2.1 A typology of negation

Zeijlstra (Reference Zeijlstra2004) lays out a typology of negation based on the following detailed version of Jespersen's Cycle, which documents the diachronic change in negative forms across languages.Footnote ⁹ The phases in Jespersen's Cycle are shown to correlate with other syntactic properties including (i) whether the language exhibits double negation or negative concord, (ii) whether or not ‘true’ negative imperatives are permitted, and (iii) the interpretations that arise when the universal quantifier in subject position precedes negation. Our investigation of child data will not consider (iii), the interpretation of sentences with a universal quantifier and negation (e.g. Every boy didn't leave) and how they correlate with semantic or syntactic negation, since appropriate data from two-year-old children are not available. The important generalization that emerges from the typological study is that languages with a head form of negation are negative concord languages but not vice versa.

1. (3)

   • Phase I: Negation is only expressed by a single negative marker that is attached to the finite verb.

   • Phase II: The negative marker that is attached to the finite verb becomes phonologically too weak to express negation by itself and a second negative adverb becomes optionally available.

   • Phase III: Sentential negation is obligatorily expressed by the negative marker that is attached to the finite verb and the adverbial negative marker.

   • Phase IV: The negative adverb is the obligatory maker for negation and the use of the negative marker that is attached to the finite verb becomes optional.

   • Phase V: The negative adverb is the only available negative marker. The negative marker that is attached to the finite verb is no longer available.

   • Phase VI: The negative marker is available in two forms: it can appear either as a negative adverb or as a negative marker that is attached on the finite verb, though sometimes simultaneously.

The phases of Jespersen's Cycle that are most relevant to English and our investigation of child language are Phases V and VI. Adult standard English is a Phase VI language: it incorporates the negative adverb not as well as the head form of negation n't (which is taken to be equivalent to a negative marker attached to a finite verb). We will see that child development mirrors the change from Phase V to Phase VI in Jespersen's Cycle as children move from a grammar in which only a negative adverb is available to one which also incorporates a negative head.

2.2 Properties of Phase V and VI languages

Languages in which there is only one form of negation, a negative adverb, are Phase V languages. These languages include German, Dutch, Swedish, and Norwegian, which are V2 (or verb-second) languages, as well as Quebecois, Bavarian, and Yiddish, which are negative concord languages. The discussion here is limited to the V2 Phase V languages that are double negation languages.

In Phase V languages with double negation, the two forms of negation each carry negative force, and therefore cancel each other out. The example below from Dutch illustrates that double negation ends up being a positive statement (from Zeijlstra Reference Zeijlstra2004: 3).

1. (4) Jan heeft nietniemand gebeld. Dutch

   Jan has negn.body called

   ‘Jan didn't call nobody.’ = ‘Jan called somebody.’

Phase V languages all allow ‘true’ negative imperatives, as illustrated with German in (5) (Zeijlstra Reference Zeijlstra2004: 140–141). In languages with true negative imperatives, the imperative form is exactly the same in both affirmative and negative versions in (5). In those languages that do not allow true negative imperatives, the negative imperative cannot be formed by simply adding negation to the affirmative form but must be expressed using a different type of syntactic structure.

1. (5)

   1. (a) Mache es! German

      do it

      ‘Do it!’

   2. (b) Mache es nicht!

      do it neg

      ‘Don't do it!’

True negative imperatives are formed by verb movement to a Mood projection, as shown in (6) below. In later work (Zeijlstra, personal communication) suggests movement is to C rather than Mood (see also Zeijlstra Reference Zeijlstra2010). Either way, in languages in which negation is an adverb, the verb can freely raise across negation to the appropriate functional projection, since the negative adverb will not block movement (Travis Reference Travis1984).

1. (6)

Phase VI languages have different properties. The set of Phase VI languages comprises only varieties of English, both standard English and non-standard varieties with negative concord. Standard English stands as the exception to Zeijlstra's generalization that any language with a negative head will exhibit negative concord. As van Gelderen (Reference Gelderen2008) notes, however, there has long been a prescriptive tide against the use of negative concord, and this may have arrested any ‘natural’ move in this direction for standard English varieties. Nevertheless, there are varieties of both British English and African American dialects in which negative concord is the norm, or at least optional (see Poplack Reference Poplack2000, Green Reference Green2002, Iyeiri Reference Iyeiri2005).

