Up to general Table of Contents

Chapter 1 — The structure and use of negative polarity items

• 1.1 — Outline
• 1.2 — The restricted distribution of a class of verbal idioms
• 1.3 — Pragmatic entailment
• 1.4 — Scale-reversing environments
• 1.5 — Negative polarity items: a multi-facet phenomenon
• 1.5.1 — NPIs of unknown origin
• 1.5.2 — Domains of meaning
• Minimum quantity
• Maximum quantity
• Indifference
• Intolerance/dislike
• Incapacity
• Matching
• Unawareness
• Degree
• Special cases
• 1.5.3 — Idiosyncratic behavior
• Idiomatized NPIs
• Semi-NPIs
• Core NPIs
• 1.5.4 — Crosslinguistic differences and historical developments
• 1.6 — Theories of polarity sensitivity
• 1.6.1 — Ladusaw: semantic entailment
• 1.6.2 — Dowty: facilitation of inferences
• 1.6.3 — Zwarts: the relative strength of licensers
• 1.6.4 — Linebarger: negative implicatures
• 1.6.5 — Negative polarity items: one phenomenon?
• 1.7 — Issues in the psycholinguistic study of NPIs
• 1.7.1 — Users' intuition about polarity sensitivity
• 1.7.2 — The acquisition of NPIs
• 1.7.3 — The `no negative evidence problem'
• NPIs as exceptions to rules of grammar
• Correcting overgeneralizations
• 1.7.4 — Outlook: acquiring the correct licensers for NPIs
• Priority ordering of rules: from narrow to wider generalizations
• Negation in terms of entailment patterns

Chapter 1 — The structure and use of negative polarity items

1.1 — Outline

This chapter is divided in three main parts and sketches the background of the developmental study of negative polarity items. First, a characterization of the behavior of negative polarity items is given, based on data from different languages, in sections 1.2, 1.3, 1.4, and 1.5. These observations are followed in section 1.6 by a short account of various theoretical approaches to polarity sensitivity. The last part of the chapter, starting with section 1.7, is concerned with the issues which are relevant in the psycholinguistic study of negative polarity items, introducing the research questions which will be explored in later chapters.

Up to Chapter 1 Table of Contents

1.2 — The restricted distribution of a class of verbal idioms

In several languages, there is a collection of similar verbal idioms. Although the lexical content of these expressions differs, it is nevertheless clear that they are variations on the same theme. This becomes evident from their uniform structure - NP + verbal predicate - and similar distributive pattern. Examples from a few languages are given below:

English:

to lift a finger, to drink a drop, to budge an inch, to hurt a fly

Dutch:

een spaan heel laten (to leave a chip of wood intact), een strobreed in de weg leggen (to lay a straw-width in the way), een knip voor de neus waard zijn (to be worth a snap in front of the nose), een vinger uitsteken (to reach out a finger)

German:

einen Muckser von sich geben (to give a sound from oneself) (Zwarts 1995), ein Sterbenswörtchen sagen (to say a dying-word) (Kürschner 1983), einen Finger krummachen (to bend a finger), einen Pfifferling wert sein (to be worth a toadstool)

Norwegian:

å eie nåla i veggen (to own the needle in the wall), å røre en finger (to move a finger), å være verdt en øre (to be worth an øre), å se hånda foran seg (to see the hand in front of oneself)

French:

manger une bouchée (to eat a mouthful), boire une goutte (to drink a drop), verser un kopeck (to pay a copeck) (Hoppenbouwers 1983)

Spanish:

ver un alma en la calle (to see a soul on the street), pegar un ojo (to close an eye), decir ni pio (to say a peep) (Víctor Sánchez-Valencia, p.c.)

Greek:

leo leksi (say word), pino ghulja (drink drop), dino dhekara (give 10-cent-coin), klino mati (close eye) (Anastasia Giannakidou, p.c.)

Hungarian:

a kisujját mozdítani (to move his/her little finger), egy cseppet érdekelni (to care a bit), egy falatot enni (to eat a mouthful), a füle botját mozdítani (to move the `stick of the ear') (Ildikó Vaskó, p.c.)

Provisionally, the distribution of these expressions can be characterized as restricted, since they only appear in a subset of the configurations in which they, by virtue of their syntactic properties as noun/verb combinations, would be grammatical. This is illustrated in the following examples, with some of these idioms from English and Dutch:

The a)-examples are perfectly normal utterances, but the b)-examples certainly are not. Yet, both the a)- and b)-sentences are syntactically correct strings of words. It thus appears that the distribution of noun/verb combinations such as to lift a finger (een vinger uitsteken), to drink a drop (een druppel drinken), to utter a sound (een kik geven), and to hurt a fly (een vlieg kwaad doen) cannot be completely predicted from their syntactic properties alone, but must be subject to additional restrictions.

The anomaly of the b)-examples obviously is related to their meaning; it is difficult to understand these sentences under their idiomatic reading. Benson et al. (1986) characterize idioms as relatively frozen expressions whose meanings do not reflect the meanings of their component parts. In the a)-examples above, this idiomatic meaning is effective and produces the intended result. Thus, the a)-sentence in (1), instead of telling something about the bystanders' fingers, conveys the meaning that Melissa was helpless. In the same way, the meanings of the other a)-sentences are, respectively, that Frank remained sober at the party, that it was completely silent in the auditorium, and that John is a harmless person. In the b)-sentences, however, the literal meaning of these expressions is forced into focus, resulting in an anomaly of semantic and pragmatic nature. Thus, we may wonder how a minimal offer of goodwill such as lifting a finger could help Melissa, how Frank managed to drink exactly one drop of wine at the party, et cetera. The literal meaning of these sentences does not make much sense, and it is indeed very unlikely that such utterances will ever occur in natural conversation.

What is it that induces this difference in meaning and acceptability between the a)- and the b)-examples? The crucial factor appears to be the presence or absence of negation. Idioms such as to lift a finger (een vinger uitsteken), to drink a drop (een druppel drinken), to utter a sound (een kik geven), and to hurt a fly (een vlieg kwaad doen) obviously thrive in the company of a negative expression, such as none (geen), not (niet), nobody (niemand), or never (nooit). It is not, however, the case that a specific negative form is an inseparable part of the idiom itself. Rather, these idioms seem to require the presence of some negative expression. Thus, the expression to drink a drop (een druppel drinken) does not form a fixed idiom with the negation not (niet), but can appear in a variety of negative sentences. See for instance the following examples:

To summarize the observations so far, languages contain a class of verbal idioms which not only resemble each other in form, but which also display a similar distributive pattern in sentences. The restricted distribution of these expressions obviously is not a matter of fixed collocation, but leaves room for some variation. The questions which arise on the basis of these observations are: what are the boundaries of this restricted distribution, and why do these restrictions apply to these particular expressions?

Up to Chapter 1 Table of Contents

1.3 — Pragmatic entailment

Fauconnier's work on implicational scales (Fauconnier 1975; 1979; 1980) explains the logic underlying the distribution of these expressions, answering both the above questions at the same time. Fauconnier shows that there is a natural reason why these expressions occur in these particular environments: in combination with each other they result in pragmatically strong statements.

Fauconnier's argumentation starts with the observation that superlatives may produce quantificational effects in certain sentences. For example:

In this sentence, the superlative the most difficult problem has the force of a universal quantifier: if Frank can solve even the most difficult problem, then we may infer that Frank can solve any problem. Fauconnier suggests that the quantificational value of the proposition in (8) is linked to a pragmatic scale of complexity:

This scale represents the decreasing complexity of problems, such that problem x3 is less complex than problem x2, problem x2 in turn is less complex than problem x1, and so on. The lowest point on the scale is denoted by the superlative: it is the most difficult problem. Now, when it is true that Frank can solve x1, we assume that Frank will also be able to solve problems x2 and x3, which are higher on the scale and therefore are less complex. In particular, if Frank can solve the most difficult problem, the lowest point on the scale, this implicates that Frank will be able to solve any problem, since all other problems are higher on the scale and therefore are less complex. Fauconnier notes that this is a pragmatic implicature rather than a strictly logical implication, since one could think of circumstances in which Frank, although he is presented with a simpler problem than the lowest point on this scale, will not be able to solve it. However, on the basis of common sense, it is appropriate to assume that the less complex a problem is, the more likely it is that a person will be able to solve it.

Note that this quantified reading is no longer effective when the sentence is negative:

On the pragmatic scale of complexity, it is not possible to make inferences from (9) to positions higher on the scale, as was possible with the positive counterpart, (8). Thus, the superlative the most difficult problem is quantificational in the affirmative sentence (8) but not in the negative sentence (9). This difference is caused by the entailment-reversing effect of negation. On the same scale of complexity, the entailment pattern of the negative sentence (9) is downward instead of upward:

When it is true that Frank cannot solve x2, we assume that Frank will not be able to solve x1 either, which is lower on the scale and therefore presents a more complex problem. Now it is the opposite end of the scale which has the meaning of a universal quantifier:

If Frank cannot solve the simplest problem, the highest point on the scale, this pragmatically implicates that Frank will not be able to solve any problem, since the other problems are all lower on the scale, and therefore are more complex.

The evidence for scale reversal with negation thus has the following form: a superlative gives rise to a quantified reading in a sentence but not in its negation, and the polar opposite of that superlative gives rise to the quantified reading only in the negative counterpart of the sentence (Fauconnier 1979: 291).

