maria-CAUSE gave a jump

When a deverbal noun with a CAUSE external argument combines with a light verb, this argument must fit the thematic structure of the light verb. This is what happens with the light verb ter (have), which inherits an external θ-role from the corresponding main verb indicating the location of the THEME internal argument, which we will label LOC (2).

(2)	a.	A Maria saltou lindamente.
		the Maria-CAUSE jumped beautifully
	b.	A Maria teve um salto lindo.
		the Maria-LOC had a beautiful jump

2. Light verbs present alternances like the ones shown by main verbs. The light verb fazer (make/do) shows the causative-inchoative alternation (3).

(3)	a.	O cirurgião fez uma operação complicada (à Maria).
		the surgeon-CAUSE made an operation difficult (to-the Maria)
	b.	A Maria fez uma operação complicada.
		the Maria-THEME made an operation difficult

Whether one accounts for these properties in terms of lexical operations on thematic structures (Samek-Lodovici 2003) or in terms of event structure (Butt 2003), they show that light verbs behave like predicates.

References

Abeillé, A., D. Godard & I. Sag (1998). Two Kinds of Composition in French Complex predicates. In Hinrichs, E., A. Kathol & T. Nakazawa (orgs.), Complex Predicates in Nonderivational Syntax. Syntax and Semantics 30. San Diego, Academic Press

Alsina, A. (1996). The Role of Argument Structure in Grammar. Stanford, Califórnia, CSLI Publications.

Butt, M. & W. Geuder (2001). On the (Semi)Lexical Status of Light Verbs. In Corver, N. & H. van Riemsdijk (eds.), Semi-lexical Categories: On the Content of function words and the function of content words. Berlim, Mouton de Gruyter. Pp. 323-370.

Butt, M. (2003). The Light Verb Jungle. Harvard Working Papers in Linguistics, Vol 9: 1-49.

Cattell, R. (1984). Composite Predicates in English. Syntax and Semantics 17. Sydney, Academic Press.

Grimshaw, J. & A. Mester (1988). Light Verbs and q-Marking. Linguistic Inquiry, 19-2: 205-232.

Gross, M. (1981). Les Bases Empiriques de la Notion de Prédicat Sémantique. Langages, 63: 7-52.

Hook, P. E. (1974). The Compound Verb in Hindi. Center for South and Southeast Asian Studies, University of Michigan.

Rosen, S. (1990). Argument Structure and Complex Predicates. Nova Iorque, Garland.

Samek-Lodovici, V. (2003). The Internal Structure of Arguments and its Role in Complex Predicate Formation. Natural Language & Linguistic Theory, 21: 835-881.

Attitudes about taste
Sergio Raul Duarte¹ and Janneke Huitink²
(¹University of Groningen; ²University of Frankfurt)

Data While (1a) is typically understood as reporting the taste of the speaker, (1b) is naturally understood as meaning that licorice tastes good to Chris (Lasersohn 2005). As Stephenson (2007) points out, the relevant experiencer need not be linked to the subject of the matrix attitude, but may also be linked to some salient individual:

(1)	a.	Licorice is tasty.
	b.	Chris believes that licorice is tasty.
	c.	Chris believes the new cat food is tasty, because the cat has eaten a lot of it.

But not all attitude predicates are created equal. Subjective attitude verbs obligatorily shift the relevant experiencer to the attitude holder (Chris):

(2)	a.	Chris finds/regards/judges/considers licorice tasty.
	b.	??Chris finds/regards/judges/considers the new cat food tasty, because the cat has eaten a lot of it.

An evidential analysis We propose that the relevant observations are explained by the evidential properties of the attitude verbs. Following Lasersohn and Stephenson, we assume that the meaning of sentences like `Licorice is tasty' depends on a judge-parameter which is set by the context:

(3)	[[tasty]]^c;w = λx:x tastes good in w to j_c

There are several possible resolutions of j_c and `Licorice is tasty' floats several propositions accordingly (though by in unembedded cases, j_c is per default set to the speaker). In attitude contexts, the evidential profile of the attitude verb helps to figure out which reading is intended. Subjective attitude verbs like find presuppose that the attitude holder has direct evidence. This is corroborated by the fact that `Chris finds licorice tasty' implies that Chris has actually tried the licorice, while this is not required for `Chris believes licorice is tasty'.

(4)	a.	[[find]]^c;w = λpλx. in all worlds w' compatible with x's beliefs in w where x has direct experience of p: [[p]]^c;w' = 1.
	b.	[[believe]]^c;w= λpλx. in all worlds w' compatible with x's beliefs in w where x has indirect experience of p: [[p]]^c;w' = 1.

In principle, the embedded predicate of personal taste has several meanings, including `tasty for the speaker', `tasty for Chris, but the direct evidence presupposition rules out the readings other than `tasty for Chris', because Chris cannot have direct evidence for the truth of the proposition that licorice is tasty to other people.

The verb believe carries an indirect evidence presupposition, which allows j_c to be resolved to someone who is not Chris or the speaker. This is witnessed by reports of the tastes of a salient experiencer, which cannot depend on direct evidence, and need a verb like believe to mark this. For example, `I believe the new brand of cat food is tasty' is better than `The new brand of cat food is tasty'.

References

Lasersohn, Peter. 2005. Context dependence, disagreement, and predicates of personal tase. Linguistics and Philosophy 28(6): 643{686.

Stephenson, Tamina. 2007. Judge dependence, epistemic modals, and predicates of personal taste. Linguistics and Philosophy 30(4): 487{525.

Using discourse information in sentiment polarity classification
Sergio Raul Duarte and Gisela Redeker (University of Groningen)

Sentiment polarity classification consists in the identification of evaluative information in human authored opinionating documents, e.g. reviews. The classification is determined by a quantitative value indicating the polarity and strength of the writer's opinion towards the item reviewed (i.e. how much a reviewer likes an item), as gleaned from the use of evaluative expressions. Common applications are the development of recommendation systems, the detection of sensitive content or antagonistic language, the mining of sentiment for business and government, among others.

One of the main approaches in the literature to perform sentiment polarity classification consists in summing up the semantic values of evaluative keywords in the text. Voll and Taboada (2007) present a sentiment polarity classifier in which the keywords are adjectives, verbs, nouns and adverbs listed in a prebuilt dictionary in which the semantic value is specified. Their classifier also considers negation, intensifiers and modality in calculating the semantic polarity value of the keywords. The accuracy obtained by their system is 76.1% on a corpus of 400 reviews (the SFU corpus).

An important drawback of this approach is that the identification of evaluative content is very narrow, since the analysis is based on isolated words, and no information is used to determine if the keywords refer to the writer's attitude toward the item reviewed or to irrelevant attitudinal information, for instance, a movie character's attitudes in a film review.

In this paper we address this problem by analyzing the use of appraisal elements in review texts and the distribution of appraisal types with respect to RST functions. Particularly, we employ the SFU corpus to identify the appraisal types that are more frequently used to express relevant evaluative information and those attitude types that may refer to off-topic information, considering also their location throughout the text. We then identify the RST functions most frequently associated with these attitude devices. Using this information, we propose strategies to assign weights to the sentiment value of the keywords according to the RST function in which the word occurs and the location of the word in the text, since some RST functions have a greater likelihood to be associated with the appraisal devices that hold relevant sentiment information. Initial results suggest that these strategies are effective, since the accuracy of the classifier is increased to 82.2% for the SFU corpus.

Reference

Voll, K. and M. Taboada (2007) Not All Words are Created Equal: Extracting Semantic Orientation as a Function of Adjective Relevance. In Proceedings of the 20th Australian Joint Conference on Artificial Intelligence. Gold Coast, Australia. December 2007. pp. 337-346

A Cognitive Technology Project: Semantically Sensitive Natural Language Processing
Matthew Elliott, Draque Thompson and Nicholas Davis (Case Western Reserve University)

The goal of the present work is to develop a semantically sensitive natural language processer. This technology can be applied to robotics by enabling an agent to express and use semantics.

This parser is based on the principles of Stemmatic Syntax, a theoretical model of language created by Dr. Per Aage Brandt that breaks down language into a canonical set of semantically meaningful nodes headed by the verb. Contrary to many parsing approaches, this system does not add words to a string based on linear order, but rather it finds local meaning structures, analyzes the relationship between the words, and finally assigns each word to a nodal position based on the schematic meaning inherent in syntax. Systematically combining these nodes accounts for linguistic embedding and recursion by creating a ‘stemmatic tree,’ which reflects local and global meaning. Stemmatic philosophy proposes that, cognitively, language has three-dimensional thought semantics (i.e. spatial cognition, cognitive maps, representations), two-dimensional semio-syntactic structure (stemmatic ‘trees’), and one-dimensional phonetic expression (linear text, speech). In writing or speaking, cognitive operations go from 3D→2D→1D, while reading (or listening) implies the opposite. Stemmatic trees are the intermediary, a bridge, between form and content.

Potential knowledge can be extracted from individual sentences and compiled to create a stemmatic network organized by grouping lexemes with their nodal values and syntagmatic similarities. An example may elucidate this concept. Take for instance this sentence, ‘Luz went back to Pordonone to open a hospital,' (Hemingway, A Very Short Story). On a generic level, the system can deduce that a subject can ‘go back’ to a location. More specifically, the subject, Luz, can ‘go back’ to the location Pordonone, etc. The verbs ‘go’ and ‘open’ are defined by the co-text. As the system stemmatizes more information, it builds up a wide depth of knowledge potentialities and several patterns of affinity emerge, eventually crossing statistical thresholds to become ‘factual’ knowledge for the system.

The presentation will first introduce the principles of stemmatic semio-syntax and next describe the language processor based on this paradigm. Finally, we will discuss how this technology can be applied to cognitive robotics for empirical verification. To test the semantic efficiency of this approach, the agent will be made to ‘think’ and ‘communicate’ in terms of stemmatics. Again, based on the three dimensional approach, the agent’s sensory input is 1D, then structured in a 2D stemmatic tree, and finally represented as a 3D cognitive map.

References

Brandt, Per Aage, 1973, L’analyse phrastique. Introduction à la grammatique, Bruxelles: AIMAV

Brandt, Per Aage, 2004, Spaces, Domains, and Meaning. Essays in Cognitive Semiotics, Bern: Peter Lang Verlag

Stemmatic Syntax (2008) from Web site: www.stemma.pbwiki.com

Interactions between elderly and caregivers
Agnes Engbersen (University of Groningen)

Conversation Analytic research on the relationships between verbal, non verbal and contextual aspects of language use in (care) interactions between elderly people and caregivers in institutional settings has not been given much attention yet (Heinemann 2006). In a recently started PhD research project I will focus on the following question from a multimodal perspective: How do elderly and caregivers manage their interactions during care activities?

In homes for the elderly many everyday care activities consist of routinized actions. The routine character is reflected in the sequential organization of both verbal and physical actions. Activities are enacted within a structured framework of related actions, a Situated Activity System (SAS; Goffman, 1981; Goodwin, 2000; see Mazeland 2007), for example ‘washing and dressing’ activities.

The base line of this SAS consists of an ordered series of physical actions being performed by caregiver and care recipient. Talk during such base line interactions may be either related to the ongoing activity (‘bound’ topics), or unbound. In my paper I will explore the interplay between verbal, non verbal and contextual elements, the way participants shape bound and unbound talk and how this reveals an orientation to the situated activity system.

References

Goffman, E. (1981). Footing. In: E. Goffman, Forms of talk. Philadelphia: University of Pennsylvania Press, pp. 124-159.

Goodwin, C. (2000). Action and embodiment within situated human interaction. Journal of Pragmatics, 32, pp. 1489-1522.

Heinemann, T. (2006). ‘Will you or can’t you?’: Displaying entitlement in interrogative requests. Journal of Pragmatics, 38, pp. 1081-1104.

Mazeland, H. (2007). De noties Situated Activity System en Participation Framework i.v.m. de zorginteracties. Collegetekst CIW. Groningen: Rijksuniversiteit.

Elderly people talking
Christina Englert (University of Groningen)

In my research project I investigate situations and forms of talk of elderly people. My main research question is: Does the talk of elderly people display forms of interactional organization and uses of linguistic practices that are characteristic for and/or constitutive of the social identity of being an elderly person. I am thinking of linguistic and interactional practices recurrent in our daily lives, such as taking turns, telling stories, asking and answering questions, or initiating and doing repair.

I want to study the language use of healthy, not demented elderly people in their natural settings. My primary aim is not to determine what elderly people are not able to do anymore compared to younger adults, but to first describe what kinds of communication situations elderly people take part in, how they contribute to these situations and what kind of interactional and linguistic practices they deploy to shape their interaction.

I am still at a stage in which I am collecting the data, that consists of video recordings of natural occurring interactions among elderly people. I describe preliminary analytical observations by showing examples from my dataset, and I would like to discuss theoretical and methodological issues for the study of language use in the elderly population.

Comprehension of scalar implicatures in five year-old Dutch-speaking children
Myrthe Faber, Jessica Overweg, Margreet van Koert and Angeliek van Hout (University of Groningen)

Much developmental work has been devoted to scalar implicatures. These are implicitly communicated propositions linked to weak terms on a scale of informativity. For example, when a speaker says Some oranges are in the boxes, the hearer is entitled to infer pragmatically that not all of the oranges are in the boxes. The general consensus is that the weaker term (the quantifier some), while logically compatible with a stronger term from the same scale (the quantifier all), prompts the inference because the speaker did not use the stronger term (Pouscoulous et al, 2007). The description is underinformative. In comprehension, children are more likely than adults to treat the weak term as compatible with one that is stronger on the scale. Adults, on the other hand, are also not at ceiling in their performance (Noveck, 2001).

Our aim is to determine the abilities of five year-old children (native Dutch speakers) to draw scalar implicatures with quantifiers and we investigate why some learners seem to be unable to fully comprehend these. The experiment is conducted within the COST A33-framework Crosslinguistically Robust Stages of Children’s Linguistic Performance, with Applications to the Diagnosis of Specific Language Impairment. The goal of COST is to identify and describe areas of cross-linguistic uniformity that can then be used to for cross-linguistic tests of language impairment. The focus of the research is therefore on syntactic, semantic and pragmatic development.

In this study, 25 children (age 5;0-5;6), all native Dutch speakers, plus 20 native Dutch adult controls, are tested using a Truth-Value Judgement Task (TVJT) designed for COST by Napoleon Katsos to be applied cross-linguistically. In this task, Mrs. Caveman asks the child to help her improve her Dutch. Mrs. Caveman makes a statement about a picture, and the child has to say whether the statement is true or false. If the child rejects the statement, we ask for an explanation. An example is given below:

Statement: Sommige sinaasappels liggen in de dozen (Some oranges are in the boxes.)

The child has to say whether it is true or false that some oranges are in the boxes with respect to the presented scene. In this example, true would reflect the logical answer, and false would reflect the pragmatic answer, showing that the child can draw the implicature.

Our goal is to contribute to the cross-linguistic research by providing a systematic analysis of the comprehension of scalar implicatures with quantifiers in 5;0-5;6 year old Dutch children. We are presently collecting data, having tested 9 children so far. In our presentation we will report our results, and discuss methodological issues and implications for further research.

From interlanguage to First Nations English? Grammatical aspects of Joe Little Chief’s Blackfoot-English writings
Inge Genee (University of Lethbridge, Canada)

Joe Little Chief was a Siksika Blackfoot man who learned English as a second language around the turn of the 20th century; in the 1950’s he wrote a collection of stories which he later sold to the Glenbow Archives in Calgary.

This paper argues that the non-standard English features found in Little Chief’s writings reflect the grammar of his native Blackfoot and may have been characteristic of the English spoken by the first generation of Blackfoot people who attended English residential schools. The features under investigation include: (i) non-standard tense and aspect; (ii) non-standard infinitive and participle forms; (iii) non-standard number and gender agreement; (iv) non-standard uses of personal and possessive pronouns, articles and demonstratives; and (v) non-standard uses of conjunctions, copula and prepositions.

What makes these features interesting is the fact that they survived beyond the first few generations of non-native speakers: not only may they still be observed in the English of elderly individuals whose first language is Blackfoot, but, more importantly, some of them are also preserved in the English spoken by Blackfoot people who speak English as their first language and have no or limited fluency in Blackfoot: they have become part of a Blackfoot First Nations English dialect.

Recent work on First Nations Englishes has argued against individual ancestral languages as the main source of First Nations Englishes or emphasized other contributions to their development (e.g. Craig 1991; Flanigan 1985, 1987; Wolfram 1984; Rowicka 2005). I argue, with e.g. Alford 1974 and Leap 1993, that the origin of features which have a clear parallel in the ancestral language is likely to be that ancestral language, especially in communities in which the major shift to English has happened within living memory.

References

Alford, Dan. 1974. The Cheyenne dialect of English and its educational implications. Box Elder, Montana: Northern Cheyenne Bilingual Program.

Craig, Beth. 1991. American Indian English. English World-Wide 12.1:25-61.

Flanigan, Beverly Olsen. 1985. American Indian English and error analysis: The case of Lakota English. English World-Wide 6,2:217-236.

Flanigan, Beverly Olsen. 1987. Language variation among Native Americans: Observations on Lakota English. Journal of English Linguistics 20,2:181-199.

Leap, William L. American Indian English. Salt Lake City: University of Utah Press.

Rowicka, Grażyna. 2005. American Indian English. The Quinault case. English World-Wide 26,3:301-324.

Wolfram, Walt. 1984. Unmarked tense in American Indian English. American Speech 59,1:31-50.

Argumentative writing in L1 and EFL of Dutch secondary school students
Kees de Glopper (University of Groningen)

Research problem
Writing research has demonstrated resemblances and discrepancies between writing processes and writing performance in a first and a second or foreign language. Schoonen et al. (2003, 2009) found very strong correlations between the writing skills in Dutch and English of secondary school students in the Netherlands. Their studies identified differences as well: the development of writing proficiency between grade 8 and grade 10 seems to be larger in English than in Dutch and writing proficiency in English seems to be better predictable with (psycho)linguistic measures than in Dutch. Writing processes in Dutch and in English have been studied by Stevenson (2005). Her study demonstrates similarities and differences in planning, translation and revision processes in Dutch als L1 and English as a foreign language. A detailed analysis of students’ revisions while writing in English did not demonstrate inhibition of higher order revision processes, but writing fluency was reduced and inhibition was found for conceptual processes such as generating ideas and setting goals.

The aforementioned studies do not delve into text characteristics that are genre specific. The writing proficiency scores in the studies of Schoonen et al. (2003, 2009) are so called primary trait scores: general impression ratings of text quality. Stevenson (2005) did address genre specific features of argumentative writing, but her analyses are restricted to a comparison of frequencies of claims and arguments in texts in Dutch and English. This study does provide a detailed analysis of argumentative writing in L1 and EFL by Dutch students.

Method
Argumentative letters of 40 secondary school students (a systematic sample from the larger data set of Schoonen et al. (2003, 2009) were analysed. Four letters per student (two from grade 9 and 10 each, two in Dutch and English each) were subjected to analyses of the argumentatieve structure and the use of indicators for claims and arguments. The outcomes of these analyses have been compared (i) between grades and within languages and (ii) between languages within grades. Further, the correlation was determined between aspects of argument structure and use of indicators and the primary trait scores for writing quality.

Results
The analyses of the argumentative features of the letters written in Dutch and English shed more light on parallels and contrasts in L1 and EFL text production and allow for a more specific answer to the question to what extent written argumentation in a foreign language is inhibited.

References

Schoonen, R., Van Gelderen, A., De Glopper, K., Hulstijn, J., Simis, A., Snellings, P. & Stevenson, M. (2003). First language and second language writing: the role of linguistic fluency, linguistic knowledge and metacognitive knowledge. Language Learning, 53, 1, 165-202.

Schoonen, R., Van Gelderen, A., Stoel, R., Hulstijn, J. & De Glopper, K. (2009, submitted). Modelling writing development: L1 and EFL Writing proficiency in secondary school years

Stevenson, M. (2005). Reading and writing in a foreign language. A comparison of conceptual and linguistic processes in Dutch and English. Amsterdam: SCO-Kohnstamm Instituut.

How do you feel today? The effect of mood on language processing.
Diewke de Goede¹, Petra van Alphen¹, Emma Mulder², José Kerstholt³ and Jos van Berkum^1,4
(¹MPI for Psycholinguistics; ²University of Amsterdam; ³TNO Human Factors; ⁴Radboud University Nijmegen)

Many aspects of cognition, such as memory retrieval, decision-making, and the use of stereotypes, have been found to be sensitive to mood, the diffuse, objectless affective state the person is in (Clore & Huntsinger, 2007). Although the exact mechanisms are hotly debated, the evidence suggests that people in a happy mood are more inclined to rely on heuristic processing strategies than people in a sad mood. Here we investigate whether mood also affects the use of heuristics (or 'educated guesses') to anticipate upcoming language as a sentence unfolds. If it does, this would show that language processing -- a classic example of 'cold' cognitive computation -- is not immune to affective variables.

