Up to general Table of Contents

Chapter 2 — The onset

• 2.1 — Introduction
• 2.2 — The corpus of spontaneous speech
• 2.3 — Method of data collection
• 2.4 — The earliest occurring NPIs
• 2.5 — To license or not to license, that is the question
• 2.6 — A note on the prerequisites for correct NPI use
• 2.7 — Hoeven
• 2.7.1 — The earliest occurrences
• 2.7.2 — Hoeven in an unanalyzed unit with negation
• 2.7.3 — Evidence of a restricted distribution
• 2.8 — Meer
• 2.8.1 — Solving ambiguities in early utterances with meer.
• 2.8.2 — The earliest occurrences
• 2.8.3 — Separate distributive patterns
• 2.8.4 — Meer in an unanalyzed unit with negation?
• 2.9 — Summary
• 2.10 — Outlook

Chapter 2 — The onset

2.1 — Introduction

Having established in the first chapter the complex distribution of NPIs and the issues which are relevant in studies on their acquisition, it is now time to start investigating how NPIs are used in child language. To that end a corpus of Dutch child speech will be explored. The data from this spontaneous speech study will be presented and discussed in several chapters, covering subsequent stages in the acquisition process.

In this chapter, the focus is on the onset, the way in which NPIs make their appearance in child language. The fundamental question to be explored is whether NPIs, as soon as they start to occur, have a restricted distribution or not. It is important to deal with this basic issue first, since it sets the scene against which background later developments take place. If children start using NPIs too broadly, in non-licensing environments, they will somehow have to retreat from these overgeneralizations without receiving information about where NPIs cannot appear. Such problems with the absence of negative evidence will not occur, however, if children follow a conservative strategy, starting with the narrowest possible restriction on NPIs. In that case, the licensing rules could be widened through exposure to positive evidence only.

It will become clear that although the first NPIs to appear in the children's speech are completely different expressions which, as a consequence, have different stories to tell, their behavior with respect to licensing is remarkably similar.

Up to Chapter 2 Table of Contents

2.2 — The corpus of spontaneous speech

The spontaneous speech study was carried out using a Dutch corpus that consisted of transcripts of tape recordings and diary notes. Some of the diary notes have been published (Goorhuis-Brouwer 1978; Kaper 1975; 1985; Schaerlaekens 1980; Schaerlaekens and Gillis 1987; Van Ginneken 1922), other notes were made available by parents (J. Hoeksema; Ch. Koster; J. van Kampen). The transcripts of recorded speech contain data of in total 45 children. Basic information about these corpora is summarized in Table 2.1.

All recordings have been made in the children's homes in the presence of one or more adult interlocutors (parents, family members, or investigators), during daily activities such as playing, getting dressed, eating, et cetera. The data were transcribed using the standardized system CHAT (Codes for the Human Analysis of Transcripts; MacWhinney 1995), and stored in computer files. Some of the files are available in CHILDES (the Child Language Data Exchange System; MacWhinney 1995). The children's ages in the electronic corpus range from 1;05 to 3;10. Within this total age range there is a wide variation in the number, duration, and frequency of speech samples across the children. For instance, the corpora compiled by Elbers and Wijnen span several months of development of three children, in which each child was recorded in weekly sessions of an hour on the average. The corpus compiled by Bol and Kuiken, on the other hand, contains files of 31 children in the age range from 1;07 to 3;07, each of which was recorded for approximately one hour during one session. The columns in Table 2.1 containing the frequency counts of words uttered by the children and the adults, give a rough measure of the size of the various corpora. More details regarding the compiling and transcription of these files can be found in the references listed in the rightmost column of the table.

Up to Chapter 2 Table of Contents

2.3 — Method of data collection

In the electronic corpus, the COMBO program, provided as part of the CHILDES tools (MacWhinney 1995), was used to search for occurrences of NPIs in both the adults' and the children's speech. This search included all expressions in Table 1.2, as well as many others, which are not listed in this table. Each NPI occurrence was extracted from the transcript within a small context of two preceding and two following utterances. This search yielded almost 2300 occurrences of NPIs, 1766 in adult utterances and 499 in child utterances. These were then incorporated into a database, in which each record was coded for information such as meaning domain (cf. Table 1.2), licenser, age, and provided with relevant notes about the context of the utterance, its grammaticality or ungrammaticality. Also the child utterances with NPIs from the diary notes, 243 in total, were added to the database and coded in the same way. This made a database of in all 742 child utterances with NPIs. The majority of these records fall between the ages of 2;00 and 3;04, since this is the age range which is covered most often in the recorded data. As the latest age of which recordings are available is 3;10, the data from that age onward consist of diary notes only, which run until the age of 5;08.

Since diary notes of child language consist of utterances which are, to a large extent, for some reason deviant or exceptional (Kaper 1985), they do not give a representative and proportional view of daily language use. Therefore, in the following presentation and analysis of the data, the diary notes will be excluded from frequency or percentage calculations. These data will nevertheless be very useful as additional evidence for patterns found in the recorded speech.