Thus, scalar end points which of themselves do not have any universal entailments may obtain quantificational force in the entailment-reversing environment of negation. It is exactly this combination of a scalar end point and an entailment-reversing environment which gives noun/verb combinations such as to lift a finger (een vinger uitsteken) and to drink a drop (een druppel drinken) the pragmatic force of a universal quantifier. The nouns in these expressions stand for the idiomatic low points of implicational scales. Thus, lifting a finger can be regarded as the low extremity on a scale of physical activities. Likewise, a drop indicates the minimum on a quantity scale of liquids, as is illustrated in the figure below:[1]

This scale is implicational with respect to the sentence Frank didn't drink a drop:

In this sentence, a drop is equivalent to a universal quantifier: saying that Frank did not drink the least amount of liquid is equivalent to saying that he drank nothing at all. The combination of a scalar end point with negation thus appears to be one which is only natural in the view of successful communication, since it results in a strong statement in terms of maximization of the conventional implicatures which can be made.

Up to Chapter 1 Table of Contents

1.4 — Scale-reversing environments

As was shown in the previous section, the idioms of minimum quantity introduced in section 1.2 obtain universal force by appearing in the entailment-reversing environment of negation. This was illustrated with the sentence negation not, which also is used in the formalization of reversed entailment in the contraposition law in propositional logic: if P => Q, ¬Q => ¬P

There are, however, more environments in natural language which have the ability to reverse entailment, such as the negative expressions none, nobody, never, nowhere, no, and nothing. As can be seen in examples (1) - (4), such environments also give rise to quantified readings of idioms of minimum quantity. Fauconnier (1979) has pointed out that it is not only sentence negation and negative quantifiers that are entailment-reversing. There is a wide variety of linguistic environments which also has this property, for example conditionals, universal statements, comparatives, before-clauses, and too...to.

Indeed, as the following examples show, verbal idioms like to drink a drop (een druppel drinken), to lift a finger (een vinger uitsteken), and to utter a sound (een kik geven) can appear in this whole variety of environments:[2]

The restricted distribution of idioms of minimum quantity thus appears to be semantically and pragmatically functional: by appearing in environments which have the ability to reverse entailment, these expressions induce strong statements.

Up to Chapter 1 Table of Contents

1.5 — Negative polarity items: a multi-facet phenomenon

The foregoing constituted a clear example of how languages are shaped by their users to meet certain communicative demands. The behavior of verbal idioms like to drink a drop and to lift a finger is not merely an arbitrary fact of grammar, unaffected by considerations of meaning or pragmatics. Rather, the restricted distribution of these expressions can be understood in functional terms, as meeting the requirements of successful communication. Interestingly, it appeared that the restrictions on these idioms have a universal basis: they are founded in a language-independent entailment pattern. Therefore, it is not surprising that this class of idioms is a recurrent phenomenon in different languages.

The scale principle thus offers a satisfactory account for the restricted distribution of a crosslinguistic class of idioms. The picture, however, is not that clear anymore when these idiomatic expressions are seen as being part of a much more extensive phenomenon. It is a well-known observation that languages contain quite varied collections of words and expressions with a behavior comparable to that of the verbal idioms discussed in the foregoing (cf. Van der Wouden 1994a for an inventory of observations from a variety of languages). The distribution of these items is restricted to the same environments, with negation being the most dominant one; hence the name Negative Polarity Items (NPIs, for short). Intuitively, it is felt that these items constitute one coherent phenomenon and that there must be a unified account for the patterns characterizing their distribution. The account in terms of scalar implicatures, however, which fits so well for the distribution of the class of negative polarity idioms, turns out to be not simply applicable to all NPIs. As will be set out in the following paragraphs, the phenomenon of polarity sensitivity is many-sided: NPIs cluster in widely different domains of meanings, are subject to historic changes, and show - to some extent - idiosyncratic behavior, both within and across languages.

Up to Chapter 1 Table of Contents

1.5.1 — NPIs of unknown origin

Investigation of large corpora of written speech - carried out within the research project Reflections of Logical Patterns in Language Structure and Language Use at the University of Groningen - has yielded an inventory of several hundreds of NPIs in Dutch. A substantial number of these expressions is listed and categorized in Table 1.2. This inventory will be discussed later on; for the time being, a closer look is taken at only four of the expressions that appear on this list: hoeven (have to/need), meer (anymore), kunnen schelen (to care), and kunnen uitstaan/velen (can stand). As the following examples show, these expressions are NPIs, since they are not allowed in straightforwardly affirmative sentences (cf. the a)-examples), but always need a certain licensing environment to appear correctly (cf. the b)-examples):

The distribution of these four NPIs in actual language use is shown in Table1.1: [3]

From this table it becomes clear that hoeven, meer, kunnen schelen, and kunnen uitstaan/velen appear with a limited set of licensers, which by and large correspond to the environments in which the idioms of minimum quantity can occur. However, where the restricted distribution of these idioms was shown to follow naturally from their status as end points on implicational scales, there appears to be no such natural connection between meaning, pragmatics and distributive properties for these four NPIs. It is not clear how these expressions could be regarded as scalar end points, and neither does there appear to be any semantic or pragmatic reason why their distribution should be limited to only these, for the most part negative, environments. To see this, compare the b)-examples in (17) - (20) with the Dutch b)-examples given in (1) - (4). The common characteristic of these sentences is the fact that they are aberrant, without being syntactically incorrect. There is, however, a difference between the two sets of examples with respect to the underlying reason for their anomaly. The anomaly of the first set of examples, the b)-sentences in (1) - (4), is of semantic and pragmatic nature: affirmative sentences with idioms of minimum quantity are not implicational with respect to scales and therefore result in weak statements. Such a motivation, however, is not available for the anomaly of the b)-sentences in (17) - (20), containing other NPIs. There is no semantic or pragmatic reason why hoeven, meer, kunnen schelen, and kunnen uitstaan/velen should be correct in the negative a)-sentences, but not allowed in affirmative sentences, as in the b)-examples. Depending on the situation, it makes perfect sense to talk about something being necessary, as well as about something being not necessary (as in (17) a) and b)). In the same way, one might want to talk about an ongoing situation as well as a situation which is no longer the case (as in (18) a) and b)), about personal involvement as well as indifference (as in (19) a) and b)), and about personal preference as well as dislike (as in (20) a) and b)). As a consequence, one would expect a language to offer the means to express the meanings of both the a)- and b)-sentences. In fact, Dutch provides its speakers with the opportunity to express both meanings, but under the specific grammaticality condition that different expressions are used in the a)- and b)-sentences. In the a)-sentences, the NPIs are correct, but in order to express the positive counterparts of these sentences, other, non-polarity sensitive expressions are needed: moeten (have to/must), nog (still), erg vinden (to bother), and kunnen opschieten (to get along).[4] The correct counterparts of the b)-sentences in (17) - (20), containing these non-polarity sensitive expressions, are given in (21) - (24):

Thus, there appears to be no obvious semantic or pragmatic necessity for the grammaticalization of hoeven, meer, kunnen schelen, and kunnen uitstaan/velen as NPIs. The unacceptability of the b)-examples in (17) - (20) remains unmotivated; such sentences are just `not done' in the grammar of Dutch. The underlying reasons for the restricted distribution of these expressions is as yet unclear.

Up to Chapter 1 Table of Contents

1.5.2 — Domains of meaning

Although the distribution of NPIs does not follow in an obvious way from semantic, syntactic, or pragmatic properties - at least not for some of them, as was shown in the previous paragraph - it is intuitively clear that the phenomenon of polarity sensitivity is not merely an arbitrary fact of grammar, a linguistic limitation apropos of nothing. It is claimed that grammaticalization processes in a language are motivated by communicative needs and speaker-hearer interactions (Hopper and Traugott 1993), and the same background must be assumed for the grammaticalization of certain expressions as NPIs.

Hoeksema (1994a) has suggested that meaning might be the crucial factor in the grammaticalization of NPIs. He notes, as does Van der Wouden (1994a), that NPIs are not arbitrarily distributed over the lexicon and that there are intuitions about the meaning domains in the lexicon in which NPIs most likely are to be found, as well as about the domains which will not be a rich source of NPIs. For example, in learning a foreign language, one would not expect the words for lamp or to travel to be restricted to only negative sentences, questions, et cetera. It should be noted, however, that this knowledge is skewed: intuitions are rather robust when it comes to excluding the possibility that a word or expression with a given meaning will be polarity sensitive, but they are weak when it comes to actually predicting that a given expression will be an NPI. Although the meaning of an expression may give a hint of possible polarity sensitivity - as Hoeksema (1994a) put it, certain expressions are predestined to become polarity sensitive due to their semantic properties - one can never be sure whether such a possible NPI candidate actually will develop polarity sensitive behavior. The factors which lead to the grammaticalization of certain expressions as NPIs are only potential, and certainly not absolute: the phenomena that give rise to language change are so complex that they will perhaps never be understood in enough detail for us to state precisely why a specific change occurred in the past or to predict when one will occur and if it does what it will be (Hopper and Traugott 1993: 63).

The clustering of NPIs in specific domains of meaning is shown in Table 1.2. This inventory of Dutch polarity sensitive expressions is based on a large corpus of mainly written Dutch, compiled by Jack Hoeksema. The table contains a representative sample of the most frequently occurring NPIs in Dutch, categorized according to the basic meaning they have in common.