In constructions like "David praised Linda because...", verbs like "praise" heuristically lead readers to expect more information about the person who is praised (in this case, Linda), not the person praising. This so-called implicit causality bias can be so strong that gender-marked pronouns that subsequently disconfirm the expectation -- "he" in "David praised Linda because he..." -- actually elicit a P600 effect in ERPs (Van Berkum et al., 2007), indicating that such pronouns are briefly taken to be problematic. We reasoned that if people process information more heuristically in a happy mood than in a sad mood, and if such mood-dependent shifts in processing strategy can also affect the mechanism involved in language comprehension, a change in mood should modulate the size of this heuristics-based P600 effect.

In a two-session EEG experiment, we used short film clips to manipulate the mood of participants just before they read short stories in which verb-based expectations were sometimes confirmed or disconfirmed with a gender-marked pronoun. When readers were in a happy mood, bias-inconsistent pronouns elicited a P600 effect, as in the abovementioned ERP study. However, when the same readers were in a sad mood, no such P600 effect was observed. Importantly, standard morpho-syntactic subject-verb agreement violations (e.g., "The boys was…") elicited a P600 effect in either mood.

Our findings support the general idea that mood modulates the degree of heuristic processing, and they reveal that such mood effects also percolate into basic language comprehension mechanisms. A change in mood has selective consequences for language processing: whereas heuristics-based conceptual anticipation can be abolished in a sad mood, more algorithmic syntactic parsing mechanisms continue to do their job.

Are you stressed? Children’s comprehension and production of marked stress
Frederike Groothoff and Sanne Kuijper (University of Groningen)

The interpretation of marked stress can be problematic for children (e.g. Szendröi, 2003). For adults the sentence “Tigger only threw a CHAIR to Piglet” can only mean (1) Tigger only threw a chair, and threw nothing else, to Piglet (narrow focus interpretation). For children this sentence is ambiguous and could also mean: (2) Tigger only threw a chair to Piglet and did nothing else (wide focus interpretation). While having problems with interpretation, children do not seem to have problems with the production of marked stress (e.g. Hornby & Hass, 1970; Cutler & Swinney, 1987). However, production and comprehension of marked stress have never been investigated within the same experiment.

In our experiment we did investigated production and comprehension of marked stress in the same experiment, with the same children and the same materials. We used pictures and stories in which things were bought and others were left in the store. The contrast was between the different products (neutral stress); their different color (marked stress); or both product and color were contrastive (neutral stress). In comprehension we expected that children would accept a wide focus interpretation for marked stress, this in contrast to the adults. In production we expected that the children would have adult like stress patterns.

The results of our experiment are not in line with previous studies. For comprehension, we found that in the marked stress condition subjects more often had a wide focus interpretation than in the experiments of Gualmini et al. (2002) and Szendrői (2004). However, we found different answer patterns for adults and children. For production, the results were not as we had expected: children and adults frequently placed main stress on the noun, independent of the focus in the test item. An explanation for these production results could be that the contrast, and thus the focus to be expressed, was not clear enough in the experiment.

Our follow-up experiment put more emphasis on the contrast between the objects: the participants were instructed to say what the character picked up, and also what he did not pick up, so they would produce sentences with two conjuncts (e.g. “The clown only bought the yellow PANTS and did not buy the yellow HEAD”). The results of the stress patterns for the first conjunct differed from the results for the second conjunct. In the second conjunct children did use the expected stress pattern.

Investigating the prosodic correlates of contrast. Comparing Greek to Italian
Stella Gryllia (Leiden University)

Contrast has a double nature, as it can combine with topic (ex.1b) and focus (ex.2b). The question in (1a) can be interpreted as containing two implicit sub-questions, (i)what did you give to your mum for Christmas and (ii)what did you give to your dad for Christmas. (1b) answers the question partially, addressing only the first sub-question; to my mum is contrasted to other persons to which one could have given a book.

(1)	a.	What did you give to your parents for Christmas?
	b.	[To my mum]_{Contrast+Topic} I gave [a book]_Focus.

Examples like (2b) can also be seen as an answer to an implicit sub-question. In this case,
(2a) is interpreted as containing two sub-questions, (i)what did you give to Helen and (ii)what did you not give to Helen. (2b) answers the question addressing only the first sub-question; in (2b) a cd is contrasted to other things that one could have given to Helen.

(2)	a.	What did you give to Helen?
	b.	I gave her [a cd]_{Contrast+Focus} (not a book).

This double nature of contrast raises a question about its status in information structure (Marandin et al. (2002), Molnár (2002)); should contrast be treated as an independent feature of information structure?

Marandin et al. (2002), based on prosodic evidence from French argue that contrast should be treated as an independent feature of information structure. In French, a C-accent seems to be present in contexts where topic combines with contrast and in contexts where focus combines with contrast. Marandin et al. argue that this C-accent is characterized by a sharp pitch-rise, longer duration and higher intensity, and signals contrast.

The aims of this paper are twofold; first, to investigate whether there is experimental evidence from Greek and Italian for Marandin et al.'s claim. Second, this paper aims at examining whether there is cross-linguistic variation with respect to the prosodic marking of contrast, comparing data from Greek and Italian. To fulfill these aims two production experiments were carried out (one experiment per language). The main question of the production experiments was: do speakers prosodically mark focus or topic for being contrastive?

References

Marandin, J.M. et al. (2002) “Discourse marking in French: C accents and discourse moves.”, Proceedings of Speech Prosody, Aix-en-Provence.

Molnár,V. (2002).“Contrast –from a contrastive perspective”, Information Structure in a Cross-Linguistic Perspective, H.Hallegard et al. (eds.) Amsterdam, Rodopi:147-61.

Audience design in documentary film interviews.
Marieke Haan (University of Groningen)

Investigations of film dialogue show that actors’ interactions normally contain more information than is necessary for the characters in the film in order to understand each other. This extra information is added for the audience, so the dialogues are aimed towards the viewers. Therefore these dialogues are not ‘natural’, but specifically designed for the spectators (Bubel). Also in documentaries, dialogues are designed for the audience. Since many different verbal utterances in documentaries are used, e.g. realized by presenters, narrators, voice-overs, etc. (Nichols, Bruzzi). This investigation concentrates on interviews in documentaries. These interviews do not only show talking heads (with the interviewer often cut off), but also pictures that are related to interview topics.

In terms of the ‘participation framework’ (Goffman, Clark), documentary film audiences may be seen as ‘overhearers’: looking at conversations between actors or interviewers and interviewees, but being no part of the ongoing communication. However, as overhearers, spectators are enabled to understand the interviews in full. There are two relevant cinematographic elements used: verbal communication designed for overhearers, and visual communication designed for audiences during the process of overhearing interviews. Moving images are used to make the overhearers’ knowledge (more) complete, during or in spite of the verbal utterances. The question then is: How is overhearer design in documentary interviews applied? It is investigated by means of a corpus analysis of 71 interview sequences of 10 documentary films.

Verbal utterances in interviews may be understood by the ‘third party’ of the overhearers in front of the TV screen. However, the mutual knowledge between the interactors on the screen – that makes their utterances smooth and easy – must be no longer mute but ‘talkative’. This is the core of the audience design, that the ‘common ground’ (Clark) between the interactors on the screen, their shared knowledge, is made accessible to overhearers.

Speakers (in the interviews) act so that overhearers may profit by their utterance design: they add examples to the verbal text, make expressions concrete, paraphrase propositions, etc. But when interviewees are speaking often images are shown (other than talking heads): e.g. images that are telling the same as the text, that show persons which are referred to, or that provide the viewers with a setting of place or time related to the verbal text. In such a way, the interview information is made available to overhearing and viewing audiences, verbally as well as visually.

References

Bruzzi, S. (2006): New Documentary: A Critical Introduction. London.

Bubel, C.M. (2008): Film Audiences as Overhearers. Journal of pragmatics 40, 55-71.

Bubel, C.M. (2006): The linguistic Construction of Character Relations in TV Drama. PhD Thesis, University of Saarbrücken.

Clark, H.H. (1996): Using Language. Cambridge.

Goffmann, E. (1976): Replies and Responses. Language and society 5, 257-313.

Nichols, B. (2001): Introduction to Documentary. Bloomington.

The origin of Afrikaans pronunciation: a comparison to west Germanic languages and Dutch dialects
Wilbert Heeringa¹ and Febe de Wet²
(¹Meertens Institute Amsterdam; ²Stellenbosch University)

This paper aims to find the origin of Afrikaans pronunciation with the use of dialectometry. First, Afrikaans was compared to Standard Dutch, Standard Frisian and Standard German. Pronunciation distances were measured by means of Levenshtein distances. Afrikaans was found to be closest to Standard Dutch.

Second, Afrikaans pronunciation was compared to 361 Dutch dialect varieties in the Netherlands and North-Belgium. Material from the Reeks Nederlandse Dialectatlassen was used. Afrikaans was found to be closest to the South Holland variety of Zoetermeer, which largely agrees with Kloeke (1950, Herkomst en Groei van het Afrikaans).

Finally, vowel and consonant similarity were studied separately. The strong relationship between Afrikaans and the South Hollandish varieties can be explained by their vowels. With regard to the consonants, the Town Frisian varieties are most closely related to Afrikaans, probably since they still maintain features which were lost in the South-Hollandish dialects.

A Diachronic Study of Irish Prepositions
Geraldine Herbert and Carl Vogel (University of Dublin)

The recent application of computational models to the problem of language change has demonstrated that quantitative measures of language change can be yielded. We present an attempt to quantify aspects of Language change in Irish. To this end a model of language change proposed by Lieberman et al (2007) is employed.

Lieberman's work focuses on tracking the rate of regularization in English verbs and concludes that the rate of regularization and the frequency of usage of any word are inversely related to each other; a verb that is 100 times less frequent regularizes 10 times as fast. While evidence of an inverse relationship between frequency and complexity is not new the use by Lieberman of a computational model to quantify the relationship is novel and allows predictions to be made about regularization in the future.

We explore the survival of Irish prepositions from Old Irish through to Modern Irish. Prepositions were selected as a closed class category to view in isolation from the rest of the grammar. Although relational in general, as verbs are, unlike verbs, prepositions to not tend to have inflection patterns to regularize. Thus, the ontogenesis and survival of particular prepositions is of interest, and we wish to establish whether the survival rates relative to frequency of attested use of prepositions replicates the Lieberman model. However, Irish also has prepositions that incorporate pronouns and which are relevant to study in terms of possible inflection paradigms in the same way as the verb study.

We observe that extant analysed diachronic corpora of Irish suggests that the overall survival of the Old Irish Prepositions conform to the spirit of the Lieberman model, but there are exceptions with the survival of some prepositions that do not record high frequency of occurrence while a number of prepositions with high frequency did not survive.

How depictives and appositive adjectives are different
Herman Heringa (University of Groningen)

Depictives and appositive adjectives, illustrated in (1) and (2) respectively, seem to be very similar constructions. The only apparent distinction between the two is the intonation. The appositive adjectives are separated from the intonation contour of the matrix by so-called comma-intonation. Depictives lack this special intonation pattern.

(1)	a.	John arrived home, drunk.	[appositive adjectives]
	b.	They dragged Mary, totally unconscious, into the ambulance.

(2)	a.	John arrived home drunk.	[depictives]
	b.	They dragged Mary unconscious into the ambulance.

In this talk, I will show that depictives and appositive adjectives differ also in their restrictions and possibilities, both syntactically and semantically. Appositive adjectives, for example, can be preceded by subordinators, whereas depictives cannot (3). Also, depictives are always stage level modifiers, whereas appositive adjectives can be used to modify on the individual level (4).

(3)	a.		They dragged Mary, though entirely healthy, into the ambulance.
	b.	*	They dragged Mary though entirely healthy into the ambulance

(4)	a.		Peter, really smart, won the quiz.
	b.	*	Peter won the quiz really smart.

The main focus of the talk will be on case marking on the constructions in question in the slavic languages, in particular Czech and Russian. In Czech, both depictives and appositive adjectives always get the same case as their anchor. In Russian, on the other hand, depictives have a sameness/instrumental alternation, depending on aspect. Appositive adjectives, however, do not show this alternation and are obligatorily marked with the same case as their anchor. The example in (5) is from Richardson (2007: 116).

(5)	Druz'ja	priveli	ego	domoj	p'janymi / p'janogoi	[Russian]
	friends	brought	him:acc	home	drunk:instr / drunk:acc
	'Friends brought him home drunk.'
	a. Appositive interpretation (Case agreement)
	b. NP modifier/attribute interpretation (Case agreement)
	c. Depictive interpretation (instrumental Case or Case agreement)

I will argue that both depictives and appositions are control structures where the adjective functions as the predicate of the PRO subject. Appositive adjectives are adjoined to the subject or object itself. Depictives, on the other hand, are adjoined to a projection of the verb. Furthermore, I propose that appositive adjectives involve a complete CP, whereas depictives are small clauses. In order to explain the case marking patterns, I will use the system developed for case marking predicates from Matushansky (2008). In this system, heads assign case to their sister and everything it includes. Also, elements can receive more than one case feature and PF has rules to decide which of these features will be spelled out.

References

Matushansky, O. (2008). A Case Study of Predication. In F. Marušič and R. Žaucer, eds., Studies in Formal Slavic Linguistics. Contributions from Formal Description of Slavic Languages 6.5. Frankfurt am Main: Peter Lang, pp. 213-239.

Richardson, K. (2007). Case and Aspect in Slavic. (Oxford Linguistics). Oxford: Oxford University Press.

Discontinuous conjunction of nominal modifiers
Jack Hoeksema (University of Groningen)

There are two competing views on the basic parallelism requirements in the area of conjunction: (1) conjuncts must have identical categories, based on internal constituency, (2) conjuncts must have identical categories, based on distributional properties. The former view is that of X-bar grammar and its various descendants, whereas the latter is enshrined in the categorial grammar tradition (cf. e.g. Steedman 1985). I will present new evidence for the categorial perspective, from a type of coordination involving prenominal adjectival phrases and postnominal PPs and relative clauses in older stages of Dutch.

(1)	De Heer Bodisco was een zeer beschaafde man, en op wiens oordeel, in verschillende betrekkingen, groote prijs gesteld werd.
	“Mr Bodisco was a very civilized man, and whose judgment, in various matters, was highly esteemed.”
(2)	Het was eene seer lange brief, en die door alle drie die heeren nog al een wijle wierde bestudeert.
	“It was a very long letter, and which was studied by those three gentlemen for quite a while”

In categorial terms, both conjuncts, though internally quite divers, as similar because they as adnominal modifiers of the general category NP/NP. In current Dutch, sentences such as (1) are out. I will argue that this is due to conflicting requirements of directionality, which became gradually more important. A categorial model, with an OT superstructure, can successfully account for this type of change.

Reference

Mark Steedman, 1985, ‘Dependency and Coordination in the Grammar of Dutch and English,’ Language 61-3, 523-568.

Verb-Raising and the Head Final Filter
Erik Hoekstra (Fryske Akademy, Royal Netherlands Academy of Sciences)

The following patterns represent the ordering possibilities for 3-verb clusters (without the infinitival marker "to") in West Germanic languages past and present (inspired by Koopman 1990):

(1) Verb cluster break-up scheme

V1 ... V2 ... V3 ...		V2 ... V3 V1
V1 ... V3 V2 ...		V2 V1 ... V3
V3 V2 V1 ...		V3 V1 ... V2

V3 is the main verb; V1 and V2 are auxiliaries such as perfect auxiliaries, modal auxiliaries. The dots represent break-up of the verbal cluster by non-verbal material. I will present data in support of (1) from Modern Dutch, Modern Frisian, Old English, Old Frisian, as well as from several Dutch dialects of the 19th and 20th centuries. The generalisation covering the facts in (1) is given in (2):

(2) Extended Koopman's Generalisation
A head-final ordering relation between two verbs may not be broken up by non-verbal material

Thus the following suborders are never broken up: 32, 31 and 21.

(2) is not as strange as it may seem. A similar, though not identical, generalisation holds with respect to prenominal AP's, whose heads must be adjacent to (the projection of) the noun. It goes by the name of the Head Final Filter. After presenting the empirical basis for the Head Final Filter (Williams 1981), I go on to discuss the following questions:

(3-i) What sort of principle (semantic, syntactic, phonological) is responsible for this ordering generalisation?
(3-ii) Does the concept "underlying order" need to exist in syntactic theory?
(3-iii) What factor correlates with the difference between verb ordering and adjective ordering.

The answers will be: syntactic, no, selection versus iteration; but the motivation for these answers are, hopefully, more interesting than the answers themselves. Given time, I will criticize some existing account of V-Raising.

References

Koopman, W. (1990). Word order in Old English. With special reference to the Verb Phrase. Dissertation University of Amsterdam. Amsterdam Studies in Generative Grammar 1.

Williams, Edwin (1982) “Another Argument that Passive is Transformational,” Linguistic Inquiry 13, 160–163.

It’s a tie. The influence of gestures and their visibility on speech.
Marieke Hoetjes, Emiel Krahmer and Marc Swerts (Tilburg University)

Gestures can be defined as the symbolic movements that we make when we speak. They are such an integral part of speech that people are generally not aware of producing them or of perceiving them. Once we start paying attention, however, it becomes clear that speech-accompanying gestures are omnipresent. Despite this omnipresence, gestures tend to be overlooked when studying language use and many aspects of their relationship to speech and their role in communication have only recently been studied.

Various studies have shown that there is a close relationship between speech and gesture, but several questions remain. For example, do people produce gestures primarily for themselves or for the addressee and to what extent does performing a gesture have an effect on speech itself? To try and answer these questions we might ask whether people are still likely to gesture when they cannot see their addressee and whether people’s speech changes when they cannot gesture.

An experimental paradigm was developed in which thirty eight first year student pairs took part. One of the participants, the instructor, had to watch video clips depicting a person tying different kinds of tie-knots and instruct the other participant, the matcher, to tie a tie in the same manner as in the video clips. Half of the instructors were unable to see the matcher because an opaque screen had been placed between the instructor and the matcher and all instructors had to sit on their hands for half of the experiment. The instructor was videotaped during the experiment and photographs were taken of the end state of each tie-knot on the matcher.

In this talk I want to discuss whether visibility leads to better instructions and whether it is possible to hear whether an instructor is able to gesture or not. In other words, are matchers better at tying a tie-knot when they can see the instructor and does the instructors’ speech change when they are unable to see the matcher or when they are ‘tied down’ and unable to gesture? Results show that not being allowed to gesture during this kind of instruction giving is difficult, and that gestures can have a noticeable impact on speech. In addition, it was found that instructors continue to gesture even when the matcher cannot see them.

Relearning in the elderly: Age-related effects on the size of savings
Nienke van der Hoeven-Houtzager (University of Groningen)

This paper reports on a study on learning new words and relearning old words in young (mean age 22.40), middle-aged (mean age 50.33) and elderly speakers (mean age 76.00) of French as a foreign language. The study focuses on the question if there is a difference between learning new words and relearning words that were learned in the past but seem to be forgotten. The results show that the three groups perform rather similarly in relearning old words, but that the younger learners are significantly more efficient in learning new words. In addition a questionnaire on contact with French and a test for working memory capacity were administered, but neither language contact nor age-related decline in working memory capacity can be seen as single factors explaining the difference between learning and relearning. The decline in older adults to learn new lexical information is related to theories that explain age-related memory deficits, and is conceived of as an age-related deficiency to form associations between unrelated concepts.

Poor nonword repetition in specific language impairment: cause or consequence of small vocabulary size?
Martine Jong¹, Judith Rispens² and Gerard Bol¹
(¹University of Groningen; ²University of Amsterdam)

Children with Specific Language Impairment (SLI) have severe problems with nonword repetition (NWR). Controversy exists on the causal mechanisms of this deficit. A dominant hypothesis is that NWR is constrained by phonological storage capacity, and that this plays a key role in learning new words. However, difficulties with NWR have also been interpreted as a consequence of a small vocabulary size rather than as a causal factor. The present study is an investigation into the association between NWR, vocabulary and word learning in SLI.

Two experiments have been carried out, one in which the items of an NWR task have been varied on (1) length to study the capacity of the phonological short term memory and (2) phonotactic probability. The latter measure has shown to influence NWR and has been taken as evidence that the repetition of nonwords makes use of the lexical representations in the mental lexicon. In the second experiment, the ability to learn novel words was investigated by a long-term repetition priming task in which nonwords were either presented once or twice.

Three groups of children participated: one group of 8-year-old children with SLI, one group of age-matched typically developing (TD) peers, and one group of younger TD children matched with the children with SLI on vocabulary size.

The results of the first experiment show a significant influence of phonotactic probability on NWR. This effect is the same for all three groups. Length also has a significant effect on NWR. There is a significant interaction between group and length as the SLI group is more severely affected by nonword length. No main effect of priming was found in the second experiment, but there was a significant interaction between phonotactic probability and priming, and a marginally significant interaction between phonotactic probability and group. The TD children showed a priming effect on low phonotactic probability items only. The SLI children did not show a priming effect in general. These findings underline that NWR is affected by lexical influences, but the group differences are better explained by a limited phonological short-term capacity of the SLI children. The SLI children furthermore demonstrate more difficulties with novel word learning. The results of our experiments will be discussed with respect to the NWR deficit and word learning in SLI.