Up to Chapter 2 Table of Contents

2.4 — The earliest occurring NPIs

Table 2.2 lists the rank-ordered frequencies of the different NPIs found in the recorded speech of both the children and the adults.

Table 2.2 — Frequencies of NPIs in recorded speech of children and adults

expression	children	adults
meer (anymore/more)	316	1128
hoeven (need/have to)	171	454
expressions of indifference	10	116
expressions of intolerance/dislike	2	24
ooit (ever)	10	116
expressions of minimum quantity		10
expressions of incapacity		9
rest		12
total	499	1766

The adults' column in this table shows a large variety of NPIs. Of this variety, only a few types are represented in the children's speech. The frequency pattern of the few NPIs that occur in the children's speech, however, mirrors that of the NPIs in the adult input. In both adult and child speech, the high frequency of meer[1] and hoeven is striking. These two NPIs together make up 90% of all occurring NPIs in the adults' speech and are almost the only NPIs used by the children (98% of all occurrences found). Although other NPIs in this early stage in the children's speech are nearly negligible, the rare occurrences that do occur (expressions of indifference and intolerance/dislike) still appear to follow the frequency rank order of those in adult speech.

From Table 2.3, which shows the spreading of NPIs in the children's speech over different age ranges, it becomes clear that hoeven and meer not only are the most frequently occurring NPIs, they also show up earliest.

The first occurrence of hoeven is found at the age of 1;08.06, and the first occurrence of meer at the age of 1;10.04. After a preliminary period of rare occurrences until the age of about two, they appear regularly throughout the corpus.

Other NPIs than meer and hoeven occur not only very rarely, but also relatively late in the children's speech. This seems to be only logical, if we take into account what the wider variety of NPIs occurring in the adults' speech consists of. These are for the most part rather formal expressions, such as überhaupt (at all) and kunnen velen (can stand), as well as collocations, such as ergens iets aan kunnen doen (to be able to do something about something), kunnen verdragen (can bear) and NPs of minimum quantity, such as een moer (a nut), and een barst (a crack), as in:

(1)	Daar begreep ik dus geen ene moer van. There understood I thus no single-one nut of.
(2)	Ik versta d'r geen barst van. I understand there no crack of.

As it is a well known fact that the acquisition of such expressions starts late (Cacciari and Levorato 1989; Clark 1995), it hardly comes as a surprise that they do not yet show up in this corpus of early child speech. The rare expressions of indifference that do occur at this early age in the children's speech are all occurrences of iets geven (to matter), a rather informal, non-collocational expression. Its occurrences are scattered across the corpus, with one occurrence as early as 2;01.14, in simple, two-word sentences:

(3)	adult:	nou, 't valt wel mee. `well, it's not that bad.'
	child:	ja. niet. geef niet. nee. weet e niet. geef niet. (2;01.14) yes. not. matter not. no. know e not. matter not.

Up to Chapter 2 Table of Contents

2.5 — To license or not to license, that is the question

Table 2.3 makes it clear that only hoeven and meer display an early enough appearance and high enough frequency in the corpus to be eligible for a study of the onset of NPIs in child language. In the following, the appearance of these two NPIs will be presented and discussed. The question to be explored in these early data is basic, but its answer will nevertheless have important consequences for the continuation of this study at later ages: is the distribution of hoeven and meer from the onset restricted to correct, licensing environments? As was pointed out in the preceding chapter, this first step in the acquisition of NPIs sets the scene for later developmental patterns. As a consequence, the answer to this question will also determine which questions are relevant in the further study of NPI acquisition.

The data regarding the appearance of hoeven and meer will be discussed in two sections. The reason for this separate treatment is that hoeven and meer are widely different expressions; apart from their polarity sensitivity, they share no common syntactic or semantic characteristics. Hoeven being a modal verb and meer a temporal adverb, they occur in widely different structures and contexts.

Section 2.7 discusses the onset of hoeven. It will become clear that hoeven cannot be seen in isolation from other modal verbs which occur in the children's early speech. Also, the question whether hoeven is licensed from the beginning cannot be answered by looking at the earliest data alone. Later occurrences of hoeven will have to be taken into account as well.

In section 2.8, the early data of meer will be reviewed. Also here, a broader range of utterances has to be taken into account to answer the question whether the NPI meer is licensed from the onset. A confounding problem that must be dealt with is the fact that meer in Dutch is a homonym with two related meanings. One meaning is polarity sensitive, the other is not. This may result in ambiguity, in particular in negative sentences, between a polarity reading and a non-polarity reading.