Up to Chapter 1 Table of Contents

Minimum quantity

Category A contains a selection of the idioms of minimum quantity, which were presented in more detail in section 1.2. The nouns in these expressions all denote an idiomatic minimum, such as the minimum width of een speld (a needle) in een speld tussen kunnen krijgen (to be able to get a needle in between) and een strobreed (a straw-width) in een strobreed in de weg leggen (to lay a straw-width in the way). In combination with their licensing environments, these expressions result in the pragmatic effect of not even A, let alone B (where A is a minimum, and B represents a point higher up the scale), as was shown in section 1.3.

The same applies to the NPs, which constitute the second subcategory under A. These NPs do not appear in only one specific NP/verb combination but can occur with a variety of verbs. For instance, een moer (a nut) and een klap (a slap) appear in combination with - among others - doen/uitvoeren (to do), snappen (to understand) and waard zijn (to be worth).

The NPs listed under A in the table are only a fraction of the NPs with NPI meaning which are encountered in daily speech; the list can be extended ad infinitum. Moreover, these expressions appear to be a source of inventiveness and fluctuate with the fashion of the moment; new creations succeed each other at a rapid rate (see also Note 1).

Up to Chapter 1 Table of Contents

Maximum quantity

The expressions listed under B may seem to form a rather incoherent group, since they have few similarities in form. These expressions nevertheless share the basic property that they can be linked to pragmatic scales, just as the expressions under A. There is one important difference between A and B, though. Whereas the expressions under A denote a scalar minimum, the expressions under B denote the other end of a scale, the maximum. For instance, the expressions in jaren (in years) and in eeuwen (in ages) can both be regarded as a maximum on a scale of duration, depending on which measurement of time is taken. This group of in + measurement of time also appears as a cluster of NPIs in English, for example: in hours, in days, in weeks, in months, in years, in decades, in ages, in a hundred years, in a million years, in a coon's age. As Von Bergen and Von Bergen (1993) note, these expressions, in combination with their licensers, result in strong statements. See the following example:

The combination of a high scalar end point with negation maximizes the conventional implicatures which can be made. If it is true that you have not seen someone in what can be considered a maximum amount of time (e.g. in ages), then it must indeed have been very long since you saw that person. The same effect can be achieved with the other expressions listed under B, as for example with voor goud (for gold). This expression is typically used in statements such as (26):

By saying this, you express in a strong way that you do not want to live in Amsterdam; not even for gold, an offer which can be regarded as denoting a maximum value. This emphatic statement can be even further strengthened by adding another NPI of maximum quantity, ter wereld (in the world):

Up to Chapter 1 Table of Contents

Indifference

The NPIs listed under C are described by Hoeksema (1994a) as psychological verbs which assess the affective aspects of the relation between a human subject and something else. The heading of this category is `indifference', not so much because these expressions themselves denote indifference (on the contrary, their meaning per se is that of something making a difference), but rather since a large part of their licensing environments results in such a negative meaning. See the examples below:

These verbs of indifference may occur in combination with the NPs of minimum quantity, listed under A, to strengthen the statement. Compare (29) with (30):

Up to Chapter 1 Table of Contents

Intolerance/dislike

The expressions listed under D assess the ability of a subject to get along with someone, or to endure a person or a situation. In the same way as with the expressions listed under C, the meaning of these verbal combinations is not negative by itself, but is supplied by their licensing environments. Hence the negative heading: `intolerance/dislike'. An example is given below:

As Hoeksema (1994a) notes, the NPI of intolerance/dislike kunnen uitstaan (can stand) has a related adjective, onuitstaanbaar (insufferable). The prefix on marks the negative meaning of this expression. This negative marker is obligatory; there exists no such word as uitstaanbaar. Therefore, Hoeksema regards onuitstaanbaar as an NPI at the morphological level. For more discussion, see Van der Wouden 1995. [5]

Up to Chapter 1 Table of Contents

Incapacity

These expressions denote the extent to which a subject is capable of dealing with a certain situation. Again, as a result of the licensing environments for these expressions, this capacity turns out to be either low, insufficient, or completely absent. Hence the heading `incapacity'. A distinction can be made between personal expressions, in which a human subject has to deal with a situation, such as iets kunnen helpen (can help something) and iets kunnen laten (to be able to desist something), and impersonal ones, in which the subject is non- human, such as baten and iets uithalen (English: to be of avail). An example of each variant is given below:

It should be noted that there is a recurrent pattern in the NPIs in categories C, D, and E: many of the verbal expressions listed here are formed with the modal verb kunnen (can/to be able). This points to a common characteristic underlying the expressions in these categories: the notion of being able (or rather, the notion of not being able) to accomplish something. It might therefore be appropriate to regard C, D, and E as subcategories under the more general heading of `inability': the expressions under C as expressing inability to make a difference, the expressions under D as expressing inability to endure someone or something, and the expressions under E as expressing inability to deal with a certain situation.

Up to Chapter 1 Table of Contents

Matching

The NPIs listed under F express a notion of comparison, pertaining to the subject or object of the sentence in which they appear. Both possible outcomes, equality or inequality, are represented in these expressions. The NPI ergens voor onder doen (to knuckle under for something), for instance, establishes a relationship of equality, precisely by means of the licensing environments in which this expression can occur. See the example below:

The essence of the other three expressions under F is that there is no equality. Again, it is the licensing environments which establish this effect. The `bare' expressions - that is, the expressions without the context in which they normally appear - denote a match: kunnen tippen aan, kunnen evenaren (can hold a candle to), zijn weerga kennen (literally: to know one's match). It is by means of their licensers that these expressions of match are turned into expressions of mismatch, for example: niet kunnen tippen aan, niet kunnen evenaren (not be able to hold a candle to), zijn weerga niet kennen (to be without match). See the following examples:

Up to Chapter 1 Table of Contents

Unawareness

In this category, the central notion is awareness, or rather unawareness, due to the licensing restrictions on these expressions. These NPIs assess the extent to which a subject is aware of a certain situation, for example as in benul hebben and idee hebben (English to have a notion/an idea):

The notion of unawareness can be strengthened by means of the adjective flauw (in English, the superlative the faintest):

Another expression of unawareness is the NPI weet hebben van (literally: to have knowledge of; cf. (39)), which is closely related to van...of...weten (to know of...or...). The dots in the latter expression can be filled by idiomatic combinations of NPs of minimum quantity, which lend rhetorical force to the statement (cf. (40)):

Up to Chapter 1 Table of Contents

Degree

There is a relatively small cluster of NPIs that assess the degree of a certain property, or the extent to which a certain state of affairs is the case. For example noemenswaardig (literally: worth mentioning), which can be used both as an adjective, as in a), and as an adverb, as in b):

Another NPI of degree is the adverb bijster (all that). Besides its polarity sensitivity, there is an additional restriction on its use: bijster can only modify the positive element of a pair of antonyms (Klein and Hoeksema 1994). Compare a) with b):

Up to Chapter 1 Table of Contents

Special cases

Last but certainly not least, it should be noted that not all NPIs cluster in certain meaning domains. There is also a number of NPIs, listed in the table as `special cases', which cannot be placed in a specific category on grounds of a shared meaning. These NPIs should by no means be considered as a negligible remainder. The relatively high frequency of these expressions in daily language use makes the category of special cases one of importance. [6]

Qua meaning, these expressions are on their own and appear to be unrelated to other NPIs. For example, the NPI verb hoeven (have to/need), of which an example was given in (17), expresses a notion of necessity, a meaning which does not return in other Dutch NPIs. Likewise, the meaning of the temporal adverb meer (anymore) remains rather on its own among NPIs. True, there is another temporal NPI adverb, ooit (ever), but ooit and meer have no further characteristics in common: whereas meer expresses the fact that some change has taken place over time, ooit is an existential quantifier over points of time, either in the past, as in (43), or in the future, as in (44):

Another special case is ook maar (even/so much as). Although the distribution of this expression is closely linked to the NPIs in category A (cf. examples (12) - (16), it clearly has a different meaning content. Whereas the NPs in category A all denote lower end points of scales, ook maar has no clear lexical content; as an adverbial prefix, it adds emphasis to a statement by accentuating low scale extremities.

Up to Chapter 1 Table of Contents

1.5.3 — Idiosyncratic behavior

As was mentioned earlier, the inventory in Table 1.2 is by no means complete. Limited space is not the only reason for this. A number of polarity sensitive expressions has not been included in the table since it is not clear just whether they, strictly speaking, should be regarded as NPIs. To be sure, the distribution of these expressions is restricted to negative sentences, but the problematic factor is that their distribution is more restricted than is usually expected for NPIs; it appears that they can only occur in combination with one specific negative expression. Deviation from this standard negative expression results in anomaly. See the examples below:

The c)-sentences are ungrammatical or at least aberrant, but they are not uninterpretable; they express approximately the same meaning as the corresponding a)-sentences. It seems that the expressions at stake - mals zijn, als een ander, kant of wal raken, slag of stoot - have become so conventionalized with a particular negative expression, though for unknown reasons, that they are now used as simple units: niet mals zijn, als geen ander, kant noch wal raken, zonder slag of stoot. `Deconventionalizing' such expressions, as is done in the c)-examples, which express the same meaning only in an innovative way, results in odd utterances.