Functional categories in learner languages
Peter Jordens (VU University Amsterdam)

Language acquisition is a developmental process which occurs in stage-wise progression. At the initial stage both in first (L1) and second (L2) language acquisition learners make use of a relatively simple language system to communicate. Later on, as the result of a process of language development interacting with target language input, this basic learner variety will normally be given up in favour of a more complex version.

I will argue that both in child L1 and in adult L2 Dutch, learner varieties develop from a lexical system to a functional system. At the lexical stage, functional categories are absent. Hence, functional properties of the linguistic system such as auxiliaries, inflection, word order variation and the determiner system are not part of the grammatical system of learners at the relevant stage, at least not productively.

My claim is that at the initial stage of language acquisition, utterance structure is based on linguistic knowledge of predicate-argument structure as it is stored in the mental lexicon. At the relevant stage, this lexical knowledge is simultaneously used for the purpose of information structuring. However, this is unlike the target language. In the target language, information structure is expressed by means of functional elements. Therefore, I will argue that it is the principles of information structure that are the driving force causing learners to develop their lexical variety at the initial stages of language acquisition into the functional variety of a fully-fledged target system.

The data of the present study originate from investigations on the acquisition of Dutch by children learning this language as their native language and adults learning it in an untutored second language learning environment. Both the L1 and L2 data come from longitudinal studies of utterances produced spontaneously. At the lexical stage, topicalisation cannot be expressed with the functional means of the target system. However, it can be expressed with the structure of an agentive lexical projection as in disse hoeniet meeneme (this have-to-not withtake) or die magwel kopen (that may-indeed buy). Reanalysis of the initial position as a topic position can be accounted for in terms of a functional projection (FP). FP serves as the structural prerequisite for the expression of the functional properties of information structure of the target language. Hence, topicalisation is the driving force in the development of learner languages from a lexical to a functional system.

Understanding Discriminative Perception of English Consonant Minimal Pairs from the Perspective of Two Levels of Processing: Sound and Meaning
Hirokatsu Kawashima (Nagasaki University of Foreign Languages)

Recent years have seen an increasing number of studies dealing with ESL/EFL listening and steady developments in both theory and practice (e.g., Rost 2002). It must be noted, however, that much of the nature of learning and teaching of listening remains unclear and unexplored. Buck (2001: P.51) points out that there have been no “complete unified descriptions” of the sub-skills of listening. Phoneme identification is claimed to be one important sub-skill of listening in the literature (such as Flowerdew & Miller 2005), for example, but systematic understanding of its nature has not yet been obtained.

An investigation was conducted in order to make a contribution to this research situation, focusing upon discriminative perception of English consonant minimal pairs. Japanese learners of English (non-English major university students, aged 18-19, N=42) participated in this investigation, in which they were asked to discriminate 7 types of English consonant minimal pairs phonetically (e.g., light vs. right, N=42) and to judge if English sentences, which include such pairs (N=42), are semantically correct (e.g., I sink the story is interesting vs. I think the story is interesting, N=42). Their discriminative performances at these two processing levels and general listening proficiency were measured, and relationships among sound-based and meaning-based discriminative performances and general listening proficiency were examined employing simple and multiple regression analyses.

The analysis of the collected data shows, for example, 1) that relationships between discriminative performances at semantic and phonetic levels should be viewed as more non-linear rather than linear, 2) that such relationships become stronger as general listening proficiency increases, and 3) meaning-based discriminative performance is more related to general listening proficiency than sound-based discriminative performance. These kinds of relationships are deemed to represent an important aspect of the nature of discriminative perception of English consonant minimal pairs. The presenter, based upon multiple analyses of the data, will discuss such findings in terms of the sub-skills of listening.

References

Buck, G. (2001) Assessing Listening. Cambridge University Press

Flowerdew, J. & L. Miller (2005) Second Language Listening: Theory and Practice. Cambridge University Press

Rost, M. (2002) Teaching and Researching Listening. Longman

Modeling Ordering Effect in Binary Judgment Experiment
Yu-guang Ko (National Chung Cheng University)

Ordering has been found to be influential on acceptability of sentences. For example, Luka and Barslou (2005) reported a structural priming effect caused by prior presentation of sentences with identical structure in a series of acceptability judgment experiments.

To factor out the effect of ordering, Myers (2007) and Ko (2007) analyzed order of sentence presented in a experimental session as a continuous and linear factor. A significant increase in acceptability for grammatical sentences was found in Ko (2007), replicating the result of Luka and Barsalou (2005). However, contrary to what Luka and Barsalou (2005) argue, the significant ordering effect can also be caused by a “practice effect”, namely, an effect caused by subjects’ increasing familiarity with the experimental paradigm.

Although the structural priming effect and the “practice effect” are both reflected on ordering of presentation, they can be torn up by a statistical technique discussed in Baayen (2008) called “Break-point analysis”. The break-point analysis looks for a specific point in a continuous scale and computes the effects of factors to data before and after this point separately. To demonstrate its usefulness, the present study conducted a binary judgment experiment and applied the technique above to analyze it to detect possible discontinuity caused by practice and structural identity and measures the respective effect size of them separately. The result showed that a break-point came in at the early stage of the experimental session, while significant ordering effects were found in two groups of data points.

Based on the result, it can be argued that the subjects became familiar with the experimental paradigm of binary judgment with ease, since not many trials were used for practice. The ordering effect was still present after that, arguably to be caused by repeated exposure to sentences with identical structure.

References

Baayen, R. H. 2008. Analyzing Linguistic Data: A Practical Introduction to Statistics. Cambridge University Press, Cambridge, UK.

Ko, Y.-G., 2007. Grammaticality and Parsability in Mandarin Syntactic Judgment Experiments. Master's thesis, National Chung Cheng University, Chiayi, Taiwan.

Luka, B & Barsalou, L. W. 2005. Structural facilitation: Mere exposure effects for grammatical acceptability as evidence for syntactic priming in comprehension. Journal of Memory and Language, 52, 436-459.

Myers, J. 2009. The design and analysis of small-scale syntactic judgment experiments. Lingua, 119,425-444.

A chart generator for the Dutch Alpino Grammar
Daniël de Kok and Gertjan van Noord (University of Groningen)

Natural language generation is a useful technique for grammar verification, machine translation, sentence compression, sentence fusion, and other tasks. In this talk we will introduce a chart generator for the wide-coverage Alpino grammar for Dutch.

Chart generation has been applied succesfully to natural language generation for other grammars, such as the English Resource Grammar (Caroll and Oepen, 2005). But characteristics of the Alpino grammar requires a different methodology than previously described in literature. For instance, Alpino represents the analysis of a sentence as a dependency tree rather than using a formalism such as Minimal Recursion Semantics. Additionally, the Alpino grammar builds dependency structures in a manner where we can not always check immediately whether the required dependency relations are present.

We will report on the abstract dependency trees used as input for our generator, which are underspecified variants of CGN dependency trees. We will also discuss the bottom-up generation algorithm, and how we incoorporate top-down knowledge such as expected dependency relations to guide the generation process. Finally, we will report on our first experiences with generation from abstract dependency trees derived from the Alpino test suite.

Need I say more? On factors causing referential overspecification
Ruud Koolen and Emiel Krahmer (Tilburg University)

Referring expressions (e.g. ‘the green chair’) are ubiquitous in human speech production. It is often assumed that speakers include just enough information in their references for the addressee to single out who or what the speaker is referring to (‘the target’). However, recent research (e.g. Engelhardt et al. 2006) has revealed that speakers often overspecify their references and include more information than is strictly speaking necessary for identification. Yet, when, why and how speakers overspecify is still largely unknown.

We hypothesize that at least three factors may cause speakers to overspecify: the complexity of the domain, the complexity of the target object(s), and the communicative setting. In order to investigate to what extent these factors influence the information load of referring expressions, the D-TUNA corpus was collected: a corpus of 2400 Dutch referring expressions. Data was collected in a large elicitation experiment, in which participants were presented trials consisting of pictures in two possible domains: the furniture domain and the more complex people domain. Complexity of the trials was manipulated by including both singular and plural trials. For singular trials, participants were asked to describe one target object in such a way that it could be distinguished from six distractor objects; for plural trials, participants were asked to describe two target objects. Sixty participants were randomly assigned to three conditions, which represented three communicative settings: text, speech and face-to-face. Participants in the text condition typed their identifying descriptions, while participants in the speech condition and the face-to-face condition were asked to speak out their descriptions to a confederate. Only in the face-to-face condition, the participants could see the confederate. Each identifying description was annotated semantically in XML: it was provided with information regarding its own properties and the properties of the distractor objects.

Results show effects on overspecification for all three factors mentioned. Referring expressions are more frequently overspecified in the complex people domain. The same goes for expressions describing complex plural trials, for which descriptions are more frequently overspecified than for singular trials. Written and spoken referring expressions appear not to differ in semantic content, but in number of words: speakers need more words to provide the same information as people who type their expressions. Current analyses focus on how speakers overspecify exactly, and these results will be presented at TABU as well.

Reference

Engelhardt, P., Bailey, K. and Ferreira, F. (2006). Do speakers and listeners observe the Gricean Maxim of Quantity? Journal of Memory and Language, 54, 554-573

Training a statistical parser for parsing French for use in syntax-based machine translation
Gideon Kotzé (University of Groningen)

In this presentation, we investigate various techniques on successfully training a statistical parser using the MaltParser system on a French treebank. First, we give an overview of the data available to us, mentioning statistics and some preprocessing necessary for training the parser. Next, we represent the most optimal settings to date, discussing all significant improvements and the reasons for them. We also discuss the tools we used to optimise training, which include a separate part-of-speech tagger and lemmatiser, and give an overview of the important linguistic features needed for training. The work was done in the context of the STEVIN project Parse and Corpus Based Machine Translation (PaCoMT), in which we aim to integrate efficient parsers for French, English and Dutch into the MT engine.

Morphology = Syntax = Morphology
Joost Kremers (University of Frankfurt)

The difference between morphology and syntax is theoretically expressed as a distinction between the types of structures they involve: morphology deals with X-structures, syntax with XP-structures. In bare phrase structure, however, projection levels are derivative, which means that it is not possible to distinguish between the two. When we merge 'boil' and 'water', the initial merge does not tell us whether the derivation will eventually yield 'to boil water' or 'water boiler'.

The fact that in the former case, [boil water] is embedded under functional material and in the latter case under the nominaliser '-er' won't help us, either. Agglutinative and polysynthetic languages such as Turkish, Nahuatl or Inuktitut suggest that "morphological" structures may include functional material as well.

I propose that the distinction between "syntax" and "morphology" is actually an optical (or rather acoustic) illusion. There is only one structure-building mechanism, let's call it Syntax. This module manipulates heads, which are bundles of morpho-syntactic features, WITHOUT the phonological material associated with those heads (cf. Beard's 1988 'Separation Hypothesis' or Late Insertion in DM).

The phonological material is dealt with in the phonological component. Lexical mapping rules associate syntactic structures (not just heads but also larger structures, e.g. idioms) with phonological material (cf. Jackendoff 1997, 2002, Ackema & Neeleman 2004). Some syntactic structures are associated with segmental material or syllabic structures (cf. prosodic morphology, McCarthy & Prince 1996), while other syntactic structures are mapped onto phonological (prosodic) structures in terms of prosodic words and phonological phrases. The former we customarily call "morphology", the latter "syntax".

This model allows a much more straightforward description of certain types of syntax-morphology interaction. Gerund-like deverbal nouns, for example, have syntactic properties, e.g. in that they can assign accusative case, but the noun itself appears a product of morphology. In the current model, we can simply say that the nominaliser attaches to vP (much in the spirit of Abney 1987). Crucially, the process that combines the phonological material associated with the two heads takes place in phonology, not in syntax. This is especially helpful in the case of Arabic, where something like head movement of V or lowering of the nominaliser is undesirable and where the nominalising morphology is prosodic, not affixal.

Another example of interaction are so-called facial adverbs in sign languages. These are adverbs expressed on the face, simultaneously with the verb they modify. As adverbs, they seem to be syntactic, but their form suggests they are morphological. The current model resolves this conflicting status.

Recalling extended metaphors in news discourse
Tina Krennmayr (VU University Amsterdam)

Journalists may play with words; they may use metaphors as a rhetorical device and employ them to persuade readers or to explain an abstract topic to a wider audience. Metaphors, in particular extended metaphors, help create a coherent text structure.

Manual metaphor annotation of 200,000 words from the registers fiction, conversation, academic texts and news of the BNC-Baby corpus has shown that, especially compared to literary texts and spoken language, journalistic writing is rich in metaphorical expressions (Steen, Dorst, Herrmann, Kaal, Krennmayr & Pasma, in preparation). However, merely because a word or phrase counts as a metaphor on a symbolic level, we cannot automatically conclude that readers activate underlying conceptual structures every time they come across metaphorical expressions in news texts.

A number of reaction times studies have tested whether people use conceptual mappings when processing metaphorical language (e.g. Nayak & Gibbs 1990; Glucksberg, Brown & McGlone 1993; Allbritton, McKoon & Gerrig 1995; Keysar, Shen, Glucksberg & Horton 2000; Thibodeau & Durgin 2008). The findings are inconsistent. Some present evidence that even conventional metaphorical expressions (e.g. Thibodeau & Durgin 2008) and idiomatic expressions (e.g. Nayak & Gibbs 1990) activate mappings while others claim that people rely on conceptual metaphors for novel expressions only (e.g. Keysar, Shen, Glucksberg, Horton 2000).

Adding an alternative perspective, I present an off-line study – a cued delayed recall task – to examine weather readers may rely on metaphoric schemas when processing extended metaphor, drawing on research by Bower, Black & Turner (1979) on script recall. The test passages resemble authentic news reports. Using a between-subject design, I manipulate the degree of conventionality of metaphorical expressions as well as signaling of an extended mapping (auto racing). For instance, in the conventional condition, a subject would read “(…) China and India pulled ahead economically (…)” while subjects in the novel condition would read “(…) China and India turbocharged ahead economically (…)”. In the condition with signal, the source domain is made explicit by a simile (“Economic development is (…) like auto racing”) while subjects in the no-signal condition do not receive such a cue. We assume that in a two-day delay the text base will have decayed and the subjects need to rely on the situation model (van Dijk & Kintsch, 1983) when recalling the text. The recall results will allow tentative conclusions about the use of metaphorical schemas in language comprehension.

References

Allbritton, D. W., McKoon, G. & Gerrig, R.J. (1995). Metaphor-Based Schemas and Text Representations: Making Connections Through Conceptual Metaphors. Journal of Experimental Psychology, 21 (3), 612 – 625

Bower, G. H., Black, John B. & Turner, T. J. (1979). Scripts in Memory for Text. Cognitive Psychology, 11, 177-220.

Gerard J. Steen, Lettie Dorst, Berenike Herrmann, Anna Kaal, Tina Krennmayr, Tryntje Pasma (in preparation). Linguistic metaphor identification in natural discourse: A casebook. Amsterdam: John Benjamins.

Glucksberg, S., Brown, M. & McGlone, M.S. (1993). Conceptual metaphors are not automatically accessed during idiom comprehension. Memory & Cognition, 21, 711-715.

Keysar, B. Shen, Y., Glucksberg, S. & Horton, W.S. (2000). Conventional Language: How Metaphorical Is It? Journal of Memory and Language 43, 576-593.

Nayak, N.P. & Gibbs, R.W. (1990). Conceptual knowledge in the interpretation of idioms. Journal of Experimental Psychology: General, 119, 315-330.

Thibodeau, P. & Durgin, F. H. (2008). Productive figurative communication: Conventional metaphors facilitate the comprehension of related novel metaphors. Journal of Memory and Language, 58, 521-540.

Van Dijk, T. A., & Kintsch, W. (1983). Strategies of discourse comprehension. New York: Academic Press.

Linear and Hypertext Discourse Structure: Printed and Online Support of Newspapers
Anne Küppers (Université Catholique de Louvain)

The talk reports on ongoing work dealing with the linguistic impact of putting the traditional newspaper news online. In this framework we investigate the differences in structuring printed and online newspapers with respect to their linear and hypertextual organization of information. The aim is to answer the question to what extent the paper version differs from the Internet version of the same newspapers regarding the way in which information is displayed.

Whereas the structure of printed media like newspapers, scientific journal articles or books is fixed historically and refers to given models (titles, table of contents, summaries, text structure etc.), hypertext is not (yet) based on such well established, historically proven organization patterns (Fastrez & Peeters, 2001, 167).

Our endeavour will be a methodological one. We present a grid that was built up in order to analyse the linguistic structure of printed and online “frontpage” news. In line with Jucker (2003, 133), we assume that the hypertextual organization of information in online newspapers is an effect of the overall increase of information which requires a modularization of this information. We will show that differences can be observed on the macro-level of the document structure as well as on its micro-level or hyperstructure in the sense of Lugrin (2001). Typical characteristics of the macro-level include the section disposition, the peritext, the presence and placing of titling and images etc., while the micro-level is analysed in terms of features like the function of titling and images, the text types introduced, or the discourse structuring by the peritext and intratextual referencing. Some of these features play a role on both levels of analysis thus requiring a holistic method linking up the two structural levels.

On the basis of an empirical investigation of a corpus of printed and online versions of Belgian, French-speaking newspapers, we will show how the structures of linear and hypertext information organization differ between the two media.

References

Fastrez, P. & H. Peeters (2001): Proposition d’une critériologie dans le choix des modes de structuration des hypermédias. In: Recherches en Communication, 16.

Jucker, A. (2003): Mass media communication at the beginning of the twenty-first century: Dimensions of change. In: Journal of Historical Pragmatics, 4 (1). 129-148.

Lugrin, G. (2001): Le mélange des genres dans l’hyperstructure. In: Semen, 13. (http://semen.revues.org/document2645.html).

The effect of teacher elicitation techniques during reading aloud sessions
Aletta Kwant, Jan Berenst and Kees de Glopper (University of Groningen)

Language learning is a major goal in educational settings for young children. In these settings reading aloud sessions play an important role. These sessions offer rich opportunities for language learning because the picture books that are used contain a wider vocabulary than occurs in ordinary conversations. Picture books are therefore part and parcel of intervention programs for early childhood education. Picture books also have rich contents for the social and emotional developmental. However, a specific support of picture books for the language development in the social and emotional domain is no common practice.

To be effective in supporting the social and emotional development of young children the social and emotional content and the corresponding language of the picture books must be within the reach of children of this age. This is not a matter of course because most children in preschool and kindergarten can not read themselves and need a mediator to become familiar with the content and language of a picture book. In an educational setting it usually will be the teacher who mediates between the children and the book(s).

During the reading aloud sessions in the intervention part of the research project described here, teachers have ‘keys’ at their disposal. These keys contain instructions for the reading aloud session. The keys specify elicitation techniques that aim at stimulating children to think about and reflect on the books in a more conscious way and providing children with opportunity in participating in the reading aloud sessions in an active and attuned way.

This presentation addresses the effects of the keys and gives an answer to the following question: What are the effects of different kind of teacher elicitation techniques on the contributions of children during a picture book reading aloud session?

We will discuss the effects of different types of teacher elicitations on the contributions of the children during reading aloud sessions. It will be shown that delayed reactions by the teacher, longer pauses and non/verbal communication positively affect the contributions of the children. Due to the keys, the role of the teacher changes from instructor to conversational partner. Especially a wondering and unsure attitude of the teacher with respect to social and emotional book events elicits a greater amount of contributions of the children and more attuned reactions. Video fragments will illustrate our presentation.

A Dialectometric Study of Swedish Vowels
Therese Leinonen (University of Groningen)

This paper offers a synchronic study of Swedish dialects based on acoustic analysis of vowel pronunciation. Among Swedish dialects vowels differ a lot in contrast to consonants, and the variation in vowel pronunciation is important for characterizing dialectal identity (Bruce, Engstrand and Eriksson 1998). The data for the present study come from the project Swedia2000 (http://swedia.ling.gu.se/) where vowels were elicited with mono- and bisyllabic words with vowels in a coronal context. 105 sites in Sweden and Swedish-language areas in Finland are included and each site is represented by 12 speakers (3 elderly women, 3 elderly men, 3 young men and 3 young women).

The vowel pronunciations are analyzed acoustically by means of principal component analysis of Bark filtered spectra, a method which shows high correlation with formant measurements but can be automated more reliably to a higher extent (Jacobi 2009).

Comparison of the older and the younger speaker group shows a significant ongoing dialect levelling. In peripheral parts of the language area where the older speakers speak very divergent dialects the younger speakers show strong convergence to standard Swedish. In some urban areas, on the other hand, the younger speakers seem to develop varieties that diverge from the standard language.

Distances between varieties are analyzed by means of cluster analysis and multidimensional scaling (Heeringa 2004). Results are visualized on choropleth maps.

References

Bruce, G., Engstrand, O. and Eriksson, A. (1998), De svenska dialekternas fonetik och fonologi år 2000 (Swedia 2000) en projektbeskrivning, Folkmålsstudier 39, 33–54.

Heeringa, W. (2004), Measuring Dialect Pronunciation Differences using Levenshtein Distance, PhD thesis, University of Groningen.