Up to Chapter 2 Table of Contents

2.6 — A note on the prerequisites for correct NPI use

When the early utterances with hoeven and meer are reviewed in the next two sections, it must be kept in mind that they occur during the early multi-word stage, when the children have just started to form productive syntactic structures (Goodluck 1991; Schaerlaekens 1980). The type of speech at this age has been characterized as telegraphic (Brown and Fraser 1963), since it has been noted that functional constituents are missing in the children's utterances. There is still much debate on how mature the child's grammar at this early age is, and whether certain functional categories may be present, although they are not yet realized in speech (cf. Radford 1995). The details of this discussion are of no further concern for the present investigation. Of importance here is that the children's utterances at this age only rarely exceed a length of three words. As these sentences offer little manoeuvering space, hoeven and meer cannot be expected to appear in the structurally more complex licensing environments which are found in adult speech. The focus of the present investigation is therefore on the basic requirement - that is, the question of whether there is a licenser present at all in utterances with NPIs. In this approach, just the co- occurrence of an NPI and a licenser in one utterance - be it ever so simple - is sufficient to count as correct NPI use. Whether the licensing environment in addition is structurally correct or not, is of no crucial importance at this point.

As a consequence, the rudimentary syntax of children's utterances at this age need not be an impediment to correct NPI use. The two- and three-word utterances are elaborate enough to host both an NPI and a lexical licenser. Also, another prerequisite for correct NPI use, negation, is present at this age. Some months before the first NPIs appear, the children have started to express negative meanings in sentences, at the age of about 1;03 (Pea 1980a). The question to be explored now is whether children observe the basic rule regarding NPIs, the fact that they cannot occur in isolation, without a licenser.

Up to Chapter 2 Table of Contents

2.7 — Hoeven

The first NPI to appear in the children's speech is the modal verb hoeven. Hoeven has no exact English equivalent; it is a combination of have to (cf. the a)-translations) and need (both the main verb and the auxiliary; cf. the b)-translations):[2]

(6)	Je hoeft je geen zorgen te maken.
	a)	`You don't have to worry.'
	b)	`You don't need to/need not worry.'
(7)	Je hoeft geen Sherlock Holmes te zijn om dat te weten.
	a)	`You don't have to be a Sherlock Holmes to know that.'
	b)	`You don't need to/need not be a Sherlock Holmes to know that.'

As was pointed out in Chapter 1, the grammaticalization of hoeven as an NPI cannot be accounted for on the basis of its semantic properties; its distribution does not follow self-evidently from its meaning. Moreover, hoeven has a close semantic neighbor, the verb moeten (have to or must), which expresses almost the same meaning as hoeven but which is not polarity sensitive (see also Note 4 in Chapter 1).

In the absence of a semantic anchor, what reason could there be for the child to assume that hoeven cannot appear in straightforwardly affirmative sentences? Although the input provides the child with correct examples of hoeven, this does not necessarily mean that this verb is only allowed in such environments. After all, licensing environments have no exclusive relationship with NPIs, and hoeven is just one among many other, non-polarity sensitive verbs the child may hear being used in the presence of negation, negative quantifiers, in comparative constructions, et cetera. The question then is how the child, who has to draw out generalizations from input speech - in this specific case involving the assignment of hoeven to the category of verbs - can discover that hoeven is exceptional, with a distribution which is not completely predictable from its syntactic properties as a verb (cf. NPIs as exceptions to rules of grammar). Consider for instance the following sentences:

(8)	Niemand hoeft vandaag naar school. `No one has to go to school today.'
(9)	*Iedereen hoeft vandaag naar school. `Everybody has to go to school today.'
(10)	Niemand moet vandaag naar school. `No one has to go to school today.'
(11)	Iedereen moet vandaag naar school. `Everybody has to go to school today.'

Up to Chapter 2 Table of Contents

2.7.1 — The earliest occurrences

The first occurrence of hoeven in the corpus appears at the age of 1;08.06 (example (12)). Until the age of about two, only rare occurrences are found; in the data of children of whom such early recordings or notes are available, hoeven occurs only once or twice per child. The following examples are the earliest utterances with hoeven from different children:

(12)	ik hoef niet. (1;08.06) I need not.
(13)	torentje bouwen. hoef(t) niet doen. (1;10.24) little-tower build. need not do.
(14)	nee. hoeft niet. (1;11.21) no. need not.
(15)	hoe(ft) -nie(t). (1;11.22) need not.
(16)	hoef e niet vla. (1;11.27) need e not custard.
(17)	Dirje hoef e niet handje Sinteklaas. (2;00.12) Dirje need e not little-hand Sinteklaas. `Dirje doesn't want to shake hands with St. Nicholas.'
(18)	'k hoef nie(t) s(l)ape (n). (2;00.14) I need not sleep.
(19)	hoe(ft) -nie(t) bad? (2;01.01) need not bath? `don't I have to take a bath?'
(20)	hoef e niet. (2;01.07) need e not.
(21)	hoef(t) niet. (2;01.14) need not.
(22)	hoef(t) niet. (2;01.26) need not.

Although these are very simple utterances with only a rudimentary structure, they allow for two important observations. In the first place, they are correct, as far as the presence of a licenser is concerned. Secondly, the licenser, without exception, is the classical negation niet (not).

Up to Chapter 2 Table of Contents

2.7.2 — Hoeven in an unanalyzed unit with negation

The question which now emerges is whether the appearance of hoeven in negative utterances really must be attributed to some licensing rule in the children's grammar. If we take a closer look at these utterances, and take into account what is known about early modal usage in child language, it can be seriously doubted that the children at this age already employ an NPI restriction.