The question that now arises is whether these expressions with their close connection to one specific negative expression are eligible for NPI status. Characteristic of NPIs is that the restrictions on their distribution leave room for a varied pattern of usage, provided it is within the borderlines of licensing, as was shown in examples (5) - (7) and (12) - (16). In the above mentioned expressions, however, such variation is lacking. It could therefore seem more appropriate to regard these expressions, with their negative host included, as idioms or collocations, instead of as NPIs which are licensed in only one environment. There is, however, a reason for not drawing such an absolute line between NPIs on the one hand and negative idioms on the other. When the distributive properties of different NPIs are compared to each other, it becomes clear that they all to some extent show idiosyncratic behavior: not all NPIs are allowed in the same environments but some expressions clearly have a more restricted distribution than others. The licensing conditions on different NPIs can be regarded as a sliding scale, ranging from relative freedom to very strict constraints. The further down this scale expressions are, the more licensing may begin to look like collocational use. I regard all expressions as NPIs, that is, as expressions which have the basic characteristic in common that they need a licensing environment. Some of these expressions, for reasons as yet unclear, have developed into more restricted expressions than others.

On the basis of these varying licensing conditions, I divide NPIs loosely into three types, which are dubbed, in order of decreasing strictness, idiomatized NPIs, core NPIs, and semi-NPIs.

Up to Chapter 1 Table of Contents

Idiomatized NPIs

Some examples of idiomatized NPIs have been given above in (45) - (48). Characteristic of these expressions is that a negative environment is an absolute requirement for grammaticality (cf. the difference between the a)- and b)-examples in (45) - (48) and moreover, that there is a strong preference for one particular negative expression to act as a licenser (cf. the difference between the a)- and c)-examples). The favorite licenser may differ from one NPI to the other: for instance, the typical licenser for the NPIs in (45) - (48) is, respectively niet (not), geen (no), noch (nor), and zonder (without). It should further be noted that the preference for a particular licenser is strong but not absolutely compelling. For instance, Hoeksema (1992a) and Van der Wouden (1994a) maintain that mals zijn (to be soft) is not exclusively acceptable with the negation niet (not), but also with geenszins, allerminst, and allesbehalve (English: not at all, not in the least, and anything but):

Until now, only one such occurrence has been reported from actual language use (e.g. mals zijn with allesbehalve as a licenser), and the Hoeksema corpus contains only rare instances of mals zijn with other licensers than niet.[7] If combinations with other licensers were common, more occurrences should have been expected. The fact that so few examples are found indicates that the licensing conditions on mals zijn must indeed be very strict.

Up to Chapter 1 Table of Contents

Semi-NPIs

The other extreme in the hierarchy of strictness is constituted by semi-NPIs (a term first used by Hoeksema (1994a)), so called since the behavior of these expressions appears to be not too constrained. Examples of such semi-NPIs include various expressions of indifference, intolerance/dislike, and incapacity. Although these expressions have a strong tendency to occur in negative sentences, conditionals, questions, et cetera, they are also now and then encountered in affirmative sentences (cf. Hoeksema (1994a) for corpus-based percentages). Some affirmative utterances with semi-NPIs are given in the a)-examples below:

And yet, the distribution of semi-NPIs in affirmative sentences is not completely free, but subject to a subtle restriction, as is shown by the distinction between the a)- and b)-examples. The a)-examples are truly grammatical, but the acceptability of the b)-examples - provided these sentences are pronounced with a normal intonation - is questionable. The difference between the two sets of examples is that the b)-sentences are neutrally affirmative, whereas the a)-sentences are emphatically affirmative, due to the presence of (stressed) adverbs such as wel (really/indeed),[8] zeker (certainly), juist (only too), nogal (quite), and best (very well). These adverbs are typically used to mark affirmative contrast. This explains why the a)-sentences display an echo effect: the affirmative intensifiers create the impression that the negative meaning of some former proposition or assumption is counterbalanced. This effect is especially strong in the case of wel, zeker and juist, and more vague with nogal and best. As for wel, zeker and juist, there will in fact often be a preceding negative proposition, for instance Het kan jou toch niks schelen (`You don't care'), in the case of (50), or Jij geeft daar toch niks om (`You don't care about that'), in the case of (52). As for nogal and best, the echo effect is less pronounced; in many cases, there will be no actual negative proposition, but rather a contextually salient negative assumption. For example, (54) might occur in a context where both the speaker and the hearer hold the view that Frank is not always as easy to get along with.

Up to Chapter 1 Table of Contents

Core NPIs

The two types of NPIs discussed above, idiomatized NPIs and semi-NPIs, can be regarded as representing respectively the lower and upper boundary of variety in NPI licensing. The core NPIs fall in the area between these two extremes. Some examples are hoeven (have to/need), meer (anymore), and ook maar (even/as much as). The distribution of these expressions is not limited to one particular licensing environment; instead, they can occur with a variety of licensers. At the same time, they are excluded from affirmative utterances, such as the ones discussed in the preceding paragraph. See the following examples:

It may be maintained that core NPIs are not always ungrammatical in affirmative utterances, in consideration of the fact that some of them may occur in utterances with highly stressed wel. See for instance the following example with hoeven (have to/need):

For one thing, it should be pointed out that hoeven `licensed' by wel is considered as infelicitous, especially in a formal register. The combination of hoeven with stressed wel is only considered as correct when it is implicit, in an elliptic affirmative clause following a negative clause with hoeven. For instance:

Such elliptic utterances are rather common, as opposed to utterances like (58), which occur rarely and only in informal settings. Moreover, at such informal occasions, hoeven can only be licensed by wel in sentences with the strongest possible echo effect. This requires, firstly, a directly preceding negative utterance. A negative assumption alone is insufficient. Thus, (58) as an independent utterance without an earlier explicitly expressed negative proposition is not acceptable. Secondly, the affirmative utterance containing hoeven and wel has to be echoic up to deixis (Drozd 1995): alternations of deictic proforms to preserve reference are allowed, but all other substantial modifications are prohibited. Thus, given the following question Q, only A1 is acceptable as answer, but not A2:

In this respect, wel has characteristics in common with metalinguistic negation. In fact, Horn's (1989) definition of metalinguistic negation is equally well applicable to wel: a device for objecting to a previous utterance on any grounds whatever, including the conventional or conversational implicata it potentially induces, its morphology, its style or register, or its phonetic realization (Horn 1989: 363). We might therefore call wel a metalinguistic affirmation, analogous to metalinguistic negation. They are two sides of the same coin: while metalinguistic negation reverses an affirmative proposition (p becomes ¬p), metalinguistic affirmation reverses a negative proposition (¬p becomes p). The operation involved in both is the same, i.e. contradiction.

Van der Wouden (1994a) notes that almost all laws of grammar may be offended in metalinguistic use, and that it is therefore not ad hoc to stipulate that polarity items do not have to be licensed in cases of metalinguistic use. Indeed, almost all NPIs from Table 1.2 are allowed to appear with metalinguistic affirmation. Yet, I think wel should not be considered as completely irrelevant to the rules underlying the distribution of NPIs. As an affirmative intensifier, just as zeker (certainly), juist (only too), nogal (quite), and best (very well), wel certainly has a role to play in the `licensing' of NPIs, since neutrally affirmative utterances with NPIs are not acceptable. The difference between core NPIs and semi-NPIs with respect to such intensifying affirmation thus appears to be that core NPIs require the affirmative intensifier with the strongest echo effect, wel, whereas semi-NPIs also accept intensifiers with a weaker echo, such as zeker (certainly), juist (only too), nogal (quite), and best (very well).

With the three types of expressions identified thus far - idiomatized NPIs, core NPIs, and semi-NPIs - only the rough outlines of variable licensing conditions on NPIs can be drawn. To do justice to the full complexity of their distribution, a much more fine-grained categorization would be called for. The fact is that the NPIs which fall within the boundaries of these three types are still far from homogeneous in their licensing. Within these categories, the expressions are marked off from each other by further preferences and aversions for specific environments. This idiosyncratic behavior of NPIs will be further pursued in section 1.6.3.

Up to Chapter 1 Table of Contents

1.5.4 — Crosslinguistic differences and historical developments

Polarity sensitivity is not a random feature of arbitrary expressions. This became clear from Table 1.2, where Dutch NPIs were shown to cluster in certain domains of meaning. Such semantic clustering is not a characteristic of Dutch NPIs alone, but is evident in other languages as well. This is illustrated in Table 1.3, which contains a few examples of NPIs from English, German, and Norwegian (more German NPIs are given in Kürschner 1983, and more English examples in Von Bergen and Von Bergen 1993; Hoeksema 1994a; Israel 1994). The crosslinguistic comparisons in this table should not be taken as evidence that NPIs always have exact equivalents in other languages. As will become clear shortly, this is not necessarily the case. What this table is meant to show in the first place is that NPIs in a variety of languages can be found in the same meaning domains as the ones which were defined earlier in Table 1.2: minimum and maximum quantity, indifference, intolerance/dislike, et cetera.

The crosslinguistic similarities between NPIs go further than semantic clustering alone; NPIs across languages may be each other's exact or very near equivalents. For instance, the verbal idioms of minimum quantity listed under A in the table are subtle variations on the same theme: een vinger uitsteken (to reach out a finger) in Dutch corresponds to to lift a finger in English, to einen Finger krummachen (to bend a finger) in German, and to å røre en finger (to move a finger) in Norwegian. In the same way, all four languages have an equivalent NPI of intolerance/dislike, stemming from the same root: Dutch kunnen uitstaan, English can stand, German ausstehen können, and Norwegian å kunne utstå (see also Hoeksema 1994a for some remarks on the etymology of polarity sensitive verbs in Dutch and English).