Jacobi I. (2009), On Variation and Change in Diphthongs and Long Vowels of Spoken Dutch, PhD thesis, University of Amsterdam.

Dorsal Fricatives in German: Derivation and Representation
Kathrin Linke (Leiden University)

We present the well-known case of ich-laut vs. ach-laut alternation in German, and show that its interaction with related phonological processes leads to opacity in such a way that no derivational version of OT developed so far can account for it. We propose a representational solution, to which we provide additional evidence from Greek.

1. The issue. The German consonantal phoneme inventory provides a voiceless dorsal fricative with a velar and a palatal allophonic variant as illustrated in Buch [buːx] `book' and Bücher [by:ç ɐ̥] `books'.

The dorsal fricative surfaces as the velar variant [x] following non-front vowels and as the palatal variant [ç] after front vowels and, additionally, sonorant coronal consonants, which suggests that they are regressively assimilated to the tautomorphemically preceding front or non-front segment.

2. Interacting Processes Leading to Opacity. A second allophonic process in German is ʀ-vocalisation, in which /ʀ/ is replaced by [ɐ̥] in syllable codas (Itô & Mester 2001), resulting in incidences in which the palatal fricative opaquely surfaces after 'back' segments.

Furthermore, DFA also interacts with the phonological processes of g-spirantisation and coda devoicing, which account for the differences in word pairs such as König [køːnɪç] 'king' / Könige [køːnɪgə] 'kings'. In Standard German, underlying /g/ in syllable codas with nucleus /ɪ/ are spirantised, opaquely surfacing as the palatal fricative [ç] (Hall 1989, Itô & Mester 2001).

3. Account Using Derivational Versions of OT. Firstly, we will discuss OT based approaches to DFA in German and their accounts for the opaque results of the interacting processes mentioned above. It will be shown that e.g. the Richness-of-the-Base principle causes problems that cannot be overcome even by using Candidate Chain Theory (McCarthy 2006), an approach for treating opacity in OT that adopts a maximal degree of derivationalism.

4. Account Incorporating Representations in OT. Evidence from Modern Greek, in which a similar (but reversed) dorsal fricative assimilation phenomenon can be witnessed, suggests that this assimilation process is not actively enforced by constraints but instead heavily relies on the feature geometric structure of the dorsal fricative and its preceding and/or succeeding elements and processes such as feature economy. Therefore, secondly, an alternative analysis of DFA (and related processes) in German will be introduced that incorporates representations in the OT framework. We argue that dorsal fricatives lack a vocalic place specifcation and need to borrow this specifcation from their neighbouring segments.

Modelling early phonological development in a foreign language
Wander Lowie (University of Groningen)

In this paper I will report on a longitudinal study investigating several aspects of early phonological development in L2. Where previous studies into L2 phonology have mainly concentrated on phonological variables in isolation and cross-sectionally, the current study focuses on the development of pronunciation from a dynamic point of view. To this end, the study uses variability analyses (cf. Verspoor, Lowie, & van Dijk 2008) to investigate a range of phonetic correlates, followed by simulations to test hypotheses derived from the data. This paper reports on a longitudinal analysis of the L2 development of a 7-9 year old Dutch learner of English. The analyses concentrate on the Voice Onset Time and the /ε/ - /æ/ vowel contrast over a period of two years. The data reveal that the developmental patterns of early L2 learners show a high degree of variability and an interesting interaction between the phonetic correlates investigated and the speaking situation.

The discourse marker like in clause-final position
Kathrin Luckmann (University of Duisburg-Essen)

The rapid spread of discourse like in urban centres and predominant use by young people has led it to be stylistically stigmatized, particularly the quotative be like, which has recently been subject to numerous investigations. Despite having been reduced to a filler or hesitation marker, in more recent studies, its function is usually explained primarily in terms of marking focus. In my presentation I address the topic from a pragmatic variationalist perspective: how do use and function of the pragmatic marker like differ in the north of England and the south of Ireland? Unlike most unconventional forms of usage of like in spoken discourse, like in clause-final position is, surprisingly, not a recent development. It forms part of two sets of regional varieties of English, located in the north of England and the southwest of Ireland. In spite of being a dialectal feature, this specific discourse marker, however, is not used predominantly among older dialect speakers, but is just as much a characteristic of the language of younger people from these regions, as well as used extensively in the modern-day urban areas Newcastle and Cork.

In my corpus-based study, I show first, what discourse marking functions clause-final like has in spoken discourse, completing the network of meanings already assembled in previous research on the discourse marker like in other positions, and second, analyse its distribution according to the sociolinguistic variables of age, gender and level of education. My findings will then be contrasted with those for clause-initial and internal position.

An Extended Method for Iterative Error Mining in Parsing Results
Jianqiang Ma, Daniël de Kok and Gertjan van Noord (University of Groningen)

Error mining is a technique that detects incorrect or incomplete linguistic descriptions that cause incomplete parsing of sentences in natural language parsing systems. Van Noord (2004) proposed a method for finding suspicious n-grams of arbitrary length. Sagot and de la Clergerie (2006) described an extension that uses an iterative process to gradually shift blame to specific unigrams or bigrams, rather than blaming each form occurring in an unparsable sentence. De Kok and Van Noord (2009) generalized the method beyond the scope of unigram/bigram by producing n-grams that are as long as necessary to identify problematic forms, but not longer. All the above approaches define errors as n-grams of specific words. This fails to account for patterns that include more abstract features such as part of speech tags and lemmas. Besides n-grams of words, we introduce hybrid n-grams of words, part of speech tags and lemmas in the iterative process, in the hope of discovering interesting patterns. This presentation describes the new method and discusses the results in comparison with previous approaches.

The Effect of Bilingual Immersion Programmes on Early Childhood Identity
Beth Martin (University of New England / PHORMS Management AG)

Within the social sciences, in particular sociolinguistics, research involving bilingual programmes for children such as immersion programmes or two-way bilingual programmes, although abundant, has largely focused on the outcomes of such programmes, i.e. the cognitive changes or academic achievements and levels of language proficiency attained. There has been a paucity of studies investigating what involvement in such programmes means to the participants or, in other words, what the act of becoming bilingual has meant to the child learner and his or her identity. This paper then presents the findings from research conducted on the effect immersion situations have on the identity of children of primary school age. Supporting a chaos theory perspective on identity, the study analyzed the way the children experience their own linguistic identities by using language silhouettes and questionnaires and then comparing results from normal state school systems, which make no provision for mother tongues other than German, with the immersion programme run by the PHORMS Schools in Berlin, Germany.

Two types of expletive negation
Alice Middag (Leiden University)

The term ‘expletive negation’ refers to a negative marker that occurs in a construction but does not contribute a negative interpretation to it. In (1a), French ne ‘not’ does not negate the sentence as it does in (1b) when combining with pas ‘not’.

(1)	a.	Je	crains	que	Marie	ne	vienne
		I	fear	that	Mary	not	comes
		‘I’m afraid Mary will come’

(1)	b.	Je	crains	que	Marie	ne	vienne	pas
		I	fear	that	Marie	not	comes	not
		‘I’m afraid Mary will not come’

In a language like Dutch, negation cannot be expletive in such constructions: in (2) negation only has negative force.

(2)	Ik	ben	bang	dat	Marie	niet	komt
	I	am	afraid	that	Mary	not	comes
	‘I’m afraid Mary will not come’

This does not mean that Dutch niet is unable to express expletive negation – in the rhetorical question (3) it does.

(3)	Hoe	vaak	heb	ik	het	al	niet	gezegd
	How	often	have	I	it	already	not	said
	‘How many times didn’t I tell you!’

Interestingly, French ne ... pas also expresses expletive negation in this context:

(4)	Que	de	fois	n’a-t-il	pas	couru	de	risques	inutiles
	How	of	times	n:has-he	not	run	of	risks	unnecessary
	‘How many times didn’t he run unnecessary risks’

Basing myself on research in six languages, I advance two hypotheses that account for these observations:

- There are two types of expletive negation (contra Espinal (1997) and Brown (1999))
- The type illustrated in (3)/(4) occurs in all languages; that in (1a) does not

Several arguments support this analysis. First, there is a semantic difference between the two types of expletive negation: the first (1a) expresses negative hope (cf. Muller (1991) and Rooryck (to appear)), whereas the second (3)/(4) expresses the affirmation (not the negation!) of the sentence. Second, type one only occurs in subordinate clauses, whereas type two occurs in main and subordinate clauses. Finally, a correlation between the first type of expletive negation and Negative Concord (two instances of negation expressing a single negation together) shows that there are two types of expletive negation that behave differently.

References

Brown, S. 1999. The syntax of negation in Russian. Stanford: CSLI Publications.

Espinal, M.T.1997. Non-negative negation and wh-exclamatives. Negation and polariy. Forget et al. (eds) Amsterdam: J. Benjamins.

Muller, C. 1991. La négation en français. Geneva: Droz.

Rooryck, J. To appear. On the scalar nature of syntactic negation in French.

An Evaluation of Phrase-based SMT and Finite State Transducer models for Translating Transliterations
Peter Nabende and Jörg Tiedemann (University of Groningen)

Transliteration is used to obtain phonetic equivalents in a target language for a given source language word. However, transliteration often results in different spellings in target languages using the same writing system. A problem that has not been addressed in the literature at least to our knowledge is that of automatically translating transliterations. Translating new entity names is important for improving performance in Natural Language Processing (NLP) applications such as Machine Translation (MT) and Cross Language Information Retrieval (CLIR). We attempt to address the problem of translating entity names whose origin is in a language using a different writing system. We train and evaluate various Weighted Finite State Transducers (WFSTs) and state of the art Phrase-based Statistical Machine Translation (PSMT) models for a characterbased translation of the transliterated entity names. In particular, we evaluate the WFST and PSMT models on Russian person names between Dutch and English, and between English and French. From our experiments, PSMT models perform best with consistent improvements compared to a baseline method of copying strings.

Vowel reduction in Dutch: an alternative perspective
Aleksei Nazarov (Leiden University)

The process of vowel reduction in Dutch optionally reduces full vowels in weak prosodic positions to schwa. This (lexically governed) process is sensitive to differences in vowel identity and prosodic positions, so that the are some possible forms where not all unstressed vowels are reduced. Reduction correlates with stylistic factors: the more reduction, the more informal the style.

(1) fònoloɣí ~ fònəloɣí ~ fònələɣí (but *fònoləɣí) - ‘phonology’

The standard interpretation of these data is that reduced and unreduced forms all have the same metrical structure, and vowel reduction is a purely segmental process. Kager (1989) even calls vowel reduction a ‘window into (Dutch) metrical structure’. In the current proposal, however, this assumption is challenged. A model is presented where reduced and unreduced forms differ in their metrical structure, and the segmental difference between full vowels and schwa is a consequence of the metrical differences. For this purpose, a slightly modified version of the prosodic hierarchy is presented, where the level traditionally called the metrical foot is split out into two levels: the level of the Superfoot, and the (lower) level of the Foot, where only Superfeet project (primary or secondary) stress; a language-specific assumption for Dutch is that the Foot may contain only one full vowel (this idea had already been presented in Van der Hulst & Moortgat (1981)). This yield different metrical structures for the different realisations in (1) (with Superfeet being indicated by square brackets, and Feet by parentheses):

(2) [(fò)(no)(lo)][(ɣí...)] ~ [(fònə)(lo)][(ɣí...)] ~ [(fònələ)][(ɣí...)] ‘phonology’

An Optimality Theory account of the data will be presented, which models the variation between reduced and unreduced forms by different rankings of constraints concerning the metrical structure (according to a special model of variation in OT). This model naturally predicts the facts about the stylistic value of reduced and unreduced variants. A few advantages of the model presented here are the following. Lexical exceptions to vowel reduction (i.e., words in which vowels never reduce) can be represented using underlying prosodic material (Feet); underlying Feet are already assumed for exceptional stress in Dutch in many approaches - thus, no generic diacritical marks are needed which prevent a word from being reduced. Also, the assumption that Feet are domains with excactly one full vowel allow for a simple account for form restrictions on so-called Germanic stems in Dutch.

References

Hulst, H. van der & M. Moortgat. 1981. Prosodische fonologie en de accentuatie van Nederlandse woorden, of: Leeft het Nederlands op grote voet? Ms., U Leiden/U of Connecticut.

Kager, R.W.J. 1989. A Metrical Theory of Stress and Destressing in English and Dutch. Diss, U Utrecht.

Phi-feature valuing in discourse ellipses
Mari Nygård (NTNU Trondheim)

Spoken Norwegian displays a high frequency of discourse ellipsis. In particular, arguments in topicalized position are vulnerable to omission:

1.	Ø tjente	rått	på	telefonhealing.
	Ø earned	plenty	on	telephone healing.

2.	Har	du	gjort	leksene dine?	Ø gjorde jeg i går.
	Have	you	done	homework?	Ø did I yesterday.

The issue to be addressed in this talk is whether it is plausible to assume a full sentence structure for these cases, and in particular, what this structure contains. More specifically, I discuss the presence of phi-features. I suggest a separationist analysis where grammatical features are in principle torn apart from lexical concepts (Harley & Noyer 1999, among others). Hence, phi-features are assumed to be present in syntactic nodes independently of lexical insertion. Support for this view comes from examples where anaphors (4), verbs (5) and predicatives (6) show agreement with null subjects:

3.	Ø har	valgt	å	ikke	trene	meg	i	hjel.
	Ø have	chosen	to	not	exercise	myself	to	death.

4.	Ø suis	tellement	énervée	que	Ø me	suis	assise	sur	la télécommande.
	Ø am	so	nervous	that	Ø me	am	seated	on	remote control.
	(Haegeman & Ihsane 2001)

5.	Ø Ø slitne	etter ferien.
	Ø Ø tired (pl)	after the holiday.

Another group of examples which point in the same direction are sentences displaying so called semantic agreement (6, 7), where there is an apparent mismatch between the number features of the subject and those of the verb (see also Den Dikken 2001, Sauerland & Elbourne 2002, Josefsson 2006):

6.	Peter and Mary travel in Royal circles.
7.	Politiet er snille.
	The police (sg) are nice (pl).

It appears that in these sentences, the agreement process is sensitive to semantic information about plurality or collectivity of the subject DP. Hence, both these groups of data demonstrate a separation between visible morphology on lexical items on the one hand, and abstract grammatical features on the other. I therefore propose a unified analysis for the two groups, where undefined phi-feature matrixes are linked to syntactic positions. These matrixes are valued depending on what lexical items are inserted, or in certain cases depending on how these items are semantically conceived of. For the discourse ellipses, this entails that phi-features are present independently of lexical insertion, and that they are valued from information from the conceptual-intentional interface.

References

Chomsky, Noam (2001) “Derivation by phase,” in M. Kenstowicz (ed.), Ken Hale: A Life in Language, Cambridge, MA: MIT Press, 89-155.

Den Dikken, Marcel (2001) ““Pluringulars”, pronouns and quirky agreement” in The Linguistic Review 18, 19-41.

Haegeman, Liliane and Tabea Ihsane (2001) “Adult Null Subjects in the non-pro-drop Languages: Two Diary Dialects”, in Language Acquisition, 9(4), 329-346.

Harley Heidi and Rolf Noyer (1999) “Distributed Morphology” in GLOT International Volume 4 Issue 4.

Hornstein, Norbert, Jairo Nunes and Kleanthes K. Grohmann (2005) Understanding Minimalism, Cambridge: Cambridge University Press.

Josefsson, Gunlög (2006) “Semantic and grammatical genders in Swedich – independent but interacting dimensions” in Lingua 116, 1346-1368.

McShane, Marjorie J. (2005) A Theory of Ellipsis, Oxford: Oxford University Press.

Sauerland, Uli and Paul Elbourne (2002) “Total Reconstruction, PF Movement, and Derivational Order” in Linguistic Inquiry, Volume 33, Number 2, Spring 2002: 283-319.

An Evaluation of TermPedia on Data from EMEA Reports and the Merck Manuals
Proscovia Olango and Gosse Bouma (University of Groningen)

TermPedia is a supervised document enrichment tool which is designed to link a technical term to an encyclopedia entry, while taking into account the meaning of that term in context. TermPedia was trained on medical English data from Wikipedia. Wikipedia is a free open content online encyclopedia created through collective effort of a community of voluntary contributors. In an effort to evaluate the performance of TermPedia, data from European Medicines Evaluation Agencies (EMEA) reports and the Merck manuals online library were used.

If a technical term occurs in a document without any definition or explanation, it may hinder the understanding of that document by users who are not experts in the knowledge domain that the document discusses. Even if a term is defined in a document, the definition may contain other technical terms that require explanation. Hopefully by linking terms to encyclopedias, not only definitions, but also explanations of the terms can be provided and this shall provide sufficient information for document understanding. Furthermore, by integrating this information as hypertext links in the text, term definitions and explanations are brought just a "click" away. Linking terms to an external resource like an encyclopedia should automatically take into account the fact that terms can be single words as well as phrases, that terms in a text may differ in spelling from entries in an encyclopedia, and that terms may be ambiguous.

We collected all internal links from Wikipedia; this is to say, links from one Wikipedia page to another Wikipedia page. These links tell us which words and phrases have been used to point to specific Wikipedia pages. This information has been used to train a system for automatic term recognition. Term sense disambiguation is performed by either selecting the most frequent target page for a term, or by using contextual information.

TermPedia was tested on data from two medical sources in order to evaluate the performance of the system on data from outside Wikipedia. We evaluated the performance of the term recognition and term disambiguation processes. Results will be delivered in the presentation.

Some remarks on the aspectual properties of complex predicates with light verbs and deverbal nouns
Fátima Oliveira, Fátima Silva, Purificação Silvan, Luis Filipe Cunha and Idalina Ferreira (Universidade do Porto)

The main goal of this paper is to describe some semantic properties of the light verbs fazer (‘make’), dar (‘give’) and ter (‘have’) occurring with deverbal nouns in European Portuguese. In order to do so, we will take into account the interaction between the light verbs, the aspectual classes of the verbs from which the nouns are derived and the final interpretation of these complex predicates.

In this type of structures, the light verb ter (‘have’) differs from the other two because it shows more flexibility in the combination with nominalisations derived from all aspectual classes. Nonetheless, the complex predicate in which this light verb participates must always be followed by modification. Furthermore, they often convey two different final interpretations (for instance, resultative or eventive) depending to a great extent on the selected adjectival modification, as illustrated by (1) and (2).

(1)	O João teve um discurso emocionado.
	João had an emotional speech.
(2)	O João teve um discurso útil.
	João had a useful speech.

The light verb fazer (‘make’) is compatible with processes, culminated processes and culminations (cf. Moens (1987)). Typically, it does not require modification. Moreover, this light verb preserves or assigns a complex structure to the final interpretation. This is the reason why fazer (‘make’) cannot occur with points.

In contrast to fazer (‘make’), dar (‘give’) is much more restrictive insofar as it can not be combined with culminated processes, states, some processes and some culminations. This light verb usually selects situations with a simpler phase-structure, characteristic that justifies the co-occurrence with points.

The characteristics adduced above for fazer (‘make’) and dar (‘give’) are illustrated by the following examples.

(3)	A Maria fez um salto.
	Mary made a jump.
(4)	A Maria deu um salto.
	Mary gave a jump.

In (3), since fazer (‘make’) assigns structure to the situation, “salto” (‘jump’) can only be understood in the context of athletic competition. In its turn, dar (‘give’) maintains the non-durative aspectual profile of the verb from which the deverbal noun is derived.

The complex predicates under consideration seem to be a kind of a “jungle” (Butt (2003)), because of the different possibilities of combinations with deverbal nominalisations and because of the different variables that contribute to the final reading (cf. Duarte et al. (2006)). Nevertheless, this paper puts forward some ideas concerning the role played by aspect that can contribute to a better understanding of this type of complex predicates.

References

Butt, M. (2003) “The Light Verb Jungle”, in Harvard Working Papers in Linguistics, 9, p. 1-49.

Duarte et al. (2006) “Verbos leves com nomes deverbais em português europeu”, in Textos seleccionados do XXI Encontro Nacional da Associação Portuguesa de Linguística, Lisboa: APL, p. 315 – 328.

Moens, M. (1987) Tense, aspect and temporal reference, PhD dissertation, University of Edinburgh, Centre for Cognitive Science.

Remnant movement in a world without traces
Dennis Ott (Harvard University)

Remnant movement (RM) plays a central role in Germanic syntax (Müller 1998). Abstractly, RM constructions have the following structure, with a trace occurring in the fronted “incomplete” XP:

(1) [_XP ... t_Y ... ] ... Y ... t_XP ...

Beyond topicalization, RM has been employed in the analysis of a wide array of phenomena, such as verb movement (e.g. Nilsen 2003, Koopman & Szabolcsi 2000), and more generally as a means of accounting for word-order variation (Kayne 1994, Mahajan 2003) and/or eliminating head movement (Platzack 2009) and covert movement (Kayne 1998). Overall, while some of these analyses have been criticized (Fanselow 2002, Zwart 2003), there is robust evidence for the existence of structures like (1).