In all these early utterances, the licenser immediately follows hoeven. This results in a rather stereotypical form, hoef(t) nie(t), which looks like a lexical chunk instead of a combination of verb and negation. The inflection -t for the third person singular is not always pronounced, and a schwa may be inserted in its slot, as in (16), (17), and (20). Hoeven and negation together may in some cases even be pronounced as if they were monomorphemic, as in (15) and (19).

These characteristics of the early appearance of hoeven exactly mirror those of two other modal verbs, kunnen (can) and mogen (may), which are not polarity sensitive. Hoekstra and Jordens (1994) note, on the basis of an investigation of longitudinal diary data from one Dutch child, that kunnen (can) and mogen (may) in an early stage occur only in negative sentences, in the forms kan-nie(t) (can not) and mag-nie(t) (may not). The affirmative counterparts do not yet occur by themselves. The same observation has been made for modal elements in early child English. Bellugi (1967) noted that can and do initially appear only in combination with negative contraction, resulting in the fixed forms can't and don't. The same observation has been made with can't and won't (Kuczaj and Maratsos 1975, cited in Stephany 1986; Leopold 1949) and can't, don't, and didn't (Bloom 1970). Since the positive counterparts of these verbs, as well as inversion with them, were absent, Bellugi claimed that these early combinations of verb and negation must constitute unanalyzed wholes - that is, negative expressions, with no internal structure.

The fact that the early examples with hoef(t) nie(t) appear in the same period and in the same form as kan -nie(t) and mag -nie(t) makes it very likely that the earliest occurrences of hoeven constitute another example of the phenomenon that some modal elements first appear as unanalyzed units with negation. The question why it is that some modal elements initially appear in negative sentences is an interesting one in its own right, but falls outside the scope of this thesis. The observation which is of importance here is that hoeven - whether it be because of the early presence of a licensing rule or not - definitely has a restricted distribution from the onset. In fact, its first use, as a negative unit in combination with niet (not), is even more restricted than in adult speech.

Up to Chapter 2 Table of Contents

2.7.3 — Evidence of a restricted distribution

The previous section showed that although the early occurrences of hoeven are correctly restricted to negative sentences, this might be for another reason than its polarity sensitive character. In the same way as some other - non-polarity sensitive - modal verbs, such as kunnen (can) and mogen (may), hoeven initially appears as a negative expression. At this early point, hoeven is not yet treated like a verb, with the morphology that comes with it, let alone as a verb with additional restrictions on its distribution.

In other words, these early examples have not yet given a clear answer to the question whether hoeven is subject to licensing or not. Therefore, the question has to be taken up again with regard to later utterances with hoeven, from the age at which it is known that children no longer treat modal verbs as unanalyzed units with negation. When these unanalyzed units eventually are split up into two syntactically distinct morphemes, verb and negation, is the distribution of hoeven - now with its fresh status of a verb - still restricted to correct environments?

We are now in the awkward position that only negative evidence can show restrictions on hoeven; when the children's utterances with hoeven, after the transition from unanalyzed to analyzed has taken place, show no changes, this is reason to assume that its distribution must be restricted. The argument runs as follows. When the stage in which modal verbs are used as unanalyzed units with negation comes to an end, the distribution of non-polarity modals, such as kunnen (can) and mogen (may), will be extended to affirmative sentences (Bellugi 1967; Hoekstra and Jordens 1994). This widening of use, however, should not take place in the case of hoeven, since this verb, due to its negative polarity character, is not allowed in straightforwardly affirmative utterances. Thus, if the children at this point used hoeven correctly, this would result in no apparent changes, since hoeven then would remain restricted to negative sentences. [3]

If, however, the distribution of hoeven from this point onward showed sudden changes and started to appear in non-licensing environments, this would be an indication that the children do not take into account the polarity restrictions on hoeven, but treat this verb in the same way as the non-polarity verbs kunnen (can) and mogen (may).

Hoekstra and Jordens (1994) locate the transition between unanalyzed and analyzed use of modal verbs in their corpus at around the age of 1;11. From that age, kunnen (can) and mogen (may) start to occur in affirmative sentences, indicating that they no longer are fixed expressions with negation, but now have the status of verbs. This transition in the use of kunnen (can) and mogen (may) was also investigated in the corpus of recorded speech, in the files of eight children of whom early longitudinal data were available. Table 2.4 lists the first productive affirmative utterances with these verbs for each child.

There is considerable variation among the children concerning the frequency of these verbs, and not all slots in the table are filled. The general picture, however, corroborates Hoekstra and Jorden's claim that the transition from unanalyzed to analyzed forms takes place towards the end of the second year. At this age, the majority of utterances with kunnen (can) and mogen (may) is still negative and appears to be unanalyzed (for instance kan e niet, kannie, kannjie, kannies), but from then on, the number of positive forms increases. See for instance the following examples:

(23)	deze bovenop. deze kan, Kees. auto bovenop. (1;10.24) this-one on-top. this one can, Kees. car on-top. `I can put this block on top of the car, Kees.'
(24)	even knippen daar. en rozen mag ook. (1;11.07) just cut there. and roses may also. `just cut there, and the roses may also be cut.'