On the other hand, there are also remarkable crosslinguistic differences with respect to polarity sensitivity. An expression which is an NPI in one language may have a non-polarity equivalent in an other language. See for instance the special cases listed under I. The much-discussed English NPI any has no equivalent in Dutch, German, or Norwegian. That is to say, any has not one exact polarity sensitive equivalent in these languages. The negative/positive polarity pair some/any in English corresponds in part to non-polarity sensitive determiners and quantifiers in Dutch, German, and Norwegian.[9] On the other hand, English shares with Norwegian a polarity sensitive variant of too (either in English, heller in Norwegian), as opposed to Dutch and German, in which the equivalent meaning is not expressed by an NPI.

Crosslinguistic comparisons may also show more gradual distinctions in polarity sensitivity. For instance, a corpus investigation of the Dutch degree adverb noemenswaardig and its German equivalent nennenswert (Klein 1994) shows that noemenswaardig has a much stronger tendency to occur in licensing environments than German nennenswert. In this respect, nennenswert resembles Norwegian nevneverdig, which also now and then occurs in affirmative sentences. More subtle still are the crosslinguistic differences with respect to Dutch hoeven, English need, German brauchen, and Norwegian å behøve. Whereas hoeven is strictly polarity sensitive, the corresponding verbs in English, German, and Norwegian are only NPIs in their function as auxiliaries. As main verbs, they have no particular restrictions on their distribution. This is illustrated in the following examples:

The distinction between a polarity sensitive auxiliary and a non-polarity sensitive main verb corresponds to two separate - though closely related - verbs in Dutch: hoeven and behoeven, respectively. Behoeven is a rather formal and archaic verb, which nowadays is little used in Dutch; much more frequent is its synonym nodig hebben (to be in need of). The similarity of behoeven with Norwegian å behøve is immediately striking. Indeed, the historical developments which appear to have taken place with Dutch behoeven are now repeating themselves with this Norwegian equivalent. In contemporary Norwegian, å behøve as a main verb is being pushed to the background, in favor of another verb expressing the same meaning, å trenge (to be in need of). Interestingly, exactly the opposite pattern is found in English, where it is not need as a main verb which is formal and archaic, but its use as an auxiliary. Contemporary German shows still another state of affairs: brauchen as a main verb occurs almost twice as often as the auxiliary brauchen (cf. Pfeffer 1973), but there appear to be remnants of an earlier non-polarity use of brauchen as auxiliary.[10]

Such observations give reason to assume that the crosslinguistic connections in polarity sensitivity will show a considerably more entwined picture when historical developments are taken into account as well. Until now, there has been relatively little research unravelling patterns in the grammaticalization process of NPIs in different languages. The results obtained so far (cf. Hoeksema (1994b) on the changes which very recently have taken place with Dutch ooit, and Hoeksema (1994a)) justify the conclusion that polarity sensitivity is not a static property of certain expressions, but rather is subject to continuous fluctuations. As time passes by, expressions may lose their polarity sensitive character, or develop an NPI variant besides a non-NPI variant.

Up to Chapter 1 Table of Contents

1.6 — Theories of polarity sensitivity

The observations discussed in the preceding subsections - which, it should be added, are by no means exhaustive - have given an impression of the many-sided phenomenon of polarity sensitivity: the sometimes obvious but in other instances obscure reasons for a restricted distribution, the clustering in meaning domains, the internal variation with respect to licensing conditions, crosslinguistic similarities and differences, and changes over time. How should such a complex phenomenon be accounted for in linguistic theory? Israel (1994) distinguishes three problems which a theory of polarity sensitivity, doing justice to the entire gamut of facts (as Israel puts it: a complete theory of polarity sensitivity) would have to address. First of all there is the licensing problem, in other words: what exactly is it that NPIs are sensitive to? The question here is whether the diverse array of licensing environments can be characterized as a natural class, in such a way that there is a common denominator in these contexts which makes it possible for them to license NPIs. The second problem is closely related to the first one, addressing the underlying reason for an expression's polarity sensitivity. Israel dubs this the sensitivity problem: while the licensing problem asks why certain contexts trigger polarity sensitivity, the sensitivity problem asks what makes certain forms so sensitive to these contexts (Israel 1994: 4). The third problem that needs to be accounted for in a theory of polarity sensitivity is the diversity problem, which bears upon the fact that not all NPIs are restricted to the same licensing environments. Why is it that some NPIs cannot occur in environments in which other NPIs are licensed?

The diverse facets of polarity sensitivity as they were presented in the preceding subsections appear to be accommodated in this tripartite division, with the exception of crosslinguistic differences and historical developments. Although Israel does not explicitly mention these two aspects, it appears natural to classify them both under a more elaborate version of the sensitivity problem. When the entwined crosslinguistic differences in polarity sensitivity are regarded as a matter of one expression being further on the way in the grammaticalization process than the other (as is done in Hoeksema 1994a), and as is also suggested by the cyclic-like movements of to need and its translations in other languages (cf. section 1.5.4), then the sensitivity problem can be extended as follows: why is it that certain expressions are so sensitive to specific contexts, and why do they in the course of time develop into NPIs, which need these contexts to be licensed?

Although polarity sensitivity in the last three decades has been a frequently addressed subject matter, which has been studied from widely different points of view, there is as yet no complete theory of this phenomenon, in the sense that there is no unified account for all three problems: licen- sing, sensitivity, and diversity. This may in part be due to the fact that the perceived boundaries of what constitutes the phenomenon of polarity sensitivity itself have stretched in the course of time: as the research proliferated and more facts became known, the range and complexity of the phenomenon has increased. Early research on polarity sensitivity was mainly preoccupied with English and only few facts were known about the situation in other languages. Moreover, the focus of attention was on a relatively small set of English NPIs; on any in particular - sometimes, but not always, in relation to its positive polarity counterpart some - and on a few others, such as ever, yet, and the class of minimum quantity NPIs. Later on, more NPI facts became known from a number of languages, and the range of expressions studied, both from English and other languages, increased. It may very well be that one of the contributory factors in this later development has been the availability of electronic speech corpora, which made it feasible to study general patterns in large corpora of actual language use, such as the relative frequency of different NPIs, the most common and marginal licensers, the varying licensing conditions for different NPIs, et cetera. Such corpora also allow for crosslinguistic comparisons in polarity sensitivity and for studies on the changes NPIs may undergo in the course of time.

No attempt will be made here to give a complete and chronological summary of the various theoretical approaches and their background assumptions. Such an overview falls outside the scope of this thesis, in which the focus is on psycholinguistic aspects of polarity sensitivity. Only the chief features of some theories will be set out in the following subsections.[11]

Up to Chapter 1 Table of Contents

1.6.1 — Ladusaw: semantic entailment

As Israel (1994) points out, theories on polarity sensitivity have tended to focus on the licensing problem, while the problems of sensitivity and diversity have received relatively little attention. An exception is Fauconnier's theory of pragmatic scales - discussed in some detail in sections 1.3 and 1.4 - which addresses both the licensing problem and the sensitivity problem. This account, however, has the limitation that it is applicable to only a subcategory of NPIs, the expressions of minimum quantity. [12] As Fauconnier himself notes: many standard polarity items (already, yet, bother, still etc...) cannot be regarded as endpoints on scales in any obvious way (Fauconnier 1975: 198). The limited range of the scale-reversal theory also explains why the diversity problem is not a matter of concern here: as the expressions in this specific subcategory so closely resemble each other in meaning and pragmatics, they adhere to the same licensing conditions.

An account which is closely connected to that of Fauconnier but which appears to be applicable to a larger class of NPIs, is Ladusaw's (1979), in which NPI licensing is cast in terms of logical implications rather than in pragmatic implicatures. Of crucial importance in Ladusaw's approach is the inference pattern known as downward entailment, which is characterized below:

(63)	Downward entailment:
	X Y Z	There are no vegetables in the fridge
	W < Y	carrots < vegetables
	X W Z	There are no carrots in the fridge

Given a premise of the form X Y Z, and given that W is a subset of Y, then one may conclude X W Z. In other words, an inference from the general to the more specific is validated. Another example is given in (64):

When it is true that John did not eat vegetables, one is allowed to substitute the specific term carrots for the more general term (i.e. the superset) vegetables. This inference is valid because of the presence of a negative expression in the premise. As can be seen in (65), downward entailment is not supported in the corresponding affirmative sentence:

The fact that John ate vegetables does not logically imply that he ate carrots, since the possibility cannot be excluded that John in fact ate cabbage, or peas, or a combination of other vegetables. Instead, in affirmative sentences, one is allowed to make inferences in the opposite direction, from the specific to the general (i.e. from subsets to supersets). This inference pattern is known as upward entailment:

Ladusaw argues that NPIs are restricted to environments which support inferences from supersets to subsets. In his own terms: NPIs must occur in the scope of a downward entailing expression. This concise rule makes good predictions about the environments in which NPIs will be licensed and the environments from which they will be blocked. For instance, it correctly predicts the grammaticality of NPIs in the scope of nobody, never, nowhere, few, without, too...to, rarely, et cetera, which are all downward entailing expressions, as well as the ungrammaticality of sentences with NPIs in upward entailing environments, such as everybody, always, everywhere, many, with, enough to, and often, or with NPIs in environments which are neither downward nor upward entailing, such as exactly n. The account also can handle rather subtle distinctions between licensers and non-licensers, as is shown in the following two examples (from Ladusaw 1979):

The grammaticality of (67) as opposed to the ungrammaticality of (68) is explained by the fact that at most is a downward entailing expression, whereas at least is upward entailing.