From a theoretical perspective, however, RM is notoriously problematic. Even when familiar problems concerning unbound traces, “shape conservation”, and Müller’s Generalization (Müller 1996) are set aside, it is unclear how a configuration like (1) would even arise under the Copy Theory of Movement (Chomsky 1993 et passim), according to which (1) is syntactically represented as in (1'):

(1') [_XP WYZ ] ... Y ... [_XP WYZ ] ...

It is not at all clear how to formulate RM once traces are abandoned as representational devices, a corollary of the idea that movement reduces to Internal Merge (Chomsky 2004): plainly, Internal Merge does not create “remnants”. Notice also that it is unclear how identical copies in RM constructions are linearized in the right way, given that there is no appropriate c-command relation (Gärtner 1998).

Proposals for deriving RM effects in a copy framework include Distributed Deletion ((2a); Fanselow & Cavar 2002) and movement of phase edges ((2b); Müller 2004, Bentzen 2007):

(2)	a.	[_XP WYZ_[Top] ] ... [_XP WY_[Foc]Z ] ...
	b.	[_XP W [_X' X⁰YZ ] ... [_XP W [_X' X⁰YZ ] ... (X⁰ a phase head)

This talk will shed light on the question how RM can be formulated in a framework based on Internal Merge rather than Move-α/trace theory. To this end, I will discuss to what extent RM can be reformulated as sketched in (2) as well as arguments purporting to show that both variable deletion and “true” RM are part of grammar (Hinterhölzl 2002). While the main purpose of this talk will be to highlight precisely the problems posed by RM, I will make some tentative proposals for rethinking the relevant analyses while preserving restrictiveness in the theory of movement (Abels 2007).

Because she said so: On the multifunctionality of direct speech in the jury room
Esther Pascual (University of Groningen)

This paper deals with the use of direct speech for purposes other than genuine quotation. Indeed, the direct speech construction may be used to introduce actual or so-called ‘constructed dialogue’ (Tannen 1986, 1989). I show that the direct speech construction may serve to present not only types of thoughts, emotions and attitudes, as assumed in the literature, but (at least) also inferences, decisions, intentions, and actions. This is claimed to be mainly motivated by a (perhaps universal) cultural model of language as informational (Sweetser 1987), establishing a fundamental link between saying, believing, and truth; as well as the understanding of talk-in-interaction as indicative of the utterer’s mental, emotional and behavioral world (Cicourel 1973). Various instances of direct speech, such as following verbs of saying or thinking, and in isolation, are examined. The construction appears as a fuzzy category within a continuum of various types of demonstrations (Clark & Gerrig 1990). Non-genuine quotations are regarded as involving what I have called fictive interaction (Pascual, 2002, 2006). This constitutes a non-genuine channel of communication, typically not involving a literal or loose quotation. Fictive interaction –for instance through direct speech– may thus serve to set up interactional as well as non-interactional realities, such as mental, emotional or attitudinal states. Attested instances of intra-sentential fictive interaction: “the attitude that, yes, I can do it”; “the attitude of yes, I can do it”; and “the ‘yes, I can do it’ attitude”.

This study is based on the full transcript of a real-life jury deliberation in a death-penalty murder trial recorded and broadcast by an American television station in 2004.

References

Cicourel, A.V. 1973. Cognitive Sociology: Language and Meaning in Social Interaction. Harmondsworth: Penguin Education.

Clark, H.., Gerrig, R.J. 1990. Quotation as demonstration. Language 66(4):784–805.

Pascual, E. 2002. Imaginary Trialogues: Conceptual Blending and Fictive Interaction in Criminal Courts. Utrecht: LOT.

Pascual, E. 2006. Fictive interaction within the sentence. Cognitive Linguistics 17(2): 245-267.

Sweetser, E. 1987. The definition of lie. In: Holland, D. and N. Quinn (Eds.). Cultural Models in Language and Thought. Chicago: UCP.

Tannen, D. 1986. Introducing constructed dialogue in Greek and American conversational and literary narratives. In: Coulmas, F. (Ed.). Direct and Indirect Speech. Berlin: Mouton.

Tannen, D. 1989. “Oh talking voice that is so sweet”: Constructing dialogue in conversation. In: Talking Voices. Cambridge: CUP.

Conflicting cues in early word learning: 14- and 24-month-olds’ use of gaze and point information to map words to referents
Markus Paulus and Paula Fikkert (Radboud University Nijmegen)

In recent years the question of how infants map words to referents has received considerable attention. It has, for example, been suggested that early in development perceptual saliency and temporal contiguity play an important role in word-referent mapping, whereas later in development social cues like eye gaze gain more importance (Hollich et al., 2000; Moore et al., 1999). In teaching words, however, caregivers use diverse social cues and the impact of the different cues on infants’ word learning remains unclear. Furthermore, infants’ sensitivity for the various social cues could also change during development. Whereas infants, for example, follow gazes from 3 to 6 months on (Entremont et al., 1997), they start to reliably follow pointing gestures around the beginning of the second year of life (Morissette et al., 1995).

Using an eye-tracking procedure we investigated whether 14- and 24-month-old infants rely more on an actor’s gaze or on an actor’s pointing behavior when mapping a new word to a referent. To this end infants observed a model on an eye-tracking screen gazing at an unknown object and pointing at another unknown object while teaching the infants a new word (tan). This action was repeated ten times. After the 6th and the 10th trial participants received two test trials. On the test trials infants were repeatedly asked where the tan is. Based on findings that language input influences looking behavior (cf. Tanenhaus & Trueswell, 2006) we expected that infants would look longer to the referent of the newly acquired word.

Preliminary analyses reveal that the 14-month-old infants showed no systematic looking behavior to one of the two objects. The 24-month-old children looked significantly longer to the object the model has been pointing to when teaching the new word.

The results suggest that the 14-mont-old infants do not show a preference for pointing or gazing when learning new word-referent mappings. This could be due to younger infants’ general preference for saliency cues and neglect of social cues in early word learning (Hollich et al., 2000). The 24-month-olds, however, show a clear preference for using the pointing cue. This could indicate that over the second year of life infants pay more attention to culturally acquired cues like pointing (cf. Wittgenstein, 1953) than to presumably from very early on working mechanisms like automatic gaze following (see Entremont et al., 1997) when learning about the referent of new words.

References

D’Entremont, B., Hains, S.M.J., & Muir, D.W. (1997). A demonstration of gaze following in 3- to 6-month-olds. Infant Behavior and Development, 20, 569-572.

Hollich, G., Hirsh-Pasek, K., & Golinkoff, R. (2000). Breaking the language barrier: An emergentist coalition model of word learning. Monographs of the Society for Research in Child Development, 65, 1-138.

Moore, C., Angelopoulos, M., & Bennett, P. (1999). Word learning in the context of referential and salience cues. Developmental Psychology, 35, 60-68.

Morissette, P., Ricard, M., & Gouin-Decarie, T. (1995). Joint visual attention and pointing in infancy: a longitudinal study of comprehension. British Journal of Developmental Psychology, 13, 163-75.

Tanenhaus, M.K., & Trueswell, J.C. (2006). Eye movements and spoken language comprehension, in M. Traxler & M. Gernsbacher (Eds.), Handbook of Psycholinguistics: second edition (pp. 863-900). Academic Press, Elesevier: New York.

Wittgenstein, L. (1953). Philosophical Investigations. Oxford: Basil Blackwell.

Embodied verbal learning? Evidence for motor simulation in verbal knowledge acquisition
Markus Paulus, Oliver Lindemann and Harold Bekkering (Radboud University Nijmegen)

Recent research suggests that language processing automatically activates modality-specific subsystems and interferes with perceptual and motor processes (see Fischer & Zwaan, 2008, for a review). Several theorists in the field of cognitive psychology and psycholinguistics have thus proposed that our propositional knowledge about, for example, actions and objects is bodily grounded in sensorimotor experiences (Barsalou, 2008). Following these so-called embodied cognition approaches, it is assumed that the processing of verbal descriptions about functional objects consists in a covert simulation of associated motor programs and a mental re-enactment of the objects’ functional use. Furthermore, neuroimaging studies of language processing have shown a somatotopically organized, i.e. effector-specific, pattern of activation in premotor cortex for words denoting actions that are related to different body parts (e.g., Hauk, Johnsrunde, & Pulvermüller, 2004).

However, it is unclear whether the activation of motor representations is indeed necessary for language comprehension or if the activation of the motor system is merely a byproduct of an amodal information processing (Mahon & Caramazza, 2008). If perceptuomotor simulations are indeed necessary for language comprehension and verbal learning we would expect that the verbal acquisition of novel functional object knowledge should be affected by a concurrent motor task but not by another attention-demanding task. Furthermore, when simulations are effector-specific we would expect that verbal knowledge acquisition of hand-related objects (e.g., a tool that you use with your hands) should be selectively impaired by a hand-related but not by a foot-related motor task.

We tested this prediction by varying the presence of motor interference in four conditions (squeezing a ball with hands vs. squeezing a ball with the feet vs. oddball detection task vs. no secondary task; each n=16) while participants verbally acquired functional object knowledge. Subsequently we examined the effects on an object detection task, which has been shown to be sensitive to functional object knowledge (van Elk, van Schie, & Bekkering, 2008). Results revealed that learning of functional object knowledge was only impaired when participants performed an effector-specific motor task while training, i.e. squeezing the balls with their hands, as expected by the embodied cognition hypothesis.

The present finding demonstrates the crucial role of the motor system in the verbal acquisition of novel object knowledge and provides support for an embodied account to language processing.

References

Barsalou, L.W. (2008). Grounded cognition. Annual Review of Psychology, 59, 617-645.

Fischer, M.H., & Zwaan, R.A. (2008). Embodied language – A review of the role of the motor system in language comprehension. Quarterly Journal of Experimental Psychology, 61, 825-850.

Hauk, O., Johnsrunde, I., & Pulvermüller, F. (2004). Somatotopic representation of action words in human motor and premotor cortex. Neuron, 41(2), 301-307.

Mahon, B.Z., & Caramazza, A. (2008). A critical look at the embodied cognition hypothesis and a new proposal for grounding conceptual content. Journal of Physiology – Paris, 102, 59-7

Van Elk, M., van Schie, H.T., & Bekkering, H. (2008). Conceptual knowledge for understanding other's actions is organized primarily around action goals. Experimental Brain Research, 189, 99-107.

Comprehending Particle Verbs: an EEG approach
Vitoria Piai, Marcel Bastiaansen and Rob Schreuder (Max Planck Institute for Psycholinguistics)

Particle verbs (e.g., look up) are verbs formed by two constituents: a main verb and a particle. Dutch particle verbs, (e.g., 'afstuderen' to graduate; 'af' finished, 'studeren' to study) are separable in the sense that the particle may be realized separated from the main verb in certain syntactic contexts (e.g., 'ik studeer af' I graduate). In principle, the Dutch syntax allows any number of words to intervene between the main verb and the particle - observed distances in the Twente Corpus (72 million words), for instance, range from 0 to 15 words. Besides, the meaning of the particle verb may be partially or completely non-compositional (compare 'afstuderen' to graduate vs. 'oppassen' to look out/after, 'op' up, 'passen' to fit). When encountering such particle verbs in their separated version, the processor must store the main verb (e.g. 'pas' fit) and integrate it with the particle (e.g. 'op' up) further downstream. Little is known about when and how this integration is realized and on which factors may influence this process and even less has been empirically tested.

We report an EEG study on the comprehension of Dutch separable particle verbs. Subjects read sentences containing particle verbs while their EEG was recorded. We manipulated a) the "family size" of the main verb (the number of particle verbs that can be formed with one same main verb by combining it with different particles) and b) the actual realization of the particle, which could either be (1) the expected one given the sentential context, (2) a particle forming an existing particle verb yet yielding a semantic violation given the sentential context or (3) a particle forming a non-existing particle verb (i.e., a morpholexical violation, for example 'toe+hangen'). For both manipulations, we found a negative peak around 300 ms which seems to be sensitive to morphological processing. Concerning b), the semantic violation yielded an N400 effect relative to the expected particle, which is in agreement with previous research on standard semantic violations. However, encountering a morpholexical violation yields both an N400 effect and an additional late positivity relative to the expected particle. We discuss the implications of these findings for when and how the integration of the particle with the main verb takes place.

Referential choices in young Thai children’s narratives
Theeraporn Ratitamkul (Chulalongkorn University)

In narrating a story, a child has to constantly make a decision concerning referential choices. A Thai-speaking child, for example, needs to decide at various points in the discourse whether to refer to an entity using a lexical, pronominal, or null form. This study examines how 4-year-old Thai children refer to entities in story-telling and discusses factors influencing their referential choices.

Data came from the Thai Frog Stories (Zlatev & Yangklang 2001), in which 10 Thai-speaking 4-year-olds told a story stimulated by a picture book. Arguments of verbs were coded for linguistic forms (lexical, pronominal, or null), and grammatical roles (S: intransitive subject, A: transitive subject, or O: object), following Du Bois (1987). Contexts in which a referring expression occurred were also specified. Adapted from Jisa (2000), the contexts were INT: introducing, REIN: reintroducing, PROM: promoting, and MA: maintaining.

Four-year-olds were found to use lexical forms most often when they introduced new characters (the INT context). Lexical forms gradually decreased when referents were reintroduced (REIN), promoted to the subject position (PROM), and continued in the subject position (MA), respectively. Overall, lexical forms were prevalent in child speech even when they appeared unnecessary. The distributional pattern of null forms, on the other hand, was in the opposite direction. Their proportion was largest in the MA context, followed by the PROM and REIN contexts, and smallest in the INT context. As for pronominal forms, they were infrequent in children’s narratives and were used only to refer to main characters.

Children also followed the Preferred Argument Structure constraints (Du Bois 1987) in that there were only a small number of clauses with two new arguments and that new arguments rarely appeared in the transitive subject (A) position. However, the number of clauses with two lexical arguments was relatively high, and so was the number of lexical arguments in the A position. This goes against the predictions made by the One Lexical Argument and Non-lexical A constraints.

This study revealed that when choosing referential forms, four-year-old Thai children to a certain extent paid attention to contexts, which were related to the status of referents in the mind of the narrator and his addressee. The overuse of lexical forms showed that children at this age were not yet fully capable of creating coherence by using appropriate anaphoric forms. This is partly because of the complexity of story-telling, which requires that children perform multiple tasks simultaneously.

First Language Thinking: The Representation of Motion in Second Language Writing
Nina Reshöft (University of Bremen)

The linguistic patterns used to describe motion events in different types of languages have been a matter of debate within scientific approaches to the relationship between language and thought. Investigations of lexicalization patterns for motion events in different languages are largely based on a widely-accepted dichotomy proposed by Talmy (e.g., 1985, 1991). According to this typological distinction, the manner of motion is lexicalized by the main verb in satellite-framed languages (e.g., English), while the direction of the movement is encoded in "satellites" outside the verb. By contrast, verb-framed languages (e.g., Romance languages, Turkish) typically encode the direction or of motion in the main verb, while the manner component is optionally expressed by additional phrases outside the verb.

It has been claimed (e.g., Slobin 1996a/b, 2000) that the different linguistic representation of lexical patterns (verb-framed vs. satellite-framed) corresponds to different mental representations of motion events. Speakers of English, for example, seem to pay more attention to the manner of motion, whereas the direction of motion is foregrounded in Romance languages.

The different constructions that are available in each language to describe motion events and the grammatical constraints related to them are said to provide speakers of different languages with different ways of "thinking for speaking" (Slobin 1996a). Accordingly, learners of a typologically different language must learn a different way of thinking for speaking. The present corpus-based study extends Talmy's typological framework to the domain of Second Language Acquisition and shows how the predominant patterns of a native language are represented in written L2 production. Specifically, the present analysis is concerned with the linguistic patterns that are to be found in English texts produced by speakers of French, Italian, and Spanish. It is further shown how languages of the same type (i.e., French, Italian, and Spanish) differ among each other in the way motion events are lexicalized.

The data from the French, Italian, and Spanish subcorpora of the International Corpus of Learner English (ICLE), compared to the patterns found in native English speaker data (British National Corpus), reflect the typical patterns of verb-framed languages in many respects. Descriptions of movement in the two types of corpora differ largely in the variety and amount of different types of motion verbs as well as in the representation of information about the direction of a movement.

References

Granger, Sylviane / Dagneaux, Estelle / Meunier, Fanny (2002): The International Corpus of Learner English. Louvain-la-Neuve: Presses Universitaires de Louvain.

Slobin, Dan I. (1996a): “From "thought and language" to "thinking for speaking"”. In: Gumperz, J. J. & Levinson, S. C. [eds.]: Rethinking linguistic relativity. Cambridge: Cambridge University Press 70-96.

Slobin, Dan I. (1996b). "Two ways to travel: Verbs of motion in English and Spanish". In: Shibatani, Masayoshi & Sandra A. Thompson [eds.], Grammatical constructions: their form and meaning. Oxford: Oxford University Press 195-219.

Slobin, Dan I. (2000): “Verbalized events: A dynamic approach to linguistic relativity and determinism”. In: Niemeier, Susanne and Dirven, René [eds.]: Evidence for Linguistic Relativity. Amsterdam/Philadelphia: John Benjamins 107-138.

Talmy, Leonard (1985): "Lexicalization patterns: semantic structure in lexical forms". In: Shopen, Timothy [ed.]: Language typology and syntactic description. Volume III: Grammatical categories and the lexicon. Cambridge: Cambridge University Press 57-149.

Talmy, Leonard (1991): “Path to realization”. Proceedings of the Seventeenth Annual Meeting of the Berkeley Linguistics Society. Berkeley Linguistics Society, University of California, Berkeley 480-519.

The British National Corpus, version 3 (BNC XML Edition) (2007): Distributed by Oxford University Computing Services on behalf of the BNC Consortium. URL: http://www.natcorp.ox.ac.uk/

Literal word meaning activation during idiom comprehension
Joost Rommers¹, Marcel Bastiaansen^1,2 and Ton Dijkstra³
(¹Max Plank Institute for Psycholinguistics, Nijmegen; ²Radboud University Nijmegen, Donders Institute for Brain, Cognition and Behavior, Centre for Cognitive Neuroimaging, Nijmegen; ³Radboud University Nijmegen, Donders Institute for Brain, Cognition and Behavior, Centre for Cognition, Nijmegen)

Idioms are usually defined as expressions of which the figurative meaning cannot be derived from the literal meaning of their component words. For instance, the idiom "to kick the bucket" has nothing to do with a bucket or with kicking, but rather means to die. Common in everyday conversation, idioms regularly confront our comprehension system with a challenging paradox: although the individual words have a meaning that is irrelevant for the overall meaning of the expression, they have to be recognized nonetheless. We investigated whether this recognition process implies the activation of literal word meanings, as predicted by some models of idiom comprehension.

EEG was recorded while participants read sentences in which the target word was (1) a correct (expected) word, (2) a word that was semantically related to the expected word, or (3) a semantically unrelated word. Both (2) and (3) were semantic violations in the sentence context. The words appeared in either a literal or an idiomatic context. Off-line pretests of the materials indicated that the selected idioms were familiar and opaque, and that idiomatic and literal contexts did not differ in cloze probability. A plausibility judgment task showed that both types of semantic violation were implausible in both types of context.

Based on earlier research (Federmeier & Kutas, 1999), the literal conditions were expected to show a 'graded' N400, with both violations eliciting an N400 effect relative to the correct word, but the semantically unrelated one being of highest amplitude. Regarding idiomatic contexts, N400 amplitude was expected to show the same pattern only if literal word meanings become active. If literal word meanings are not automatically activated during the comprehension of idioms, N400 effects for semantically related and unrelated words should be be of equal amplitude.

This study is among the first to address idiom comprehension with EEG. The results discussed in the talk have consequences for models of idiom processing, but also speak to the general issue of context-dependency of word meaning activation.

Reference

Federmeier, K.D., & Kutas, M. (1999). A rose by any other name: long-term memory structure and sentence processing. Journal of memory and Language, 41, 469-495.

Syntactic structure and activation of semantic informations in noun phrases containing a polysemic adjective.
Audrey Rudel (Université Blaise Pascal)

Our research focuses on the link between syntax and semantics in sense construction of complex expressions containing words with multiple meanings. This paper deals with the French polysemic adjective ‘cher’ within the framework of cognitive linguistics. Firstly, we analyse the semantic structure of this item in order to give a conceptual representation of it. Secondly, we examine the way sense can be constructed in expressions composed of ‘cher’ and a noun. We look for semantic regularities taking into account both semantic and syntactic structures.

The French adjective ‘cher’ presents two main meanings: 'beloved' and 'expensive'. According to our hypothesis, linguistic items are associated, into the mental lexicon of speakers-hearers, with an organized conceptual structure composed of numerous semantic pieces of information.

We can determine different main semantic pieces of information associated with the item ‘cher’:
- [SENTIMENTAL VALUE] for the meaning 'beloved' (affective sense),
- [MATERIAL VALUE] for the meaning 'expensive' (physical sense).

The sentimental and material senses are linked by a common concept which can be [VALUE]. To elaborate the semantic representation of the adjective ‘cher’ in a conceptual perspective, we were inspired by the theory of Cognitive Grammar established by R.W. Langacker (1987) and re-used by D. Tuggy (1993). Thus, the concept [VALUE] shared by the two senses is part of the "schematic sense" of the adjective ‘cher’ and the sentimental and material senses correspond to "elaborations" or "instantiations" of that schematic sense.