The question now is whether this transition in the treatment of modal verbs, from unanalyzed to analyzed use, also affects the way in which hoeven occurs. [4] Is hoeven, in the same way as non-polarity modals, now also extended to affirmative utterances? The following examples are representative of the use of hoeven from the age of two:

(27)	ik hoef niet douchen meer. (2;02.25) I need not shower anymore.
(28)	hoef(t) niet vies. (2;02.26) need not dirty.
(29)	hoef(t) niet. (2;03.01; diary note, with the remark that this phrase often occurs) need not.
(30)	'k hoef niet te s(l)apen. (2;03.02) I need not to sleep.

Compared to the somewhat earlier utterances, (12) - (22), these later examples show no drastic changes. Although hoeven now sometimes occurs in sentences with a more elaborated syntactic structure (as for instance in (27) and (30)), it still appears in negative sentences only, in the direct company of the classical negation niet (not). Whereas the distribution of non-polarity modals at this age is widened, the distribution of hoeven remains restricted to negative sentences. This is indirect evidence that hoeven does not follow the same acquisition pattern as the other modal verbs, but is now treated as a special verb with a restricted distribution.

There is also direct evidence, leading to the same conclusion. This regards the fact that hoeven as a verb now is integrated in the syntax. See the following examples:

(31)	hoeft 'ie niet # pap # opeten niet. (2;03.14) need he not # porridge # eat not.
(32)	Joke hoef(t) ook niet. (2;04.16) Joke need also not.
(33)	papa hoef(t) s(ch)eren niet. (2;04.23) daddy need shave not.
(34)	ik hoef z(w)emme(n) niet. (2;04.25) I need swim not.
(35)	hoeven wij # niet in bad? (2;06.19) need we # not in bath?

In these utterances, hoeven is syntactically integrated; it shows inversion and agreement (cf. (31) and (35); note in particular the difference between (35) and example (19), given earlier), and the fixed combination with negation is split up by intervening words (cf. (32), (33), and (34)). This clearly shows that hoeven now no longer constitutes an unanalyzed unit with negation, but instead has the status of a verb. [5]

So, there is both indirect and direct evidence in support of the conclusion that hoeven from the age of about two is treated as a verb. If children merely had treated hoeven according to the verb rules which now apparently are specified in their grammar, overgeneralization errors could have emerged. Such errors, however, do not occur. The distribution of hoeven - now with the status of a verb - remains restricted to negative sentences. This leads to the conclusion that already at this young age, hoeven must have the status of a verb with additional restrictions on its distribution, setting it apart from non-polarity verbs which can freely occur in both affirmative and negative sentences.

Up to Chapter 2 Table of Contents

2.8 — Meer

The other early-appearing NPI in child language is the adverb meer. Meer can have two meanings in Dutch. One is polarity sensitive and corresponds to anymore (cf. (36)); the other is not polarity sensitive and corresponds to more in English (cf. (37)):

(36)	Sinds dat feest heb ik Jan nooit meer gezien. `Since that party, I never saw John anymore.'
(37)	Deze brief weegt niet meer dan 20 gram. `This letter weighs no more than 20 grams.'

The distinction between the two meanings of meer is not always as clear-cut as in the above examples. In negative sentences, meer can be ambiguous between a polarity reading (from now on, `NPI-meer', for short) and a non-polarity reading (`quantitative-meer'). See the following example:

(38)		Er is niet meer eten. There is not more food.
	a)	polarity reading temporal aspect	`There is no food anymore.' `There is no food left.'
	b)	non-polarity reading quantitative aspect	`There is not any more food.' `There is not more food than this.'

The two possible meanings of meer in negative sentences are closely related. In many cases, a temporal and a quantitative meaning are not in conflict with each other, and may in fact converge. In some instances, when the two meanings make a relevant difference and may cause confusion, the ambiguity may be solved by overtly marking the quantitative meaning by means of stress (cf. (39)):[6]

(39)	Jan kon niet meer[!] eten. `John could not eat any more[!].'

From the perspective of acquisition, the distribution of meer may appear to be even more complex than the distribution of hoeven. For one thing, in the same way as the subtle meaning distinction between hoeven and moeten corresponds to substantially different distributive rules (cf. section 2.7), meer also presents the child with two close semantic neighbors, each with its own specific distribution: while quantitative-meer is allowed in both affirmative and negative sentences, the distribution of NPI-meer is restricted to specific licensing environments only. In contrast to the hoeven/moeten pair, the situation appears to be further complicated by the fact that the two meanings are indistinguishable qua form; they are both expressed by the homonym meer. When the child does not keep the two meers apart, this homonymy of a non-NPI and an NPI will provide ample opportunity for overgeneralization errors, even when a conservative acquisition strategy is followed. If meer is represented as one uniform expression in the child's lexicon, input speech will provide the child with the false evidence that this expression can occur in both affirmative and negative sentences. In that case, overgeneralizations of NPI-meer to non-licensing environments seem unavoidable.