Ladusaw's solution to the licensing problem differs from that of Fauconnier in that the mechanism considered responsible for NPI licensing is cast in terms of truth-conditional semantics rather than in pragmatic, context-dependent implicatures. The asset of this approach is that it appears to hold for a much larger class of NPIs than the scale-reversal account, which is limited to one specific class of expressions. The trade-off, however, is that this approach offers no natural solution to the sensitivity problem; in other words, it is not clear why the distribution of NPIs should be restricted to downward entailing environments.[13]

Up to Chapter 1 Table of Contents

1.6.2 — Dowty: facilitation of inferences

A recent proposal, based on Ladusaw's theory of NPI licensing, has been made by Dowty (1994), who states that NPIs allow for easier application of deductive rules. In this account, the presence of an NPI in a sentence functions as a flag, indicating that inferences from supersets to subsets are allowed: we can hypothesize that one important reason for the existence of NPI (...) marking is to directly mark positions syntactically which are subject to downward entailing inferences (Dowty 1994: 125). No examples are given as to how such shortcut inferences might work with various types of NPIs; it appears that Dowty devised this hypothesis with particular reference to noun phrase NPIs (any NP, anybody): the occurrence of a NPI-marked noun phrase may immediately tell the hearer, in advance of understanding a whole clause compositionally, that there is no possible semantically entailed discourse referent for that NP (Dowty 1994: 126). Indeed, for this specific type of NPI such inferences appear to meet a communicative purpose (see also Kadmon and Landman's (1993) analysis of any in terms of widening and strengthening). For other types of NPIs, however, it remains as yet unclear what the point is of downward entailing inferences to be made and what they possibly could have to add to the sentence meaning; let alone that it is evident why such inferencing would have to be facilitated by a shortcut. Moreover, taking into account that the relationship between NPIs and downward entailing environments is not an exclusive one - by no means all downward entailing expressions in daily language use contain NPIs - the question is still open as to why, in some downward-entailing environments, inferencing would have to be facilitated by the presence of an NPI, while a multiplicity of downward entailing environments seem able to do without such inference flags.

Up to Chapter 1 Table of Contents

1.6.3 — Zwarts: the relative strength of licensers

Another elaboration of Ladusaw's theory of semantic entailment has been proposed by Zwarts (1981; 1986; 1995). Zwarts' extension has bearing on a problem which is not addressed by Ladusaw: diversity - that is, the observation that not all NPIs behave in the same manner as far as licensing conditions are concerned (cf. section 1.5.3). Zwarts notes that there are NPIs in Dutch which are blocked in the scope of some downward entailing expressions, as opposed to other NPIs, which are licensed in these environments. See for instance the following pair of examples (from Zwarts 1995):

The sentences in (69) contain the NPI ook maar (even/as much as), which in both cases occurs in the scope of a downward entailing expression: niemand (no one) in a), and hoogstens (at most) in b). Contrary to what Ladusaw's account would predict, however, the b)-example is ungrammatical; hoogstens is not a correct licenser for ook maar. The examples in (70) show that the state of affairs is different for hoeven (have to/need): this NPI is licensed by niemand as well as hoogstens.

Zwarts attempts to explain such differences between NPIs by appealing to the relative strength of NPIs and licensing environments, with strong NPIs demanding strong licensers and weaker NPIs accepting weaker ones. In this approach, as in Ladusaw's, the licensing of NPIs is directly related to logical entailment properties, but now the class of downward entailing expressions is refined and subdivided into strong and weaker expressions, according to their ability to support certain inference patterns.

Some expressions, for instance no one and not, not only support downward entailment, but also the inference pattern of anti-additivity, given in (71). Examples with this pattern are given in (72) and (73).

The expression not, in addition to being downward entailing and anti-additive, also supports the antimorphic inference pattern, which is represented in (74). An example with this pattern is given in (75):

On the basis of their ability to support different inference patterns, Zwarts constructed a hierarchy of negative expressions. A sample of these expressions is given in Table 1.4.

In this hierarchy, not occupies the position of the strongest negation (dubbed the classical negation), in virtue of its ability to support both the downward entailing, anti-additive, and antimorphic inference patterns. One step down in the hierarchy are the somewhat weaker negative forms (regular negations), such as nothing and never, which are both anti-additive and downward entailing, but not antimorphic. Lowest in the hierarchy are the weakest negations (minimal negations), such as few and at most, which are only downward entailing.

Zwarts makes a direct connection from this hierarchy of negative expressions to the licensing of Dutch NPIs, formulated in three laws of negative polarity:

Since the various forms of negation stand in a subset-superset relation to each other (all are downward entailing, but only some of them are in addition anti-additive or even antimorphic), these laws mean that weak NPIs, which accept downward entailing expressions as licensers, can occur in a relatively large variety of environments. Superstrong NPIs on the other hand have a very restricted distribution, since they make a much stronger demand, requiring an antimorphic expression as licenser.[14] In accordance with these laws, Zwarts regards mals zijn (to be soft) with its strong preference for the classical negation niet (not) as licenser (cf. the examples in (45)) as a superstrong NPI. In the same way, the difference between (69) and (70) is accounted for by classifying ook maar, which requires a regular negation (see also Hoeksema 1983), as a strong NPI and hoeven, the NPI which takes minimal negations as licensers, as a weak one.

Although the superstrong/strong/weak distinction has been demonstrated to hold for a range of NPIs (cf. Zwarts 1995 and Van der Wouden 1994a for Dutch examples, and Van der Wouden 1994b for examples from English), it does not as yet offer a complete solution to the diversity problem. For instance, the licensing con- ditions on the two most frequent Dutch NPIs, hoeven (have to/need) and meer (anymore), do not easily fit their classification as weak NPIs (for a discussion, see Van der Wouden 1994a). Also, once expressions such as mals zijn are singled out as a special class of NPIs, the question comes up as to how this class of superstrong NPIs relates to other idiomatized NPIs with a comparably strong preference for one particular environment (cf. examples (46) - (48)), which remain unidentified as a category. Hoeksema (1992a; 1992b) argues that the diversity of licensing conditions on NPIs cannot be cast solely in logical terms, but that there might also be a classification possible along rhetorical lines. This suggestion is prompted by the observation that specific licensing conditions on NPIs appear to correlate with their rhetorical function, whether this is emphatic or not. A similar approach has been advocated and worked out for a range of English NPIs by Israel (1994), who claims that the acceptability of NPIs in sentences is determined by their informational value (i.e. emphatic or understating) in specific contexts.

Up to Chapter 1 Table of Contents

1.6.4 — Linebarger: negative implicatures

There is a number of polarity facts which are problematic for Ladusaw's account in terms of downward entailingness, and consequently also for approaches which take this theory as a starting point, such as Dowty's approach to sensitivity, and Zwarts' and Van der Wouden's treatment of diversity. The problem is that NPIs sometimes occur in environments which are not downward entailing, or in environments of which the logical entailment patterns are unclear; for instance in the complements of some adversative verbs, in questions, and in the scope of only or long after. Some of these problems are addressed by Linebarger (1980; 1987; 1991). [15]

In Linebarger's approach, which in turn is partly based on Baker (1970), it is argued that the licensing of NPIs reflects an interplay between syntax and pragmatics. The primary licensing mechanism in this account is negation; NPIs should always appear in its immediate scope in the logical form of the sentence. The acceptability of NPIs in environments which do not contain any overt negation is explained by a derivative licensing mechanism, which is pragmatic in nature. According to this secondary principle, NPIs are also licensed in environments which allude to the paradigm case of sentential negation and express a negative implicature. Linebarger argues that the latter environments include cases which have been mentioned as being problematic under Ladusaw's account, such as adversative predicates, only, and long after. Examples of the negative implicatures she has in mind are given below (from Linebarger 1987):

Under Linebarger's account, the fact that these NPIs are acceptable in the non-negative a)-sentences is explained by the availability of negative implicatures, which are shown in the b)-examples.

On the basis of such test cases, Linebarger concludes that her account in terms of negative implicatures is preferable to the downward entailing account. It should, however, be noted that the negative implicature account itself faces a serious problem. As this approach is not constrained enough, it falsely predicts NPIs to be correct in environments which in fact are non-licensers. Consider for instance the pair of antonyms rarely/often. The most plausible candidate for a negative implicature associated with rarely is not often. On the basis of this implicature, Linebarger's account correctly predicts that NPIs are excepted in the scope of rarely. Following the same line of argument, however, it is not clear how NPIs should be blocked from occurring in the scope of often, since the most plausible negative implicature for this expression is not rarely. The same problem arises with the pair at most/at least, with the respective negative implicatures not more than and not less than. Linebarger's account provides no criterion by which to decide which of these implicatures is a correct NPI licenser. On the other hand, the licensing power of at most, as opposed to at least, follows directly from Ladusaw's account (cf. section 1.6.1). Consequently, Ladusaw's argument against an account in terms of negative implicatures is that there is no principled way in which we could tell which entailments of a sentence containing a certain lexical item are relevant to establishing that lexical item as a trigger for NPI's, particularly for pairs of items which are related to each other in a way that makes each entail the negation of the other (Ladusaw 1979: 105). A principled way to distinguish between implicatures which are relevant to NPI licensing and implicatures which are not is exactly what seems to be needed for a true solution to the licensing problem.