Combined with a noun, ‘cher’ shows strong semantic variations: it presents either its affective sense, either its physical sense. It depends on which pieces of information are activated during the process of sense construction. When [SENTIMENTAL VALUE] is selected and activated, ‘cher’ takes its affective sense. When [MATERIAL VALUE] is selected and activated, it has its physical sense. Thanks to the study of many utterances, we bring to the fore the existence of semantic regularities in sense construction of expressions combining ‘cher’ and a noun. These regularities allow predicting which sense of ‘cher’ will be selected and activated. They are especially linked to syntactic parameters such as the position of the adjective within the noun phrase (anteposition or postposition).

To conclude, the semantic-cognitive analysis of the adjective ‘cher’ and the study of its combinatorial with the substantive lead us to underline the complex mechanisms of sense construction and utterance interpretation. Furthermore, this analysis highlights the link between syntactic construction and information structure.

References

Langacker, R.W. (1987), Foundations of Cognitive Grammar, Vol. 1, Stanford: Stanford University Press.

Taylor, J.R. (1992), Old problems: Adjectives in Cognitive Grammar, Cognitive Linguistics, 3-1: 1-36.

Tuggy, D. (1993), Ambiguity, polysemy and vagueness, Cognitive Linguistics, 4-3: 273-290.

Overgeneralization in second dialect acquisition and the role of type frequency
Kathy Rys (University of Antwerp / FWO Flanders)

In this paper we focus on the distinction between token frequency (i.e. the frequency of usage of individual words in language use) and type frequency (i.e. the frequency of occurrence of a linguistic pattern) as explanatory factors in the process of second dialect acquisition. From the literature it appears that both kinds of frequency play a role in language use, language acquisition and language change. Recently, it has been demonstrated that type frequency generally has more explanatory power than token frequency (Baayen, 2001; Baayen & Lieber, 1991; ; Bybee, 2001; Eddington, 2002a,b, 2003). Bybee (2001) argues that the productivity of linguistic patterns is determined by type frequency and not by token frequency (e.g. overregularization of the First Conjugation by French children). Moreover, according to Bybee, the productivity of a morphological pattern is diminished when that pattern applies to very high token frequency items.

In this paper, data on the overgeneralization errors made by standard-speaking children who learn a Flemish dialect as a second language (Rys, 2007) are used to demonstrate that type frequency is the determining factor of the productivity of dialect features. It is the phonological features with the highest type frequency that are overgeneralized most often by children who are acquiring the dialect as a second language. Further, in support of Bybee, we show that high token frequency detracts from the productivity of dialect features. In addition, we will argue that overgeneralizations may sometimes lead to dialect change, and that type frequency is the pivotal factor when this dialect change takes place.

References

Baayen, Harald. (2001). Word frequency distributions. Dordrecht: Kluwer.

Baayen, Harald, & Lieber, Rochelle. (1991). Productivity and English derivation: A corpus study. Linguistics 29: 801-843.

Bybee, Joan. (2001). Phonology and language use. Cambridge: Cambridge University Press.

Eddington, David. (2002a). Dissociation in Italian conjugations: A single-route account. Brain and language 81: 291-302.

Eddington, David. (2002b). Spanish gender assignment in an analogical framework. Journal of quantitative linguistics 9: 49-75.

Eddington, David. (2003). Issues in modeling language processing analogically. Lingua 114: 849-871.

Rys, Kathy. (2007). Dialect as a second language: Linguistic and non-linguistic factors in secondary dialect acquisition by children and adolescents. Unpublished doctoral dissertation, Ghent University.

Lexical semantic relations and linear ordering
Nathalie Schapansky (Simon Fraser University)

This work shows that the lexical semantic relations of hyponymy and holonymy determine the linear ordering of constituents across or within clauses, phrases, or words. Hyponyms must precede hyperonyms and holonyms must precede meronyms. This precedence law governs linearization. Variations in the grammaticalization processes are to be expected.

Lehman and Martin-Berthet (2005) show that, in French, the order antecedent > anaphor is determined by the order hyponym > hyperonym (1) and holonym > meronym (2).

1. Un chat entra. L’animal était malade/*Un animal entra. Le chat était malade.
2. Il regarde l’arbre; le tronc était tout craquelé/*Il regarde le tronc; l’arbre était tout craquelé.

For want of better terms, the hyponym in (1) and the holonym in (2) are called includers, while the hyperonym in (1) and the meronym in (2) are called includees. The includer takes precedence over the includee. This precedence law governs the ordering relation of constituents across clauses, as above, or within clauses, phrases, or words.

90% of the world languages have either a SVO or SOV order (Whaley 1997). In these languages, the subject has been grammaticalised as the includer and the predicate as the includee. The head/complement relation also is determined by the order includer > includee. The includer can be either the head or the complement, as evidenced in Breton, a head first language, for possessive constructions (3) and compounds (4).

3.	a.	Maria hé zad ‘Maria her father = Maria’s father’
	b.	tad Maria ‘father Maria = Maria’s father’

4.	a.	dourgi ‘otter’	from ‘dour + ki = water + dog’
	b.	kidour ‘otter’	from ‘ki + dour = dog + water’

Suffixation, as a preponderant derivational process (Whaley 1997), follows the order includer > includee, whether the base is a hyponym (5a) or a holonym (5b), as shown for Breton.

5.	a.	logod ‘mice’	+	-a (action of gathering)	→	logota ‘to catch, hunt mice’
	b.	logod ‘mice’	+	-enn (unit/part)	→	logodenn ‘mouse’

In conclusion, the precedence law includer > includee determines the linear ordering of constituents. What counts as includer or includee, however, is parameterized across and within languages. This can change overtime–Breton dourgi is an older form while kidour is a more recent one. This precedence law reflects our relation to the world and serves to cement relations from anaphora to compounding. This analysis can be extended to account for other parts of the grammar as well.

References

Lehman, A. and F. Martin-Berthet, 2005. Introduction à la lexicologie: sémantique et morphologie. Paris: Armand Collin.

Whaley, L. J., 1997. Introduction to typology. Thousand Oaks: SAGE Publications.

Variation in long-distance movement constructions: a diachronic perspective
Ankelien Schippers and Jack Hoeksema (University of Groningen)

This talk concerns Dutch historical corpus data regarding long-distance (LD) movement constructions and the variation and change these data show. LD-movement constructions are traditionally analyzed as involving a productive rule that moves an element over clause boundaries (cf. Chomsky, 1977). Recently, some studies have presented corpus material that appears to contradict this analysis (Dąbrowska, 2004; Verhagen, 2005), since their data shows that LD wh-movement constructions almost invariably involve the matrix verb ‘think’ or ‘say’ and a second person matrix subject. This has led to the idea that LD-movement constructions don’t involve a productive rule, but are formed by analogy to the template [wh + think/say + 2^nd pers [that...]].

However, our diachronic corpus data contradicts this: LD-movement is possible with a wide variety of matrix verbs and the pattern [wh + think/say + 2^nd pers [that….]] is only very frequent for wh-movement in recent periods. Moreover, in LD-movement constructions other than questions (e.g. relatives and topicalization constructions), a much wider variety of matrix predicates is attested. These constructions also have different preferences regarding the type of matrix predicate: e.g. weten ‘know’ is most frequent for relatives, while it virtually doesn’t show up in wh-movement constructions. This suggests the choice of matrix predicate is influenced by the semantics of a particular construction and the pragmatics of the speech situation.

We take the fact that the constructions under consideration show such a wide variety of matrix predicates as evidence for a productive rule underlying LD-dependencies. However, we don’t want to deny that frequency plays a role, too. Our data show a strong increase of LD wh-movement constructions relatively to other types of LD-movement over the past few decades. However, the variety of matrix predicates attested in LD wh-movement constructions in this period is very limited, mirroring the findings of Verhagen and Dąbrowska. Furthermore, LD-movement constructions appear to be constrained by semantic and pragmatic effects.

Therefore, we argue that LD-constructions are formed by a productive movement rule, but that the use of this rule is constrained by various non-syntactic factors. These factors all conspire together resulting in the formulaic-like LD wh-movement constructions we find in contemporary corpora.

References

Chomsky, N. (1977). On Wh-Movement. In: P.Culicover, T. Wasow, and A. Akamajian (eds.), Formal syntax. Academic Press: New York.

Dąbrowska, E. (2004). Language, Mind and Brain. Georgetown: Georgetown University Press.

Verhagen, A. (2005). Constructions of Intersubjectivity. Oxford: Oxford University Press.

EPP and the object obligation of particle verbs in Dutch: An acquisitional account
Rianne Schippers (Utrecht University)

Dutch transitive particle verbs obligatorily take an object (compare (1a) and (1b)).

(1)	a.	Jan eet *(de appel) op
		(Jan eats the apple up)
	b.	Jan eet (de appel)
		(Jan eats the apple)

In this talk I will (a) pursue the idea that this obligation is caused by the EPP and (b) show it results from acquisition steps. The account is based on a longitudinal study of the Dutch child Sarah (CHILDES Van Kampen corpus), aged 1;6 till 2;6. I will argue that the object obligation of transitive particle verbs is the result of three consecutive steps in acquisition: (1) the lexical classification in the two word stage, (2) the acquisition of verb second and (3) the acquisition of the EPP.

Step one
In the two word stage there is an overwhelming amount of sentences that have a topic and a predicate. The group of predicative elements contains verbs, adjectives and particles, see (2).

(2)	a.	particle	schoen weg/uit (shoe away/off)
	b.	adjective	pappa lief/vies (daddy nice/dirty)
	c.	verb	pappa slapen (daddy sleep)

At this stage in acquisition, children group the three different categories in (2) together in one lexical group. In order to be able to distinguish the predicates from each other, children need to acquire language specific knowledge.

Step two
Verb second (V2) enables the Dutch-learning child to distinguish verbs from the other predicates in (2). Acquiring V2 also leads to a reanalysis of particle verbs. Prior to the acquisition of V2 both particles and verbs form primary predicates. I argue that after the acquisition of V2, particles form complex predicates with verbs.

Step three
The acquisition of V2 is shortly followed by the acquisition of the EPP (3 weeks). The rise of the EPP reflects an acquisition step for complex predicates. The acquisition of the EPP does not only lead to the subject obligation of verbs, but also to the subject obligation of the other predicates in (2) (Stowell 1983). I propose that because particles form a complex head with verbs, their subject becomes the (obligatory) object of the particle verb. The obligatoriness of the EPP leads to the obligatory presence of the particle’s subject, the particle verb’s object. The result is an object obligation of transitive particle verbs. Due to the new configuration, the subject (as external argument of the simplex predicate) gets an explicit licensing position and becomes obligatory.

Error Analysis: The phonology of German learners of English
Anke-Elaine Schmidt (University of Essen)

Although phonetics and phonology is nowadays an integral part of university education when learning or studying a second language, teaching is usually restricted to how sounds should be produced. Students’ errors in phonology on the other hand are hardly ever analysed and it is often not explained what causes these errors. This, however, is also a result of the lack of empirical research in the area. Most phonetics and phonology books for German learners of English neglect error analysis and only provide a few lines about why students have problems pronouncing eth or theta sounds and occasionally the buzzword ‘terminal devoicing’ appears.

Thus, in my research I have now set out to analyse German students’ L2 pronunciation errors systematically with regard to their quality and quantity. While first-year students of English were asked to read out a text aloud, they were recorded and afterwards had to answer further questions. This material has then been analysed and the errors categorised. It turned out that not only terminal devoicing but especially “fortisation”, i.e. the process of making lenis sounds fortis, is very common among German learners. Furthermore, the fact that sounds in unstressed positions are not reduced and that word stress is often on the wrong syllable accounts for many of the errors. Partly, these errors are due to language interference from German but they also result from similar spelling of other more familiar English words. A detailed study of German learners’ errors together with an in-depth explanation of their origin will help to improve language courses in the future and give foreign students the chance to make their pronunciation as native-like as possible.

Perceptual Learning in Hindi-English Bilinguals
Katharina Schuhmann (Stony Brook University)

This study investigates whether perceptual learning effects that have been reported for monolingual speakers (e.g., Norris, McQueen, & Cutler 2003, Eisner & McQueen 2005, Kraljic & Samuel 2006, Kraljic, Samuel, & Brennan 2008) also occur in bilinguals. The existing literature on perceptual learning only reports experiments conducted in monolingual contexts: Listeners who are exposed to sounds that are ambiguous between for example /f/ and /s/ while performing a lexical decision task in Dutch, show a shifted category boundary in the direction of the training when tested on a continuum between /f/ and /s/ (Norris, McQueen, & Cutler 2003).

Kraljic & Samuel (2006) show that perceptual learning effects on the voicing contrast in dental stops generalize to the voicing contrast in bilabial stops in English speakers. Our study examines whether the generalization of perceptual learning effects from one stop contrast to another instance of the same kind of contrast (as in Kraljic & Samuel 2006) operates at a purely abstract, phonological level or whether language-specific phonetics also play a role. We therefore tested whether perceptual learning in bilinguals carries over to stop contrasts in another language. We exposed Hindi-English bilinguals to sounds that are ambiguous between /k/ and /g/ while they performed a lexical decision task in English. Afterwards, participants categorized ambiguous stimuli of three Hindi stop continua (dental, retroflex, and velar stops) and one English stop continuum (velar stops).

The data from 28 Hindi-English bilinguals analyzed so far suggest that perceptual learning in stops is specific to the language of training. Participants showed marginal perceptual learning effects for English velar stops (p=.100), while they showed no significant perceptual learning effects for either of the three continua involving Hindi stop contrasts.

The missing perceptual learning effects in Hindi are most likely related to the fact that the Hindi phoneme inventory distinguishes four types of stops at each place of articulation, using two contrastive features (voice and aspiration/spread glottis). Naturally, the same phonetic space (e.g., velar stops) is broken down into four contrastive sounds in Hindi (/g^h/, /g/, /k/, and /k^h/), but only two contrastive sounds in English (/g/ and/k/). The results indicate a) that perceptual learning effects do not generalize to an untrained language with a distinct set of phonetic realizations of the voicing contrast in stops, or b) that the overall use of space due to the sound inventory influences perceptual learning effects. Directions for future research will be discussed.

References

Eisner, Frank, & McQueen, James M. (2005). The specificity of perceptual learning in speech processing. Perception & Psychophysics 67:224-238.

Kraljic, Tanya, & Samuel, Arthur G. (2006). Generalization in perceptual learning for speech. Psychonomic Bulletin & Review, 13, 262-268.

Kraljic, Tanya, & Samuel, Arthur G. (2007). Perceptual adjustments to multiple speakers. Journal of Memory and Language 56:1-15.

Kraljic, Tanya, & Samuel, Arthur G. & Brennan, Susan. (2008). First impressions and last resorts. How listeners adjust to speaker variability. Psychological Science 19:332-338.

Norris, Dennis, McQueen, James M., & Cutler Anne (2003). Perceptual learning in speech. Cognitive Psychology, 47, 204-238.

Frequency and grammar in interaction
Marjoleine Sloos (Leiden University)

There are two main approaches in phonology; one is the generative approach, currently mainly in terms of Optimality Theory (McCarthy & Prince 1993/2004), the other is usage-based-oriented phonology, here identified with Exemplar Theory (Bybee 2001 et seq., Pierrehumbert 2002 amongst others). These approaches have quite opposite goals; generative grammar investigates the internal system of language, phonological rules, whereas usage-based phonology investigates e.g. frequency effects and models the lexicon. Recently some articles suggest that frequency can interact with grammar (Anttila 2006, Zhang & Lai 2008), which would be problematic for both OT as well as ET, because, to put it boldly, OT cannot handle frequency effects, whereas in ET grammar is not implemented.

In order to investigate the interaction between frequency and grammar, we performed a production experiment. As a frequency effect we selected pretonic schwa deletion in Dutch, e.g. gelijk [χəlɛɪk] ‘right/justice’ → [χlɛɪk]. Gelijk, a high-frequency word, was matched with low-frequent gelid [χəlɪt] ‘joint’. Schwa deletion was expected to occur more often in high-frequency words than in low-frequency words. Syllabification was selected as a phonological process; stimuli consisted of words that would result in phonotactically ill-formed clusters after schwa deletion and words that would have well-formed clusters: after schwa deletion, gelijk begins with phonotactically well-formed /χl/, but beneden has an ill-formed onset /bn/. The hypothesis was that deletion would not occur when the resulting word would have an ill-formed cluster.

The results show a frequency effect within the category of high-frequency words and, most importantly, this effect only occurred when the target onset cluster was a well-formed cluster. This shows that there is indeed an interaction between frequency effects and grammar. I will show that both OT and ET fail to account for these facts, but that a combined model for OT and ET can capture both the frequency effect as well as the interaction between frequency and grammar.

References

Anttila, Arto. 2006. Variation and opacity. Natural language & Linguistic theory, 24: 893-944.

Bybee, Joan. 2001. Phonology and language use. New York, CUP.

Pierrehumbert, Janet. B. 2002. Word-specific phonetics. In: C. Gussenhoven & N. Warner (Eds.), Laboratory Phonology, Vol. VII. 101-137. Berlin, Mouton de Gruyter.

Prince and Smolensky. 1993/2004. Optimality Theory: constraint interaction in generative grammar. MIT Press, Massachusetts.

Zhang & Lai. 2008. Phonological knowledge beyond the lexicon in Taiwanese double reduplication. WCCFL 27.

Is HCI Research Relevant to the Practice of Natural Language Processing?
Ielka van der Sluis, Nikiforos Karamanis, Anne Schneider, Saturnino Luz, Gavin Doherty and Stephan Schlogl (Trinity College Dublin)

Natural Language processing (NLP) is characterized as the field which aims to provide computers with the ability to process human language [1]. To get computers to perform useful language-related tasks such as conversing with a human, translating a document, answering questions using information from the Web, is the ultimate goal of NLP. In sharing the goal of making human-computer communication more natural NLP is closely related to human Computer interaction (HCI). Although the two fields have common research ancestors [2, 3], only limited interaction between the two fields can be detected [4], [5], [6]. Inspired by a preliminary study by Reiter [7] we performed a bibliometric analysis of cross-references between HCI and NLP.

In a deeper literature review we focus on biomedical NLP (BioNLP), Meeting Browsing (MB) and Natural Language Generation (NLG), three diverse subdomains of NLP. All three areas have clearly defined contexts of human activity where a need for NLP technology has been identified. However, our study reveals relatively little application of HCI methods and research developments. In the area of MB, for instance, despite the fact that meetings have been widely studied in HCI from the perspective of computer supported cooperative work (CSCW), observational methods are rarely employed. This is evident in recent efforts aimed at collecting meeting data undertaken within the MB community. Although meeting data have been recorded, transcribed and annotated at various levels, these data originated either from scenarios carefully designed so as to enable the investigation of linguistic aspects of the activity [8] or from meetings held among the researchers themselves [9]. Observation data on the information browsing task itself is much harder to find. Similar observations apply for BioNLP and NLG.

Although there have been some attempts to use HCI methods in NLP system design the strongest point of contact between HCI and NLP seems to remain the area of evaluation. There is little usage in the NLP community of HCI methods and research that could benefit the design of NLP-based systems. Observational methods, which for example are commonly employed in HCI studies, are rarely applied when data is collected for MB, NLG or BioNLP. There also remain significant gaps in HCI research concerning the design of NLP systems. We argue that understanding and addressing these gaps is necessary in order to make HCI methods relevant to NLP practice.

References

Jurafsky, D., Martin, J.H.: Speech and Language Processing. Prentice Hall Series in Artificial Intelligence (2nd edition) 2008.

Winograd, T.: Shifting viewpoints: Artificial intelligence and human-computer interaction. Artificial Intelligence 170 (2006).

Grudin, J.: Turing maturing: the separation of artificial intelligence and human-computer interaction. Interactions 13 (2006) p54-57.

Ozkan, N., Paris, C.: Cross-fertilization between human computer interaction and natural language processing: Why and how. International Journal of Speech Technology 5 (2002) p135-146.

Dybkjaer, L., Bernsen, N.O.: Usability issues in spoken dialogue systems. Natural Language Engineering 6 (2000) p243-271.

Larsen, L.B.: Assessment of spoken dialogue system usability - what are we really measuring? In: EUROSPEECH 2003. p1945-1948.

Reiter, E.: The shrinking horizons of computational linguistics. Computational Linguistics. 33 (2007) p283-287.

Carletta, J., Ashby, S., Bourban, S., Flynn, M., Guillemot, M., Hain, T., Kadlec, J., Karaiskos, V., Kraaij, W., Kronenthal, M., Lathoud, G., Lincoln, M., Lisowska, A. McCowan, I., Post, W., Reidsma, D., Wellner, P.: The AMI meetings corpus. In: Proceedings of the Measuring Behavior 2005 symposium on "Annotating and measuring Meeting Behavior". (2005).