According to Clark (1993), however, there is no a priori reason to assume that homonymy causes confusion in the acquisition of the lexicon: Although homonymy presents children with one-to-many mappings, it does not seem to cause them any difficulty in the lexicon at large. They seem to master the meanings of such homonyms as night and knight, pair and pear, light (illumination) and light (weight) without difficulty. There is no evidence that children avoid homonyms during acquisition. Rather, they appear to understand and use them from an early age. In English, even two-year-olds understand and produce many homonymous nouns and verbs (Clark 1993: 113).

What has to be explored in the early spontaneous speech data, then, is whether the children indeed can keep these two meers apart as two separate expressions, each with its own distributive pattern. More in particular, it has to be investigated whether the non-polarity part of the homonym, quantitative-meer, leads to overgeneralization of the polarity-part, NPI-meer, to non-licensing environments. Or do children from the onset treat NPI-meer apart from quantitative-meer, as having its own distributive pattern, restricted to licensing environments only?

Up to Chapter 2 Table of Contents

2.8.1 — Solving ambiguities in early utterances with meer

In order to answer this question, the children's early utterances with meer have to be analyzed for possible ambiguities, not only the negative utterances, but also the affirmative ones.

Although, as was just pointed out, in adult speech there is no ambiguity in affirmative utterances with meer, it is still an open question whether NPI-meer may occur in affirmative utterances in child speech. Since the issue at stake is whether NPIs are licensed from the onset or not, children's affirmative sentences with meer cannot be excluded from the investigation. Thus, when an affirmative utterance like (41) is found in the corpus of child speech, it has to be decided whether this is a correct occurrence of quantitative-meer, a), or an incorrect (e.g. unlicensed) occurrence of NPI-meer, b). If a) is the case, then the utterance is of no further relevance for the present investigation. However, (41) with meaning b) would be a violation of the licensing restriction.

(41)		Hij heeft meer. He has more.
	a)	`He has some more.' (quantitative)
	b)	*`He has anymore.' (temporal)

Up to Chapter 2 Table of Contents

2.8.2— The earliest occurrences

The very first occurrences of meer, which appear towards the end of the first year, are in affirmative utterances, which in general serve the purpose of obtaining something the child wants to have. See the following examples:

(43)	meer peer. (1;09.20) more pear. `I want to have another pear.'
(44)	op. die op. nog meer koffie. (1;10.09; about to fill an empty coffee cup) all-gone. that-one all-gone. still more coffee. `that cup is empty. I will pour in some more.'

(45)	child:	meer. (1;11.03) more.
	adult:	graag meer mama. please more mama.
	child:	meer. meer appel. more. more apple. `I want to have another apple.'

(46)	koek meer. meer. koek meer. meer, meer. meer koeke(n). (2;00.07) cookie more. more. cookie more. more, more. more cookies. `I want to have more cookies.'

As was pointed out in section 2.8, in adult speech there is no ambiguity in affirmative utterances with meer; only the non-polarity reading is possible. The above examples show that also the first affirmative utterances with meer in the children's speech leave no doubt as to their meaning; it must be quantitative-meer which is being used here.[7] As it is clear that no temporal meaning is expressed in these utterances, there is no sign of NPI-meer, let alone of unlicensed use of NPI-meer. [8]

Some time later, negative utterances with meer start to occur as well. Examples are given in (47) - (54). As was pointed out in section 2.8, such utterances may exhibit ambiguity between a polarity reading and a non-polarity reading. Depending on which meer is being used, the negative utterances with meer in the children's speech may represent different developments. If meer has a temporal meaning, then they mark the appearance of NPI-meer in correct, negative utterances. If, however, meer has a quantitative meaning, then they show the continuation of quantitative-meer, the distribution of which is now being extended to negative utterances. Of course, it is also possible that there is a combination of these two, thus marking the continuation of quantitative-meer, with NPI-meer coming in at the same time.

It is important to note, however, that in either case no errors are made, neither with NPI-meer nor with quantitative-meer: NPI-meer is allowed to occur in negative utterances, while quantitative-meer with its free distribution can occur in both affirmative and negative utterances. The question now is which meer is being used in the following utterances.

(47)	Context: The container of a toy truck breaks and is fixed. When the child starts playing with the truck, the container breaks again. Then, the child starts playing with a toy duck. kan niet meer. eendje taartje bakken. (1;10.04) can not more. little-duck little-cake bake.
(48)	nou nie(t) meer. (2;00.19; pretending to eat cotton pads) now not more.
(49)	nou nie(t) meer. meer me(l)k. (2;02.02) now not more. more milk.
(50)	niet meer kleien. dat nog e zoeken. (2;03.02) not more claying. that still e look-for.
(51)	nou is nie(t) meer. (2;03.10) now is not more.

(52)	adult:	wil je nog een heel klein beetje? `do you want just a little bit more?'
	child:	dan niet meer. (2;03.22) then not more.