Up to Chapter 1 Table of Contents

1.6.5 — Negative polarity items: one phenomenon?

As the preceding subsections have shown, there is no lack of proposals for a theory of polarity sensitivity. But there is, as yet, still no one complete account that, in itself, can explain how the licensing, sensitivity, and diversity of NPIs should be captured. In fact, now that the range of phenomena classified under the umbrella of polarity sensitivity has increased over the years, it is sometimes felt that such a unified account may not even be feasible. This view rests on the assumption that NPIs in essence cannot be treated all alike; polarity sensitive expressions manifest themselves as one phenomenon on the surface, but there may be a diversity of reasons underlying their common distributional patterns (Hoeksema 1992b; Van der Wouden 1994a; Von Klopp 1993). If this line is further pursued, it would not necessarily mean that polarity sensitivity as a cover term for one phenomenon would cease to exist. Rather, it would put the phenomenon - its licensing, sensitivity, and diversity - into a different light: as a phenomenon which on the surface is homogeneous in that it unmistakably displays similar distributional patterns (e.g. licensing), but which is heterogeneous qua underlying reasons for this distribution (e.g. its sensitivity); a heterogeneity which in turn might be responsible for the fact that the distributional patterns to some extent are idiosyncratic (e.g. diversity).[16] Within this perspective, it is possible for different approaches to coexist, and even seemingly incompatible accounts need not exclude each other. For instance, the categories distinguished by Israel (1994) as high scalar emphatics and low scalar understaters are not in conflict with Hoeksema's (1992b) proposal to carve the class of NPIs up in emphatic and non-emphatic NPIs.

Clearly, in theories of polarity sensitivity, the last word has not been spoken yet. We will leave this issue here, and now move on to the central topic of this thesis: the practice of NPIs - that is, NPIs as they are used by speakers of a language.

Up to Chapter 1 Table of Contents

1.7 — Issues in the psycholinguistic study of NPIs

Regardless of whether NPIs theoretically should be accommodated under different categories or not, as far as their appearance in language use is concerned these expressions are uniform with respect to one important characteristic: a restricted distribution. More precisely a distribution in sentences which is not completely predictable from their syntactic properties (Ladusaw 1979: 1). In other words: NPIs are negative exceptions to rules which normally apply to members of specific syntactic categories. For instance, the polarity sensitive verb hoeven (have to/need) differs from other verbs in that additional restrictions apply to its distribution, besides the normal syntactic rules.

Adult speakers are confident with the restricted distribution of NPIs; they use these expressions apparently as a matter of course only in licensing environments. Clearly, in one way or another, language users know in which sentences NPIs are allowed and from which sentences they are blocked. An interesting question to be asked is what it means to `know' the conditions for correct NPI use. Does the competence of adult speakers involve knowledge of licensing rules of some sort? And also: how implicit or explicit is this knowledge; do speakers know why they use NPIs the way they do? When general patterns in actual NPI use are displayed in percentages of licensing environments (cf. Table 1.1), it is clear that NPIs do not occur as fixed collocations with a particular licensing expression. Rather, speakers use these expressions creatively within the boundaries of licensing. This could mean that NPI use is indeed rule-based, but it need not necessarily be the case. It is also possible that speakers use NPIs the way they do because the meaning and function of these expressions simply leave them no other choice; the communicative function of NPIs might be so transparent for language users that their licensers so to speak automatically come with them. To put it boldly: could it be that to know the meaning of an NPI implies to know its distributive properties? This question amounts to raising the sensitivity problem again, now with the focus on user intuitions.

Up to Chapter 1 Table of Contents

1.7.1 — Users' intuitions about polarity sensitivity

In order to get an indication of speakers' awareness of the conditions for correct NPI use, several native speakers of Dutch - all non-linguists - were asked to judge the grammaticality of a number of sentences with NPIs, and to do some introspection on their own judgments. The subjects were shown three pairs of sentences with NPIs. (The order in which these pairs of sentences were given varied per subject.) Of each pair, only one sentence was correctly licensed:

In personal interviews, the subjects were asked to indicate for each pair of sentences which one of the two was better, a) or b). After that, they were questioned about the motives for their answer.

As was to be expected, none of the subjects had any difficulty with the grammaticality judgments: the licensed sentences were always, without hesitation, pointed out as being the correct ones. As for the motivations behind these judgments, a clear division emerged between (80) and (81) on the one hand, and (82) on the other. The subjects appeared not to be able to explain the reasons for their judgments about (80) and (81), with the NPIs hoeven (have to/need) and kunnen schelen (to care), respectively. The only `motivation' they came up with was a repetition of their earlier judgments. For instance, with reference to the unlicensed sentences, `it doesn't feel right', `it's simply wrong', or `technically speaking it's not right', and with reference to the licensed sentences: `there is no reason; you just say it that way'. Moreover, when the subjects were asked to consider whether the different meaning of the a)- and b)-sentence could have anything to do with their grammaticality or ungrammaticality, no new insights were gained. Some subjects, by way of answer, just gave the correct equivalent of the unlicensed sentence, in which a non-polarity sensitive expression with the same meaning was substituted for the NPI: erg vinden (to bother) instead of kunnen schelen (to care), and moeten (have to/must) instead of hoeven (have to/need).

As for (82), which contains the idiom of minimum quantity een deuk in een pakje boter kunnen schoppen (to be able to kick a dent in a block of butter), the state of affairs was quite different. Although some subjects in the first instance were not able to motivate their judgment here either, the question whether the meaning difference between the a)- and the b)-sentence possibly played a role in grammaticality clearly rang a bell. All subjects eventually came up with a motivation which - in more or less explicit terms - made reference to a minimum, and the conclusions which can or which cannot be drawn from such an end point, depending on whether the sentence is negative or not. In some cases, this explanation rather closely resembled the form of an implicature: `Butter is so soft. When you are not even able to kick a dent in butter, then it means that you are really capable of nothing.' Also, subjects made reference to the low informative value of an affirmative sentence with an NPI of minimum quantity: `B (e.g. the unlicensed sentence) doesn't mean anything. What does one want to say with it? A (e.g. the licensed sentence) means something extra[!]'. And: `Well look, if you had said that Jan can climb the highest mountain, # yes. But this (referring to the unlicensed sentence) doesn't mean anything.'

In the light of what was said earlier in section 1.5.2 (i.e. there is no apparent semantic/pragmatic necessity for the grammaticalization of some expressions as NPIs), these results hardly come as a surprise. The sensitivity problem still being a puzzle to linguists, this small-scale experiment indicates that language users do not have the key to sensitivity either. Only a subcategory of NPIs, the idioms of minimum quantity, appear to have a relatively transparent distribution, but the most frequently occurring NPIs lack such semantic or pragmatic directions for use. For the distribution of the latter, language users can think of no other reason than that it is `just the way one has to say it'. In other words, they just conform to linguistic rules.[17]

Up to Chapter 1 Table of Contents

1.7.2 — The acquisition of NPIs

How can adults be expected to have arrived at these licensing rules for NPI use? When a large number of NPIs offers no semantic or pragmatic grip on their restricted distribution, then we might ask by which means language users have come to know the licensing rules which they now, apparently as a matter of course, conform to. Is the nature of NPIs such that the ins and outs of licensing can be grasped in a relatively short period of time, or does it take the language learner through a time-consuming process of trial and error? With these questions we have arrived at the topic which will be our concern for the rest of this thesis: the acquisition of NPIs.

The basic tenet in language acquisition studies is that children acquire their mother tongue by constructing the finite set of linguistic rules which enables them to produce an infinite number of utterances. Likewise, in a study about the acquisition of NPIs, the question to be addressed is how children draw out the generalizations about NPIs which are necessary for these expressions to be used correctly and productively. This, however, is only one side of the coin. The idiosyncrasy of NPIs, characterized earlier as negative exceptions to the rules of grammar, urges an investigation of the converse question as well, that is: how do children deal with the hazards of making too broad generalizations, which would falsely predict NPIs to be acceptable in non-licensing environments? The latter question has bearing on the issue which has come to be known as the `no negative evidence problem' (cf. Bowerman 1988 for an overview).

Up to Chapter 1 Table of Contents

1.7.3 — The `no negative evidence problem'

Input language - that is, the speech as the child hears it being used around him - functions as the child's source of evidence for the language he is to acquire. Such primary linguistic data provides the child with positive evidence, information about what is possible in language, but not about what is not possible. As Braine (1971) convincingly pointed out, absence of negative evidence is problematic for general theories of learning which assume that language is acquired by testing hypothetical grammars against the input data, as this is a procedure for which both positive and negative feedback would be required. On a smaller scale, the lack of negative evidence affects a variety of linguistic rules which present partial regularities, for instance dative alternation and passivization. The question here is how children can be expected to discover that there happen to be exceptions to such rules. When no one is providing the child with evidence about what is not possible, exceptions to linguistic rules might seem to be invisible holes in grammar.