Janin, A., Baron, D., Edwards, J., Ellis, D., Gelbart, D., Morgan, N., Peskin, B., Pfau, T., Shriberg, E., Stolcke, A., Wooters, C.: The ICSI meeting corpus. In: IEEE International Conference on Acoustics, Speech, and Signal Processing, 2003. Proceedings of (ICASSP '03). Volume 1. (2003) 364-367

Identifying formulaic sequences (chunks) in learner data: methodology
Hana Smiskova and Marjolijn Verspoor (University of Groningen)

This paper will illustrate the process of definition and identification of chunks (formulaic sequences) and provide a rationale for the methodology to be used in a longitudinal study. Taking an emergentist, usage-based perspective to second language acquisition, the study aims to trace the development of chunks over time in two input condition groups: high input and low input. A trend has already been found in the successful use of lexical chunks by learners in high input conditions (Verspoor & Edelenbos 2008).

Chunks have been recognized as an essential aspect of native speaker language. Learners who demonstrate knowledge of chunks are therefore considered successful in approximating native-like proficiency. Different sources, however, acknowledge that formulaic sequences are not easy to identify. Wray (2002) points out that the process of definition and identification is inherently cyclical: first, prototypical, representative samples must be identified in a corpus to enable their effective definition. Identification thus relies less on formal definition than vice versa. Moreover, the precise “borders” of some formulaic sequences are difficult to determine, either by use of native speaker intuition or corpora search engines, as there may be overlaps (even a simple word string like thank you, besides occuring entirely alone, can also be found in longer strings such as thank you very much or thank you very much indeed, Wray 2002). The issue of “chunk borders and overlap” can be accounted for by Cognitive Grammar (Langacker 1991), which perceives grammar and lexicon as lying on a continuum and can therefore account for both a holistic and a partial storage of longer sequences, and Pattern Grammar (Hunston and Francis 2000), which focuses on the association of lexical items into whole constructions.

After presenting a working definition of “chunks” and the identification process to be used in this study, we will argue that longer, semi-fixed formulaic sequences could be a distinguishing factor between the developmental stages of individual learners. While some are clearly capable of using a whole range of chunk types as well as longer stretches of native-speaker like language, others rely heavily on their L1 Dutch constructions. We will also focus on the acquisition stages of these longer, semi-fixed sequences to illustrate differences between different stages of L2 development in individual learners.

References

Hunston, S. and Francis, G. (2000). Pattern Grammar: a corpus-driven approach to the lexical grammar of English. Amsterdam: John Benjamins.

Langacker, R. W. (1991). Foundations of cognitive grammar, vol. II.: descriptive application. Stanford, CA: Stanford University Press.

Verspoor, M. & P. Edelenbos. (2008). Tweetalig Onderwijs: Beter geschoolde leerlingen in 2024. In De Graaff, R & D.Tuin (Eds). De Toekomst van het Talenonderwijs: Nodig? Anders? Beter?. IVLOS: Utrecht

Wray, A. (2002). Formulaic language and the lexicon. Cambridge: Cambridge University Press.

Models of Language Comprehension and Pronoun Ambiguity in Discourse
Ryan Taylor, Laurie Stowe, Gisela Redeker and John Hoeks (University of Groningen)

How do we understand structures with multiple possible interpretations? Different models of on-line language processing make different predictions about ambiguous structures; which model of comprehension is right has yet to be resolved. Pronouns provide fertile grounds for research on ambiguity, and moreover serve to create a better understanding of discourse comprehension. We report on a self-paced reading experiment that tested two models of sentence processing, the Constraint-based model (MacDonald, Pearlmutter, & Seidenberg, 1994; MacDonald & Seidenberg, 2006), and the Race model (Pickering, Traxler, & Crocker, 2000; Traxler, Pickering, & Clifton, 1998). These models predict different performance at a point of ambiguity in self-paced reading. The Constraint-based model predicts a competition between potential sentences occurs at the point of ambiguity, resulting in longer reading times there, whereas the Race model predicts that one analysis is chosen at the point of ambiguity and reanalysis only occurs at a disambiguating phrase (MacDonald et al., 1994; Pickering et al., 2000; Traxler et al., 1998).

In a self-paced reading experiment, the topic of the discourse was temporarily ambiguous, thus making the pronoun ambiguous, as in (1c):

1.	a.	Andrew showed Paul his time machine.
	b.	Paul sent Andrew back in time.
	c.	He wrote from his time machine about what happened. (N.B. original materials were in Dutch)

This ambiguity was resolved with a later adverbial phrase (i.e., From his time machine). The Constraint-based model predicts a processing delay at the pronoun, whereas the Race model predicts a delay at the disambiguating adverbial. The results are discussed in relation to Constraint-based and Race models in general, with particular attention to their handling of discourse structure.

References

MacDonald, M. C., Pearlmutter, N. J., & Seidenberg, M. S. (1994). Lexical nature of syntactic ambiguity resolution. Psychological Review , 101 (4), 676–703.

MacDonald, M. C., & Seidenberg, M. S. (2006). Constraint satisfaction accounts of lexical and sentence comprehension. In M. Traxler & M. Gernsbacher (Eds.), Handbook of psycholinguistics (2nd ed., pp. 581–611). New York: Elsevier.

Pickering, M. J., Traxler, M. J., & Crocker, M. W. (2000). Ambiguity resolution in sentence processing: Evidence against frequency-based accounts. Journal of Memory and Language, 43 , 447–475.

Traxler, M. J., Pickering, M. J., & Clifton, C., Jr.. (1998). Adjunct attachment is not a form of lexical ambiguity resolution. Journal of Memory and Language , 39 , 558–592.

Testing hypotheses about the neurological mechanisms underlying Childhood Apraxia of Speech (CAS)
Hayo Terband¹, Ben Maassen² and Frank Guenther³
(¹Radboud University Nijmegen; ²University of Groningen; ³Boston University)

Introduction: Childhood Apraxia of Speech (CAS) has been associated with a wide variety of diagnostic descriptions and has been shown to involve different symptoms during successive stages of development. In a previous study (Terband et al., in press), we attempted to associate the symptoms of CAS in a particular developmental stage with information processing deficits by using computational modeling with the DIVA model (e.g. Guenther et al., 2006). The hypothesis was that the speech production system in CAS suffers from poor feedforward control and consequently an increased reliance on the feedback control subsystem. A series of computer simulations accounted for four key characteristics of speech production in CAS: stronger coarticulation, distorted vowel productions, searching articulation, and increased token-to-token variability. In the present study, two possible core deficits of degraded forward control in CAS were investigated: reduced/degraded somatosensory information (H1) and increased levels of neural noise (H2).

Method: In a series of computer simulations with the DIVA model we systematically varied the involvement of these two deficits during the acquisitional processes of babbling and imitation learning. H1 was implemented as noise in the somatosensory and motor state representations (i.e. the cell activations in the model’s somatosensory and motor cortices). In the case of H2, we added noise to the motor, somatosensory, and auditory cortices. The babbling stage was evaluated by the sum-squared Euclidean error of the trained forward model. Imitation learning was evaluated acoustically on four selected key symptoms of CAS. To differentiate the effect of noise during acquisition from the effect of system noise alone, we investigated the effect both deficits have on speech production after asymptotic learning as control conditions.

Results: Babbling: Results showed decreased forward model performance with increased intensity for both deficits. This effect is much larger for H2. Imitation learning: Results showed increased severity of four symptoms of CAS for both deficits - deviant coarticulation, speech sound distortion, searching articulation, and increased variability - but on different scales. The comparison between during and after learning shows a larger difference for H1. Detailed results will be available at the conference.

Discussion and conclusions: Results do not differentiate directly between the hypotheses. However, results imply that for increased levels of neural noise the majority of the symptoms should be attributed mainly to production/execution. Degraded somatosensory information affects the acquisitional processes to a larger extent, causing inherently unstable or deviant motor commands. These findings lead to directly testable predictions for auditory and articulatory perturbation experiments.

Evidence Based Word Alignment
Jörg Tiedemann (University of Groningen)

Automatic word alignment has received a lot of attention mainly due to the intensive research on statistical machine translation. In recent years various discriminative approaches have been proposed (Taskar et al 2005, 2006, Moore 2005, 2006, Liu et al, 2005, Blunsom and Cohn 2006). They require word-aligned training data for estimating model parameters in contrast to the traditional IBM alignment models that work on raw parallel corpora (sentence aligned only).

In this talk, we present another supervised alignment approach based on association evidence indicating relations between words in context. Each piece of link evidence is seen as independent but not mutually-exclusive probabilistic indication and their disjunctive sum is used to make alignment decisions. For example, the alignment evidence E(a|s,t) based on two clues is defined as this:

E(a|s,t) = E(a,r₁|s,t)+E(a,r₂|s,t)-E(a,r₁|s,t)*E(a,r₂|s,t)

Link evidence is based on association features, for example string similarity or co-occurrence. As usual, these features need to be weighted according to their strength in predicting links. The difference to other approaches using, for example, linear combinations of features is that in our approach strong clues have a larger influence on the alignment decisions made. The intuition behind this is that very strong pieces of evidence are hard to overrule by many (other) weak indications. It is sufficient to have only a few but solid clues to create a link. In this way, our approach can be described as a sub-linear combination of high precision features.

The feature weights have to be learned from word aligned training data. The fact that alignment clues are modeled as individual (independent) pieces of evidence makes it possible to optimize there weights independently of each other. (Note that complex features can easily be created in cases where this independence assumption is strongly violated.) This makes learning extremely simple and fast and a wide range of statistical, orthographic and positional clues can be combined in this way. I our experiments we can show that this "evidence-based" approach to word alignment can be used to improve the baseline of statistical alignment and also outperforms a discriminative approach using a maximum entropy model with the same set of features.

References

P. Blunsom, T. Cohn, "Discriminative word alignment with conditional random fields", In Proceedings of ACL, Sydney, Australia, 2006.

S. Lacoste-Julien, B. Taskar, D. Klein, and M. Jordan, "Word Alignment via Quadratic Assignment", In Proceedings of HLT-NAACL, New York, 2006.

Liu, Yang, Liu, Qun, Lin, Shouxun, "Log-Linear Models For Word Alignment", In Proceedings of ACL, Ann Arbor, Michigan, 2005.

Robert C. Moore, A discriminative framework for bilingual word alignment, In Proceedings of HLT/EMNLP, Vancouver, Canada , 2005.

Robert C. Moore, Wen-tau Yih, Andreas Bode: Improved Discriminative Bilingual Word Alignment. ACL 2006.

B. Taskar, S. Lacoste-Julien, and D. Klein, "A Discriminative Matching Approach to Word Alignment", In Proceedings of HLT/EMNLP, Vancouver, Canada, 2005.

The case of North-Occidental Catalan: towards a linguistic convergence with the Standard?
Esteve Valls (University of Barcelona)

During the last decades, the language policy carried out in Catalonia has resulted in a better knowledge of Standard Catalan among the young generations. Some experts state that this situation will lead to: (1) a process of linguistic convergence that will reduce the dialectal diversity within Catalonia; and (2) a process of linguistic divergence that will increase the differences between the Catalan spoken in Catalonia and the varieties used in the other territories of the Catalan linguistic area.

In this presentation, we want to pay attention to the current situation of North-Occidental Catalan, a dialect spoken in the Western area of Catalonia and in the Autonomous Community of Aragon. Our aim is to determine the speed of the linguistic change on the basis of quantitative data. To do that, we will show the results of applying the dialectometric techniques developed at the University of Barcelona to a corpus of linguistic data collected in the area. This corpus, which is about 20.000 items, includes data from rural and urban areas located both in Catalonia and Aragon; besides, the informants belong to four age groups, as we intend to analyze the linguistic change in terms of apparent time. Using this corpus, we would like to provide an answer to the following questions:

(1)	Has the process of linguistic convergence strengthened during the last decades or it is actually slowing down because of the major awareness of its speakers?
(2)	Are the rural areas more resistant to the pressure of the standard language than the urban areas?
(3)	Is it possible to detect the border-effect that arises due to the existence of political borders among these varieties? Do the dialects located in different autonomous communities evolve differently depending on the language policies of Catalonia and Aragon?
(4)	Does the standardization process entails a complete dialectal uniformization or it results in a “dialectalized standard”, i.e., a common variety for all North-Occidental subdialects that still keeps some of the most general features of the dialect?
(5)	Is it possible to use dialectometric techniques to determine the different rates at which morphological and phonological features evolve?

To shed light on these questions, not only will we present the results of the dialectometric analysis of the data, but we will also show some concrete examples of the former signals of linguistic provenance that have now been replaced by the Standard forms.

Grammatical Inventories
Marijn van 't Veer (Leiden University)

1. Introduction
The goal of linguistic theory is to explain both the similarities of, and differences between, human languages. The framework of Optimality Theory attempts to do so by positing the grammar as a language-specifc ranking of universal constraints. The research presented here applies this idea to the realm of the segment inventory. The status of the segment inventory of natural languages is not entirely clear within Optimality Theory (OT). Whereas it is seen as a phonological primitive in some analyses (for example, Flemming, 1995), both Prince and Smolensky (1993) and Kager(1999) treat the inventory as a byproduct of grammar (i.e., a ranking of faithfulness and markedness constraints). The latter solution is arguably more in accordance with the spirit of OT: that which is language specific is derived from the constraint ranking, and no language specific underlying representations are assumed. However, a systematic investigation into how this can be accomplished has never been undertaken.

2. Current Research
The present research demonstrates how segment inventories can be described using a set of very simple constraints: two kinds of Feature Cooccurrence Constraints (FCCs) are involved, and only one type of faithfulness constraint:
Feature Cooccurrence Constraints:

	*[F][G]
	a segment should not be specified for features [F] and [G]

	[F] → [G]
	a segment specified for feature [F] should be specified for feature [G]

Faithfulness constraints:

	IDENT[G]
	for every segment at level α that is specified for F, the corresponding segment at level β must be specified as well (where α and β are input and output levels)

These constraints are realized for every feature, and ranked for every language in the p-base database (Mielke, 2004). Applied to a subset of this database, these constraints result in a two-stratum ranking describing the inventory of the language.

3. Implications
The implications of this approach are interesting in various respects. First and foremost, it is shown that within OT, the segment inventory of natural languages can be seen as an epiphenomenon of grammar, and subject to the same constraints as other linguistic structures. Furthermore, a systematic description of inventories allows for further typological investigations. Next, the description of the inventory in terms of FCCs and ident constraints allows for predictions about the acquisition of the inventory. Finally, this approach to segment inventories raises interesting predictions when it comes to the relative markedness of segments, for example in epenthesis.

References

Flemming, E. (1995). Auditory representations in phonology. Unpublished doctoral dissertation, University of California Los Angeles.

Kager, R. (1999). Optimality theory. Cambridge: Cambridge University Press.

Mielke, J. (2004). Emergent features. Unpublished doctoral dissertation, Ohio State University.

Prince, A., & Smolensky, P. (1993). Optimality theory - constraint interaction in generative grammar. ROA, 1-262.

The building blocks of language: the impact of literacy on speech segmentation
Dorina Veldhuis (Tilburg University)

Current studies in neurolinguistics often take ‘words’ as a basis for research. However, just as gestures do not necessarily refer to ‘words’, studies of language acquisition within the tradition of Cognitive Linguistics have pointed out that the constituent building blocks children use and recognize do not always correspond to conventional words (cf. Tomasello 2006). Similarly, a large number of descriptive and experimental studies from the 1970s and 1980s has shown that children do not ‘automatically’ segment speech into words when they are requested to segment sentences. What is more, these studies have shown that children up to the age of about seven often lack the matalinguistic awareness of what our conventional ‘words’ exactly are, and that this awareness only develops when children get older.

Accordingly, researchers such as Karmiloff-Smith et al. (1996) concluded that age affects children’s metalinguistic awareness, and their awareness of conventional words. Homer (2007) and Kurvers & Uri (2006) suggested that it is probably not age which is the decisive factor in children’s ‘word’ awareness, but rather literacy. As Piaget already suggested that people only start to think according to certain conventions once they have been exposed to it for a while (in Vygotsky 1986), the hypothesis for this research is that children will only be able to segment speech into conventional words once they have been exposed to formal literacy education.

In comparing speech segmentation styles of pre-readers and readers, the influence of literacy on speech segmentation was tested in several pilot studies. Production and perception tasks show that literacy indeed seems a decisive factor in speech segmentation and in the perception of what the basic building blocks of language are.

References

Homer, Bruce D. (2000). Literacy and metalinguistic awareness: a cross-cultural study. (thesis, National library Canada).

Karmiloff-Smith, Annette et al. (1996). Rethinking metalinguistic awareness: representing and accessing knowledge about what counts as word. Cognition, 58. 197-219.

Kurvers, Jeanne & Uri, Helene (2006). Metalexical awareness: development, methodology or written language? A cross-linguistic comparison. J Psycholinguist Res, 35. 353-367.

Tomasello Michael (2006). Acquiring linguistic constructions. In D. Kuhn & R. Siegler (Eds.), Handbook of Child Psychology. New York: Wiley.

Vygotsky, Lev (1986). Thought and language. Alex Kozulin (ed.). Massachusettes: MIT press.

Language analysis as a method to determine national or regional origin in asylum cases
Sanne Verhoef and Maaike Verrips (De Taalstudio, Amsterdam)

When a foreign citizen applies for asylum in the Netherlands, doubts may arise regarding the nationality, regional origin or ethnic background of the applicant, in particular when they cannot show any documents to prove their claims. In such cases, the Immigration and Naturalisation Service can offer the asylum seeker the opportunity to dispel these doubts by means of a language analysis. In a language analysis the language(s) spoken by the foreign citizen as well as his knowledge about his claimed culture and region of origin are evaluated. The outcome of language analyses may be crucial for the Dutch government’s decision whether to grant asylum or not.

In the Netherlands, language analyses are carried out by the Bureau Land en Taal (Office for Country Information and Language Analysis) of the Immigration and Naturalisation Service. Language analyses are based on a recorded interview in which the asylum seeker is asked questions about his background, for example about local customs and names of towns and villages in the area. He is asked to answer extensively while using his own language variety. A language analyst listens to the recording and lays down his findings in a written “language analysis report”. The report discusses the asylum seeker’s knowledge of his claimed region of origin and culture and describes his language use, illustrated with examples of pronunciation, word-choice and grammar. The language analyst generally is a native speaker of the language that is analysed, and not necessarily a professional linguist. He writes his report under supervision of a linguist, who does not necessarily speak the language that is concerned. When an asylum seeker disagrees with the outcome of a language analysis, he has the right to have a contra-expertise (second opinion) carried out by an independent organisation or expert, at his own cost.

In this presentation I will discuss the use of language analysis in the Dutch asylum procedure and some of the concerns that linguists have raised about the quality and reliability of such analyses. These concerns relate mainly to the complexity of the language phenomena involved: multilingualism, language attitudes, and the scarcity of descriptive studies of certain languages.

Diachronic change in subjectivity and stance: conversationalization in journalistic texts
Kirsten Vis, José Sanders and Wilbert Spooren (VU University Amsterdam)

In different areas of research it has been suggested that public discourse is subject to change (Bhatia 2005, Biber 2004, Fairclough 1994, Swales 1990), particularly to the tendency towards a more informal and conversational style, as signalled by Fairclough in studies of public texts such as university prospectuses, political speeches and journalistic texts.

Although this theory is intuitively plausible, there is hardly any empirical evidence. Therefore we put the conversationalization hypothesis to an empirical test in a corpus analysis of Dutch news from 1950 and 2002 (first manually on 30,000 and 50,000 words respectively; second automatically on 2,000,000 words).

Since typical features of informal communication (conversation) are copresence of speaker and listener and self-expression - both aspects of (inter)subjectivity (Clark 1996, Lyons 1994) - in our view conversationalization is best understood in terms of foregrounding and (inter)subjectification. While subjectivity is the foregrounded expression of the locutionary’s agent of herself and her own attitudes and beliefs (Langacker 1990), intersubjectivity can be described as the explicit expression of the speaker’s attention to the ‘self’ of the addressee (Traugott 2003). A means to measure conversationalization lies in the frequency of (inter)subjective linguistic features such as stance (Du Bois 2007).

Diachronic studies investigating stance markers such as stance adverbials, (semi-)modals, and stance complement clauses show that in various registers speakers and writers have become more inclined to express stance, especially over the last 50 years; thus suggesting major shifts in social norms (Biber 2004).

In this study, we extend this stance model of conversationalization with linguistic features of speaker subjectivity and intersubjectivity as found in conversational data by Scheibman (2002). This includes explicit reference to speaker and listener by pronouns, questions, and exclamations. In addition, presentational coherence markers (Mann and Thompson 1988), deictic elements and intensifiers (Pearce 2005) are included. To minimize the effects of the subjectivity of other speakers than the journalist on our results, we separate direct quotations in our corpus.

The results show that, contrary to expectations, explicit reference to speaker and listener seem to have generally decreased in news, when excluding direct quotations. Other markers have generally not changed in frequency. However, direct quotations themselves are used more in 2002 than in 1950. In addition, they have become more subjective. This suggests an unexpected form of conversationalization in news: conversation is not imitated but incorporated more. In this paper these counterintuitive findings are discussed and possible explanations proposed.