(53)	nou kan niet[!] meer zien. (2;03.21; after closing the door of a toy ambulance) now can not[!] more see.
(54)	Context: the child pretends his finger is a puppet, driving around in a toy car. nou mag vinger niet meer in. nou mag die vinger er weer in. (2;04.12) now may finger not more in. now may that finger there again in.

When a close look is taken at the above examples, it becomes clear that both NPI-meer and quantitative-meer are represented. The context of (47) makes it very likely that a temporal meaning is expressed; this example thus has to be regarded as the first occurrence of NPI-meer in the corpus. [9] Also (50), (53), and (54) express temporal meanings and are thus regarded as occurrences of NPI-meer. In all these examples, niet meer can be substituted by the temporal expression no longer, leaving the meaning of the utterances intact.

There are also utterances which most probably express a quantitative meaning, such as (49) and (52). These examples are therefore regarded as negated instances of quantitative-meer. In the two remaining examples, (48) and (51), it cannot be decided what the exact meaning of meer is. In principle, these utterances could express a temporal meaning, but it is not impossible either that their meaning is quantitative. Meer in these utterances might as well be an amalgam of both a temporal and a quantitative meaning (cf. section 2.8).

Up to Chapter 2 Table of Contents

2.8.3 — Separate distributive patterns

The early data show that the two meanings of meer in the children's speech emerge separate from each other, with the quantitative meaning coming in first and the temporal meaning following somewhat later. It also appears that the children treat the two meers as separate expressions, each with its own distributive pattern. When quantitative-meer emerges, its first use is in affirmative utterances only. Some time later, its distribution is extended - as is correct - to negative sentences as well. When NPI-meer comes in, it appears in negative sentences only, thus in a correct environment from the onset.

Apparently, then, the two meers and their respective distributions are not mixed up in the children's vocabulary. Although meer presents the children with a mapping of one form onto two meanings, there is no sign of confusion. This is in accordance with Clark's (1993) claim that children quite readily accept homonymy.

What is particularly important about these data for the study of NPI acquisition is the fact that NPI-meer appears to have a restricted distribution from the onset. Although quantitative-meer, with its relatively free distribution, is the first to appear, it does not drag NPI-meer along into non-licensing environments. Obvious errors of meer with a temporal meaning, occurring in affirmative utterances, are not found. The children's affirmative utterances with meer can unambiguously be interpreted as expressing a quantitative meaning. NPI-meer only occurs in sentences in which it according to its negative polarity character is allowed to appear.

Up to Chapter 2 Table of Contents

2.8.4 — Meer in an unanalyzed unit with negation?

The children's early negative utterances with meer have one notable characteristic in common with the earliest occurrences of hoeven: the exclusive presence of the classical negation niet (not). Whereas hoeven is immediately followed by this negation, resulting in the stereotypic form hoef(t) nie(t), meer is immediately preceded by it, resulting in nie(t) meer. In section 2.7.2, it was argued that hoef(t) nie(t) until the age of two represents an unanalyzed unit, and need not be the result of the early presence of a licensing rule. The question now arises whether the same may apply to nie(t) meer. Could it be that children at this early age in their lexicon specify nie(t) meer as a negative expression, apart from the affirmative item meer?

The clues which could be used to decide whether hoef(t) nie(t) was an unanalyzed unit - its distribution compared to that of other modal verbs, and inflections or inversion with the subject on proper occasions - cannot be used here, since meer is an adverb, and as such appears in only one, static form. It thus seems that spontaneous speech data alone cannot provide evidence to demonstrate the existence of nie(t) meer as an unanalyzed unit. There is, however, indirect evidence for the opposite: that nie(t) meer cannot be an unanalyzed unit. See for instance the following example:

(55)	nou nie(t) meer. meer me(l)k. (2;02.02) now not more. more milk.

In this example, the second utterance, meer me(l)k (more milk), appears to be an elucidating elaboration of the former, indicating that the child is able to split nie(t) meer up into two separate units. This shows that in the preceding utterance, nou nie(t) meer (now not more), nie(t) functions as the negation of meer, instead of constituting an unanalyzed unit with it.

There are more examples in the data, allowing the same conclusion, that meer must be negated by niet, and does not amalgamate with it. In (56), niet is uncoupled from meer by means of stress. And in (57), there is a clear pause between niet and meer. Moreover, meer is repeated as an expression on its own, apart from niet.

(56)	nou kan niet[!] meer zien. (2;03.21) now can not[!] more see.
(57)	dat gaat Frank niet # meer meer +... (2;02.03) that goes Frank not # more more +...

So, although niet and meer in the children's utterances invariably occur in the same cluster, there are indications that they do not constitute an unanalyzed negative unit. It is much more likely then to consider the direct co-occurrence of niet and meer as a natural consequence of the syntactic structure of Dutch, which offers little opportunity for a larger distance between an adverb and niet. This becomes evident from adult utterances in which meer is licensed by niet. In the corpus of written Dutch (cf. section 1.5.1), it was found that 90% of these utterances shows the same direct clustering of niet and meer. In the remaining 10% the distance between niet and meer is enlarged by adjectival and adverbial phrases, as for example in (58) and (59): [10]

(58)	Eigenlijk ben ik niet boos meer maar wel heel verdrietig. Actually am I not angry anymore but indeed very sad.
(59)	Dergelijke argumenten hoor je tegenwoordig niet zo vaak meer. Such arguments hear you nowadays not so often anymore.