In the particular case of NPI acquisition, there are different issues for which absence of negative evidence is a matter of concern:

Up to Chapter 1 Table of Contents

NPIs as exceptions to rules of grammar

In the absence of explicit information about what is not possible, it is puzzling how a child should come to know that NPIs have a restricted distribution and are not acceptable in a large number of sentences in which they, given their membership of a certain syntactic category, should be allowed to appear. For instance, once expressions such as can stand and all that are assigned to the class of verbs and adverbs through exposure to positive evidence, the child would seem to need additional evidence to inform him that these expressions are exceptions to the rules and are not allowed in straightforwardly affirmative sentences. Positive evidence alone seems to be insufficient to discover a restricted distribution. Consider the following pairs of sentences:

Exposure to examples of how NPIs can be used (i.e. the positive evidence of the a)-sentences) does not provide the child with evidence that the b)-sentences are unacceptable. After all, exposure to sentences such as in c) neither counts as evidence against the acceptability of the d)-sentences.

Up to Chapter 1 Table of Contents

Correcting overgeneralizations

In principle, it is possible that the child, on the basis of positive evidence, draws out rules about NPIs which are overinclusive - that is, that these expressions adhere to the rules which normally apply to members of a specific syntactic category. Accordingly, the child's grammar would generate many correct utterances with NPIs, but also a number of ungrammatical ones in which NPIs occur in non-licensing environments.

If overgeneralizations with NPIs indeed occurred, then the absence of negative evidence would give rise to another puzzle: how can a child retreat from such errors without receiving direct evidence that these generalizations are not acceptable? Unlearning too general rules of language seems to require someone telling the child that his utterances are not acceptable. However, it has been documented (Brown and Hanlon 1970) that caretakers only rarely explicitly correct children's syntactic errors. Moreover, children appear not to be able to use such information when it is given, not even in apparently simple cases (Braine 1971).

There is much more to be said about this matter, in particular about the role of adult-child interaction and the possibility that the child can dispense with direct negative evidence since he receives sufficient indirect negative evidence, in the form of requests for clarification, misunderstandings, recastings, et cetera, which enables him to correct his errors (Hirsch-Pasek et al. 1984; Demetras et al. 1986). I will not anticipate these issues here, since it first remains to be seen whether children make overgeneralization errors with NPIs at all. It is part of the present study to find out whether this is indeed the case. Therefore, the first matter to be addressed in the next chapter is the very onset of NPIs in child language; are these expressions used in a restricted manner or not?

Up to Chapter 1 Table of Contents

1.7.4 — Outlook: acquiring the correct licensers for NPIs

Detecting a restricted distribution in the absence of negative evidence is one thing, but it is certainly not the end of the matter. An important issue then still has to be dealt with: how does the child come to know in exactly which environments NPIs are licensed? Against the background of the present chapter, in which the complexities of NPI licensing were laid out, the expectation is that the task of figuring out correct licensing conditions will present the child with rather complex matters, in which various modules of knowledge might interact.

In order to arrive at a truly productive rule about NPI use, it seems necessary for the child to gain insight into the characteristics of correct licensers. It is here that the theoretical approaches to polarity sensitivity enter the picture again. Do the various notions from theory which have been proposed as solutions to the licensing problem have a role to play in the acquisition process, in that they might be of help for the child who has to discover what counts as correct NPI licensers? This, of course, is a very interesting matter, which might shed new light on the interface between language acquisition and linguistic theories. A qualification needs to be made, though.

As became clear from the overview of theories, there is not one theoretical account which can handle all facts of polarity sensitivity. Within their own frameworks, different theories are capable of accounting for the most frequent patterns of licensing - that is, sentences in which NPIs occur in the scope of negation or another negative quantifier, such as nothing or nobody. Where the explanatory power of theories starts to differ is with less frequent licensing structures such as before-clauses, superlative constructions, adversative predicates like to surprise, to doubt, et cetera. The complex syntactic structure of these licensers makes it hardly likely that they will ever be encountered in young children's speech. At the age at which children can be expected to have mastered such structures, chances are big that they also have come already a long way with NPI acquisition. This means that it will be very hard to check the contrastive cases for theories against child language data.

The lack of theoretical contrastive cases in simple sentences also means that for the structures which may be expected to occur in early child speech, we have to rely heavily on error data in order to be able to infer anything about the child's progress in the definition of NPI licensers. Correct utterances at best indicate that the child apparently has mastered the system underlying NPI use, but do not provide any information as to which factors might have been decisive in order to arrive at that system. Errors, on the other hand, may give insight into the factors which are problematic to the child, and may indicate underlying attempts to adhere to intermediate rules, deviant from adult grammar.

Up to Chapter 1 Table of Contents

Priority ordering of rules: from narrow to wider generalizations

There is one theoretical approach to polarity sensitivity which lends itself to be investigated in relation to early stages in the acquisition process: the account of Zwarts (1995) in which a hierarchy of negative expressions constitutes the basis for laws on the licensing of various types of NPIs (cf. section 1.6.3). The subset-superset ordering of licensing rules in this account makes it possible to conceive of an order in the acquisition of NPI licensers which would be the most beneficial in the light of the absence of negative evidence. This alleged order, which can briefly be characterized as `conservative widening', would bypass the `no negative evidence problem' by starting out with a very narrow generalization, later on to be widened on the basis of positive evidence.

The notion of conservative widening is adopted from the subset principle (Berwick and Weinberg 1984; Berwick 1985), a proposed constraint on the order in which generalizations during acquisition should be made in order to prevent the `no negative evidence problem' from occurring: the particular principle that specific grammar rules trigger before general rules because a more general rule allows a broader range of surface sentences than a more specific one (Berwick and Weinberg 1984: 228). Briefly, the subset principle demands that a learning procedure should guess the narrowest possible language, consistent with positive evidence seen so far. By hypothesizing as narrow a target language as possible, the acquisition procedure is protected from disastrous overgeneralization (Berwick and Weinberg 1984: 233).

The subset principle has mainly been applied at the level of Universal Grammar, to deal with the problem when a correct parameter value cannot unambiguously be selected on the basis of input. Such a problem may arise in the special case when grammars generated by a particular parameter value stand in a subset-superset relation to each other. The constraint proposed by the subset principle is that, to avoid overgeneralization in such cases, children must always start out by setting the parameter value to the subset language, such that positive evidence will be available to reset it, if necessary.

The subset principle as a solution to the `no negative evidence problem' is still controversial. As the objections that have been raised have bearing on the conditions and preconditions for application of this principle to the acquisition of parameters of Universal Grammar, they will not be reviewed here (but see MacLaughlin 1995 for a discussion).

What may appear to be of relevance to the acquisition of NPIs is the basic idea that the `no negative evidence problem' can be circumvented if an acquisition order from narrow to broader generalizations were followed. Following this priority ordering of rules, it is possible to see a direct connection with Zwarts' hierarchy of negative expressions and the related laws of negative polarity. As was pointed out in section 1.6.3, these laws stand in a subset-superset relation to each other, with smaller sets representing narrower rules of licensing. The narrowest possible rule to be conceived of is that NPIs require the classical negation niet (not) as a licenser. Niet (not), the strongest negation in the hierarchy of negative expressions, can license all three types of NPIs: weak, strong, as well as superstrong ones. If children started out with this narrow restriction, which would allow NPIs to appear only in the presence of niet (not), they would be in a position where the wider licensing rules for strong and weak NPIs could eventually be discovered through positive evidence alone - that is, by sentences in the input in which NPIs occur with other licensers than niet (not). If the child, through exposure to such positive evidence, discovers that the requirement of a classical negation is too narrow for strong NPIs, the restriction could be widened to include regular negations. If this restriction in turn appears to be still too narrow for weak NPIs, it could be widened to include minimal negations as well, until eventually the adult way of licensing is achieved for each individual NPI.

This appears to be a safe and steady strategy. If children proceeded in this conservative manner - that is, never going beyond the model which is set by input speech - always hypothesizing the narrowest possible rule compatible with these positive data, they would be prevented from using NPIs too broadly, in non-licensing environments. Consequently, if children employed such a strategy of conservative widening, several predictions can be made about how the acquisition of NPIs would proceed:

• errors of NPIs in non-licensing environments would not occur
• children would start out using NPIs in a highly restricted manner, in the presence of niet (not) only; and
• the development towards more variation in licensing environments would proceed in the order from regular to minimal negations.

In the next chapter, which deals with the onset of NPIs in child language, it will be investigated whether these predictions indeed are in accordance with the acquisition data.

Up to Chapter 1 Table of Contents

Negation in terms of entailment patterns

Before progressing to an investigation of children's earliest utterances with NPIs, there is another concept from theory which needs special mentioning: the notion of entailment, as it can be found in Fauconnier, Ladusaw, Zwarts, and others. Aside from their possible role in NPI acquisition, entailment patterns may have implications in a larger context, which is indissolubly connected with the acquisition of NPIs, to wit the development of negation. Hitherto, in work on the acquisition of negation, it has been problematic finding a principled way to define the very object of study. In Pea's (1980a) words: we must confront a problem rarely acknowledged for child semantics in general, and not at all for child negation. What exactly is `negation'? Like many words, `negation' does not have any one central or defining essence, but a number of meanings that partake of family resemblance to one another (Pea 1980a: 160). Consequently, studies on the acquisition of negation have focused on quite a varied range of expressions, in many instances including just not, in other cases expanded to include no or never, but to my knowledge never including expressions which only intuitively were felt to express a negative meaning, such as few, without, or rarely.

The merit of formal semantic approaches to negation for psycholinguistics is that they allow for a natural and principled demarcation of this concept. Negation defined in terms of language-independent entailment patterns offers a new and challenging framework within which the acquisition of a large number of negative expressions and structures can be studied, language-internally as well as crosslinguistically.

On to next chapter: `The Onset'

Up to general Table of Contents

Back to Introduction