References

Bhatia, V.K. (2005). In H. Halmari and T. Virtanen (eds.) Persuasion across genres. Amsterdam: Benjamins, 213-225.

Biber, D. (2004). Historical Patterns for the Grammatical Marking of Stance: A Cross-Register Comparison, Journal of Historical Pragmatics 5 (1): 107-136.

Clark, H. H. (1996). Using language. Cambridge: Cambridge University Press.

Du Bois, J.W. (2007). The stance triangle. In R. Englebretson (ed.) Stancetaking in discourse. Amsterdam: Benjamins, 139-182.

Fairclough, N. (1994). Conversationalization of public discourse and the authority of the consumer. In R. Keat, N. Whiteley and N. Abercrombie (eds.) The Authority of the Consumer. London: Routledge, 253-268.

Langacker, R.W. (1990). Subjectification. Cognitive Linguistics, 1: 5-38.

Lyons, J. (1994). Subjecthood and subjectivity. In M. Yaguello (ed.) Subjecthood and Subjectivity: Proceedings of the Colloquium ‘The Status of the Subject in Linguistic Theory’. Paris: Ophrys, 9-17.

Mann, W.C. and S.A. Thompson (1988). Rhetorical Structure Theory: Toward a functional theory of text organization. Text 8 (3): 243-281.

Pearce, M. (2005). Informalization in UK party election broadcasts: 1966–97. Language & Literature, 14 (1): 65-90.

Scheibman, J. (2002). Point of view and grammar: Structural patterns of subjectivity in American English conversation. Amsterdam: Benjamins.

Swales, J. M. (1990). Genre analysis: English in academic and research settings. Cambridge: Cambridge University Press.

Traugott, E. (2003). From subjectification to intersubjectification. In R. Hickey (ed.) Motives for Language Change. Cambridge: Cambridge University Press, 124-39.

Who's got the floor? - speaking patterns in group conversation
Nynke van der Vliet (University of Groningen)

Group meetings form an important part of many professional activities. They are useful to plan work, solve and highlight problems and share knowledge between people. With increased possibilities of recording and archiving the meeting process, there is a growing need for meeting browsers to enable users to navigate through an archive of meetings. For this information-retrieval task as well as for summarization of meetings, the recorded talk needs to be indexed or annotated. A description at the level of single turns is usually too fine-grained for this; in particular, many turns are not perceived as being part of the main 'floor' of the conversation, but rather as side comments. Therefore, floor can be an important concept for understanding the content and discourse structure of a meeting.

Previous attempts at statistical modelling to predict the floor have been limited to considering patterns of who has spoken previously (e.g. Parker, 1988). Herein, a floor is defined as a pair-wise conversation between two participants of a group conversation. Edelsky (1981) defines floor as a psychologically developed, interactional space among interactants, which is related to the topic of the discourse, communicative actions (how things are being said in the discourse), and the participant's sense of what is happening in the conversation.

This presentation discusses the application of both the floor concepts of Parker and Edelsky to the AMI and ICSI meeting corpora to see if the theory is appropriate for modelling floor in conversation. Parker's floor model is easy to apply and describes the sequence of speaking turns, but it ignores some important aspects of conversation. The application of Edelsky's floor concept shows that a sort of ‘main’ conversation can be distinguished next to other utterances that form the background speech. In this speech there are several types of utterances that have different functions in the turn-taking process in group conversation.

Do Presupposition Triggers Influence the Felicity of Voice Mismatched Sentences?
Margreet Vogelzang and Jennifer Spenader (University of Groningen)

Kehler (2002) and Kehler et al. (2008) have shown that the type of rhetorical relation holding between two clauses influences pronoun interpretation. Pronouns in PARALLEL relations such as (1) have different interpretation preferences than pronouns in CAUSE-EFFECT relations like (2).

(1) Samuel threatened Justin with a knife, and Erin blindfolded him. (PARALLEL)
(2) Samuel threatened Justin with a knife, and Erin stopped him. (CAUSE-EFFECT)

According to Kehler et al. (2008), simple strategies for pronoun interpretation that say e.g. object pronouns prefer object antecedents, work for PARALLEL relations in e.g. (1), but are irrelevant for CAUSE-EFFECT relations like (2).

A further claim of Kehler’s theory is that PARALLEL relations prefer PARALLEL syntactic structures, while for CAUSE-EFFECT relations this is irrelevant. Thus voice mismatches, with a passive clause followed by an active clause, are dispreferred in PARALLEL relations, e,g, (3) and one reason why would be because it make anaphor interpretation difficult because simple strategies are no longer possible.

(3) Justin was threatened by Samuel with a knife, and Erin blindfolded him. (PARALLEL)
(4) Justin was threatened by Samuel with a knife, and Erin stopped him. (CAUSE-EFFECT)

It has been suggested that the presence of a presupposition marker like too might help repair the damage made by the mismatch, making such sentences more felicitous, e.g. (5).

(5) Samuel threatened Justin with a knife, and Erin blindfolded him too. (PARALLEL)

In a study testing Kehler’s theories with Verb Phrase Ellipsis, Frazier and Clifton (2006) unexpectedly found that people disliked syntactic mismatches regardless of whether or not they were PARALLEL or CAUSE-EFFECT. But their examples didn’t have too in them, and all of Kehler’s examples did.

In a sentences judgment task experiment, we ask Dutch speakers to judge the felicity of PARALLEL and CAUSE-EFFECT sentences with and without syntactic voice mismatches. An additional factor is the presence or absence of the presupposition triggers too (ook) and toch, and testing in Dutch is particularly useful because the Dutch trigger toch marks Denial of Expectation, a type of CAUSE-EFFECT relation. Each sentence also contained an object pronoun and we also study how the presence of the presupposition trigger influences the pronoun interpretation. Our prediction is that the presence of the presupposition trigger will improve the mismatch sentences because it offers listeners an alternative means for interpreting the coherence relation.

Right Dislocation and Afterthoughts
Mark de Vries (University of Groningen)

This talk discusses the properties of right-peripheral dislocation constructions in Germanic languages. Examples are (1) and (2):

(1) They were really big, those pizzas.
(2) Something beautiful appeared in the sky, a double rainbow.

The main claims are the following:

A. Two types need to be distinguished: backgrounding right dislocation (1) and afterthoughts (2). The differences, however, are essentially semantic/pragmatic and phonological, but not necessarily syntactic. In (1), those pizzas is destressed, and forms a prosodic unit with the matrix. In (2), on the other hand, a double rainbow is assigned an independent intonation contour containing a pitch accent. Semantically, those pizzas in (1) functions as a reminder. By contrast, a double rainbow in (2) provides new information. As for the syntax, I argue against a movement analysis of backgrounding right-dislocation (cf. Averintseva-Klisch 2008). To this end, I will present examples from Dutch, German, and English showing a lack of reconstruction effects, and the possibility of non-local construal (cf. Zwart 2001 and De Vries 2007).

B. Dislocation in the right periphery forms a natural class with apposition. This class can be described as anchored parenthesis: the relevant phrases are analyzed as parenthetical secondary predicates of some anchor in the matrix – often a (pro)nominal. Building on Potts (2007), Heringa (2007), and O’Connor (2008), I argue that there is a secondary proposition that can schematically be represented as <pro-anchor> = <parenthetical material>. Such a copular connection can be attributive or specific¬ational. In (2), for example, it is specificational, but in (3) – from German – it is attributive:

(3)	Maria ging aus wandern, einfach gekleidet.
	‘Maria went for a walk, simply dressed.’

C. Right dislocation and afterthoughts are not constructions in the rigid sense of the word. Especially in spoken language, phrases with the same function can be found in intervening positions. This is a relevant difference with left dislocation. In (3), einfach gekleidet can be shifted to the middle field. Another example is (4) – from Dutch –, where daar is equated to in Groningen:

(4)	En daar heeft Jan, in Groningen (bedoel ik), zijn toekomstige vrouw leren kennen.
	‘And there, in Groningen (I mean), Jan got to know his future wife.’

Thus, once we recognize the concept of anchored parenthesis, we can generalize over a number of apparently different constructions.

References

Averintseva-Klisch, Maria (2008). Rechte Satzperipherie im Diskurs: Die NP-Rechtsversetzung im Deutschen. Ph.D. Dissertation, University of Tübingen.

Heringa, Herman (2007). Appositional constructions: coordination and predication. Proceedings of the Fifth Semantics in the Netherlands Day, Marlies Kluck & Erik-Jan Smits (eds.), 67–82. Den Haag: RS Staten Kopie.

O’Connor, Kathleen (2008). Aspects de la syntaxe et de l’interprétation de l’apposition à antécédent nominal. Ph.D. Dissertation, Université Charles de Gaulle – Lille 3.

Potts, Christopher (2007). Conventional implicatures, a distinguished class of meanings. In The Oxford Handbook of Linguistic Interfaces, Gillian Ramchand and Charles Reiss (eds.), 475-501. Oxford: Oxford University Press.

de Vries, Mark (2007). Dislocation and backgrounding. Linguistics in the Netherlands 2007, 235-247.

Zwart, Jan-Wouter (2001). Backgrounding (‘right dislocation’) in Dutch. Manuscript, University of Groningen.

Paths in L1 acquisition of verb second – on the role of input and frequency
Christian Waldmann (Växjö University)

This presentation deals with similarities and differences in L1 acquisition of verb second (V2) among Swedish children. In Swedish, V2 means that the finite verb must be the second constituent in main clauses (1). In subordinate clauses, the finite verb occurs in a lower position, following sentence negation (2).

(1)	a.	Igår	ville	Per	inte	handla	julklappar.
		yesterday	wanted	Per	not	shop	Christmas-gifts

(1)	b.	*Igår	Per	ville	inte	handla	julklappar.
		yesterday	Per	wanted	not	shop	Christmas-gifts

(2)	...att	Per	inte	ville	handla	julklappar	igår.
	that	Per	not	wanted	shop	Christmas-gifts	yesterday

This study investigates the acquisition of V2 among 4 monolingual Swedish speaking children (1;6–3;0), focusing on their input as well as their production.

As regards the input for V2, all children are exposed to an equally large and stable amount of input for V2. Lexically, the input for V2 is characterized to a great extent by the same finite verbs and initial constituents.

Although exposed to a quantitatively and qualitatively similar input for V2, not all 4 children acquire V2 alike. 3 children apply V2 consistently in the earliest finite utterances, finite verbs are rarely misplaced. No lexical limitations are observed.

The fourth child, called Tea, makes abundant verb placement errors up to an age of 3;3, a period during which she applies V2 only sporadically. It is however argued that Tea’s early V2 consists of imitated chunks and that V2 is not applied systematically until just before 3;0. Moreover, it is argued that Tea does not misplace finite verbs randomly, but that her verb placement in main clauses develops systematically in 3 phases: from a low verbal position following sentence negation, as in subordinate clauses (see (2) above), via a verbal position in the middle field of the clause, as in (1b) above, to the second position, as in (1a) above. The transition to V2 happens abruptly between 3;3 and 3;4.

As regards the theoretical implications of my results, I discuss the role of input and frequency in acquisition, arguing that neither copying of input patterns based on frequency nor triggering parameters can capture Tea’s verb placement. Instead it seems that Tea formulates and evaluates different hypotheses, a strategy which allows children to take different paths in acquisition. Furthermore, I explore the relation between verb placement in subordinate clauses in the input and Tea’s deviating verb placement in main clauses.

Phonological Correlates of Emphasis in Baghdadi Arabic
Islam Youssef (University of Tromsø)

In this paper, I argue that phonological activity is dependent on the structure of the contrast system of a given language—following the Toronto school of contrast (Dresher, Piggot & Rice 1994 inter alia). Based on the system of contrast, the analysis of language into an economic and exhaustive set of distinctive features is achieved, and the resulting set of natural classes can account for all kinds of phonological behavior. I illustrate this with regard to the interaction of pharyngealization and labialization in Baghdadi Arabic (BA).

In BA, all consonants may occur pharyngealized in syllables containing a back low vowel [ɑ], while their non-pharyngealized counterparts occur in syllables not containing the back low vowel. I take this complementary distribution as evidence that the low vowel [ɑ] has an underlying pharyngealization feature. Furthermore, the restricted distribution of emphatic non-low vowels suggests that pharyngealization is contrastive for the coronals [tˤ, sˤ, ðˤ, rˤ, lˤ]. This defines the natural class of pharyngealization triggers [tˤ, sˤ, ðˤ, rˤ, lˤ, ɑ] by a V-place[dor] feature. Pharyngealization spreads from these segments within the uninflected word domain, but is blocked by non-tautosyllabic high vowels or palatal consonants (Hassan & Esling 2007). These segments constitute the set of blockers, marked with a V-place[cor] feature.

Another unique fact about BA is that in some contexts [i] and [u] exist in complementary distribution (Blanc 1964, Altoma 1969), best manifested in the epenthetic vowel (EV). If the preceding vowel is [u] or [i], the EV would also be [u] or [i] (i.e. vowel harmony). When the first vowel is [a]; the EV may be [i] or [u] depending on the consonantal environment. Interestingly, if the cluster consists of a labial followed or preceded by a velar or emphatic, the EV is always [u]. I argue that these occurrences of [u] can be derived by spreading a V-place[lab] feature (from a labial) and a V or C-place[dor] (from an emphatic or velar).

This simplification of Baghdadi Arabic phonology allows us to account for pharyngealization and labialization as correlates of emphasis. While the feature V-place[dor] defines the natural class of pharyngealization triggers in BA, it also participates in the process of labialization. The proposed set of abstract features provide a unified analysis of multiple phonological processes in the language and are solely justified on overt phonological evidence. Furthermore, this representational analysis fits neatly into a constraint-based model of phonology such as Optimality Theory.

Acquisition of Meaningful Combinations of Words from Psychiatric Texts
Liang-Chih Yu and Chien-Lung Chan (Yuan Ze University)

With the increased incidence of depressive disorders, many psychiatric websites have developed community-based services such as message boards, web forums and blogs for public access. Through these services, individuals can describe their stressful or negative life events such as death of a family member, argument with a spouse and loss of a job, along with depressive symptoms, such as depressive mood, suicidal tendencies and anxiety. Such psychiatric texts (e.g., forum posts) contain large amounts of natural language expressions related to negative emotion, which makes them a useful data resource for building emotion-based applications as well as automatic psychiatric services. For instance, a dialogue system can generate supportive responses like “Don’t worry”, “That’s really sad” and “Cheer up” if it can understand the negative life events embedded in the following input sentences.

		(1) I am very worried about my children’s health.	[Family]
		(2) I broke up with my dear but cruel boyfriend recently.	[Love]
		(3) I argued with my best friend and was upset.	[Social]
		(4) I lost my job in this economic recession a few months ago.	[Work]
		(5) I hate to go to school because my teacher always blames me.	[School]

The main characteristic of the above sentences are the meaningful combinations of words such as <worry, children, health>, <break up, boyfriend>, <argue, friend>, <loss, job>, and <school, teacher, blame>. Such word combinations carry more information than singe words, which are significant features for identifying negative life events. Therefore, the aim of this study is two-fold: (a) to automatically discover meaningful combinations of words from the sentences annotated with negative life events; and (b) to identify the type (e.g. [Family]) of negative life events of sentences using the discovered word combinations.

To discover meaningful word combinations, we incorporate the measure mutual information into a data mining algorithm, called association rule mining, to incrementally derive frequently co-occurred words in sentences. For simplicity, only the combinations of nouns and verbs are considered, and the length is restricted to at most 4 words, i.e., 2-word, 3-word and 4-word combinations. The discovered word combinations are then combined with single words as features to train classifiers such as a support vector machine (SVM) and naïve bayes (NB) classifier for negative life event identification. Experimental results show that the meaningful combinations of words are high-precision features, thus yielding better performance than the baseline system using single words alone.

POS Multi-tagging based on combined models
Yan Zhao (University of Groningen)

In a Hidden Markov Model (HMM), we compared two POS multi-tagging decoding methods: N_best paths and forward-backward methods on three different languages. We found that the forward-backward method is slightly better than the N_best paths method. Based on the forward-backward frame, we combined the HMM and Maximum Entropy Model and obtained a better result than using the individual models. This result suggests that our combined multi-tagging method should improve the Alpino parser both in efficiency and accuracy.

The Effect of Pragmatic Infelicity on Children’s Interpretation of Weak Crossover Sentences: Evidence from Dutch
Marie-Elise van der Ziel (Utrecht University)

The pronoun in sentence (1) can only be interpreted deictically. That is, the sentence can only mean that a certain mother tickled all the boys. The reading whereby each boy is tickled by his own mother (e.g. the bound-variable reading) is ungrammatical. The constraint which prohibits the latter interpretation is known as the Weak Crossover (WCO) constraint.

(1) ??His_i mother tickled each boy_i.

If the WCO-constraint derives from the innate set of linguistic principles of Universal Grammar (UG), and if children have access to the principles of UG from the onset of language acquisition (e.g. the Strong Continuity Approach), then children should have knowledge of the constraint from an early age. However, though some L1-acquisition studies report that English preschool children show adult-like sensitivity to WCO-effects (Thornton 1990, Su, 2001), the results of other studies suggest that 5-year olds have not acquired the constraint yet (Lewis, 2000). In particular, Thornton (1990) and Su (2001) report that more than 80 percent of their child participants rejected the test sentence as a correct description of a situation in which the bound-variable interpretation of the sentence would be true. In contrast, only 15 percent of the participants of Lewis (2000) correctly judged the test statement to be false.

This talk will argue that the results of both Su and Thornton may have been confounded by some methodological problems. Second, this talk will argue that though Lewis’ study does not suffer from the same problems as Su’s and Thornton’s studies, the poor performance of Lewis’ participants does not necessarily indicate that 4-6 year old children have not acquired the WCO-constraint yet. Rather, it is argued that pragmatically infelicitous use of the possessive pronoun may have negatively influenced children’s judgements.

In this talk I will present the results of two Truth Value Judgement experiments on the acquisition of Weak Crossover by Dutch preschool children, in which an attempt is made to improve the pragmatic felicity of the pronoun in the test statement. The fact that our child participants performed significantly better on WCO sentences than the children of Lewis’s study suggests that pragmatic (in)felicity is indeed a factor influencing children’s judgements, albeit not the only one.

Embedded verb-second revisited: a layered derivations account
Jan-Wouter Zwart (University of Groningen)

Synopsis This paper proposes a new analysis of embedded verb-second (EV2) phenomena (in languages like Dutch, Frisian, and German), within a version of the minimalist program that assumes layered derivations (Zwart 2008). The proposal is that EV2 indicates that the matrix clause and the embedded clause are construed in separate derivation layers, and more particularly that it is the matrix clause that is the outcome of a previous derivation. The analysis is consistent with the analysis of embedded root phenomena of Hooper and Thompson (1973), and proposes a rephrasing of that analysis in current minimalist terms.

Embedded verb-second Dutch, German and Frisian show a pattern of verb placement where the finite verb is in second position in independent clauses (V2), and in final position in embedded clauses. In circumscribed constructions, V2 also occurs in embedded clauses (1).

(1)	Jan	zei	dat	hij	kon	wel	janken
	John	said	that	he	could	AFF	cry
	‘John said that he could cry.’

(1) occurs when the matrix part has a subordinate status in terms of information structure (Hooper and Thompson 1973:473, Zwart 1997:235). EV2 has distinct properties (opacity for extraction, no cliticization/agreement on the complementizer) and restrictions (not with negative, irrealis, or interrogative matrix clauses; not in subject or adjunct clauses). EV2 is typically analysed as involving coordination (De Haan 1990), CP-recursion (Vikner 1991) or a restart (Zwart 1997).

Layered derivations A layered derivation is a network of derivations, such that the output of one derivation layer may function as part of the input of another. In the transition between derivation layers, material passes through the interfaces, creating a fixed sound-meaning correspondence and turning complex material into an atomic element (a standard example is idiom formation). Applied to EV2 it is tempting to think that the EV2 clause is the output of a separate derivation layer. This would explain its opacity for extraction (since the output is turned into an atom, rendering its component parts invisible for syntactic operations). However, subject clauses and adjunct clauses are by definition the output of a separate derivation, and they do not show EV2 (with few exceptions).

Proposal This talk therefore considers another implementation of the layered derivation architecture, where the matrix part is the output of a separate derivation layer. This predicts that the matrix clause does not allow any kind of modification (involving negation, mood, or interrogative). It also follows that the EV2 clause cannot be a subject/adjunct, as the matrix clause would then not be a constituent. It is also consistent with the idea that the matrix clause has subordinate status. I specifically propose that dat in (1) is not a regular complementizer but a kind of linker introduced at the interface, which may be absent in some varieties (e.g. German)

References

De Haan 1990, Hoofd- en bijzinnen: traditie en progressie, Handelingen Ned. Filologencongres 40

Hooper & Thompson 1973, On the application of root transformations, Linguistic Inquiry 4

Vikner 1995, Verb movement and expletive subjects in the Germanic languages Oxford

Zwart 1997, Morphosyntax of verb movement Kluwer

Zwart 2008, Prospects of top-down derivation, t.a. Catalan Journal of Linguistics.

Last modified: 2011-12-15 | 30th TABU Dag 2009