The co-occurrence of niet meer in child speech thus appears to be a reflection of syntactic conditions. In this respect the children's utterances are not deviant from those of adult speakers of Dutch. [11] However, a clear difference with adult speech at this point is the fact that meer only appears in combination with niet (not), whereas in adult speech there is more variation in licensing environments (cf. Table 1.1).

Up to Chapter 2 Table of Contents

2.9 — Summary

The two earliest appearing NPIs in Dutch child language fall in different syntactic categories: one is a verb, hoeven, the other an adverb, meer. As a natural consequence, these expressions occur in widely different structures and contexts. Different approaches were therefore required to investigate whether these expressions from the onset appear in accordance with the basic restriction they have in common: the fact that they are only allowed in a limited set of licensing environments.

The early appearance of hoeven is interwoven with that of modal verbs in general, and in the case of meer, a polarity reading has to be distinguished from a non-polarity reading. In both cases, however, the answer to the research question formulated in section 2.5 turned out to be affirmative: these two NPIs are restricted from the onset in the children's speech. From the very start, these expressions are subject to a restriction which sets them apart from their non-polarity neighbors. Whereas non-polarity modal verbs and quantitative-meer freely occur in both negative and affirmative utterances, the NPIs hoeven and meer appear in negative sentences only.

This chapter started with the open question whether the appearance of NPIs in child speech is restricted or unrestricted. In the light of the poverty of stimulus - more in particular: the absence of negative evidence - there was an a priori assumption that children would be better off if they started with a highly restricted distribution of NPIs. But in principle, children could approach the acquisition of NPIs from the other end - that is, starting out with an unrestricted distribution. In that case, the children would have to backtrack, a task for which positive evidence alone seems to be insufficient.

The study of the earliest utterances with NPIs in spontaneous speech pursued in this chapter has shown that children take the most beneficial first step: they start out with a highly restricted distribution of NPIs. With this starting point, they are now in the position that they can take full advantage of the host of available positive evidence in the input about the distribution of NPIs.

Up to Chapter 2 Table of Contents

2.10 — Outlook

The finding that NPIs in the children's speech from the onset have a restricted distribution is an important first accomplishment. At the same time, however, it is clear that the children still have a long way to go towards adult-like use of NPIs. Accordingly, new questions arise as to how the acquisition of NPIs and their licensing requirements proceeds from here. In particular, the issue now has to be dealt with as to what kind of knowledge may be underlying the restricted distribution of NPIs in the children's speech.

One conspicuous observation in the data discussed so far is that the distribution of NPIs at this early age is much more restricted than in adult language. In the children's speech, hoeven and meer are restricted to the immediate company of the classical negation niet (not) only. In adult language on the other hand, these NPIs occur in a variety of licensing environments. As such, the adult utterances all conform to the requirement of licensing, which defines the boundaries of a correct distribution of NPIs. Within these boundaries, there is room for a variety of environments in which NPIs may appear. As long as none of this allowed variety is displayed in the children's speech, it remains unclear which principle may be underlying this correct but highly restricted use of NPIs.

One possibility is that the narrow distribution of NPIs at this early age is due to a strategy of conservative widening, as discussed in section 1.7.4. It was pointed out there that if children followed such a cautious principle in the acquisition of NPIs, they would start out using NPIs in a highly restricted manner, in the presence of the classical negation niet (not) only. At a later stage, when positive evidence would induce the child to revise his grammar, the variety in licensing environments would increase, first by means of regular negations, later by minimal negations as well. The acquisition data we have seen so far appear to be in accordance with the prediction that children would start out using NPIs in the presence of niet (not) only. It is, however, much too early to conclude that children indeed employ a strategy of conservative widening. Data from later ages now have to be taken into account, to see whether the other predictions - an error-free, gradual expansion of the licensing rules - are born out as well.

What complicates the interpretation of the early exclusivity of niet (not) as a licenser is the fact that it is so frequently used in input speech. The narrowest possible generalization to be conceived of in NPI licensing clearly coincides with the most frequent distributive pattern. As became clear from Table 1.1, more than 60% of adults' utterances with hoeven and meer is licensed by niet (not).[12] Presented with this abundance of examples, children could filter out niet (not) as the prototypical environment for hoeven and meer.

In order to gain more insight into the principle which may underlie the highly restricted distribution of NPIs, the continuation after this early start has to be investigated. Questions of particular relevance are whether any errors occur during this process, and whether any particular pattern can be detected in the order in which other licensers appear.

On to next chapter: `Apparent Incompatibilities'

Up to general Table of Contents

Back to previous chapter: `The Structure and Use of Negative Polarity Items'