In standard English, sentential negation is expressed either by the negative marker n't, a head that is attached to a finite auxiliary verb, or by the negative adverb not. In fact, the literature has debated whether not should be analysed as a negative adverb or as a head (see Zanuttini Reference Zanuttini1991, Reference Zanuttini, Baltin and Collins2001; Haegeman Reference Haegeman1995; Lindstad Reference Lindstad2006). One test used by Zeijlstra (Reference Zeijlstra2004) to determine its syntactic status is Merchant's why not test (Merchant Reference Merchant2001). On Merchant's analysis, the phrase why not is formed by phrasal adjunction, and therefore is only licit with negative adverbs. As one might expect, the why not construction is fine for Phase V languages like German, Dutch, Swedish, and it is also fine for English, suggesting that not is indeed an adverb. We follow Zeijlstra in taking not to be an adverb or an XP.

Turning to imperatives, Zeijlstra's generalization is that languages with a head form of negation ban true negative imperatives. If imperatives with do-support are considered to illustrate a different syntactic structure, then true to the generalization, modern-day English in all its dialectal varieties bans true negative imperatives, but whether the addition of do-support constitutes a different syntactic structure is debatable; it may be that do-support is an independent requirement.Footnote ¹⁰ In 15th century English, however, before the introduction of do-support and the negative marker n't, true negative imperatives of the form in (7) clearly did occur (Zeijlstra Reference Zeijlstra2004: 146). At this time, English still exhibited verb movement, so we can assume that the verb moved out of the vP across negation.

1. (7) Fear not! 15th century English

   ‘Don't fear!’

In later work, Zeijlstra (Reference Zeijlstra and Biberauer2008a) raises the possibility that English is actually a Phase V language with only a negative adverb. This is because standard English goes against the generalization that languages with a negative head are negative concord languages. The argument would be that all negative auxiliary verbs like don't, can't, etc. are unanalysed whole lexical items and therefore n't is not analysed as a separate head (Pullum & Wilson Reference Pullum and Wilson1977). It is difficult to make this argument for the lexical item doesn't, however, since this particular negative auxiliary verb realizes the agreement morpheme-internally to the word, splitting the word don't.Footnote ¹¹ It seems unlikely that this form, too, is simply learned as an unanalysed whole. We will therefore continue to assume that English is a Phase VI language with both a negative adverb and a head form of negation, n't, as in the earlier proposal.

2.3 The syntax of negation

The theory of negation proposed by Zeijlstra (Reference Zeijlstra2004, Reference Zeijlstra, Cate and Zeevat2007, Reference Zeijlstra and Biberauer2008a) is couched in the minimalist program (Chomsky Reference Chomsky1995, Reference Chomsky, Martin, Michaels and Uriagereka2000). Within minimalism, semantic features are distinguished from formal features. Semantic features have semantic content and do not trigger syntactic operations while formal features may or may not have semantic content. If they have semantic content, they are interpretable and can be interpreted at LF, but if they are without semantic content, they are uninterpretable and must be deleted during the derivation before reaching LF. The deletion takes place in the syntactic component once a checking relation with a corresponding interpretable feature is established, through the mechanism of Agree (Chomsky Reference Chomsky, Martin, Michaels and Uriagereka2000). Working within this framework, Zeijlstra assumes that negative markers can carry either semantic features or uninterpretable formal features.

In Dutch and German and other V2 languages with semantic negation, the negative adverb is adjoined as a vP adjunct, as illustrated in (8a) (but see also Lindstad Reference Lindstad2006 for Norwegian). The negative adverb carries a semantic feature [NEG], as shown in (8b), that is directly interpreted at LF.

1. (8)

   1. (a) [_vP niet [_vP ]] Dutch

   2. (b) Jan loopt niet_[NEG].

      Jan walks not

      ‘Jan doesn't walk.’

English also has a negative adverb, not, so one possibility is that this is adjoined to vP, just as in Dutch. However, since adult English also has a head form of negation that motivates a NegP projection, not could also be positioned in the specifier of the NegP projection. Either way, it can be assumed that not carries a semantic [NEG] feature.

Negation implicates formal features in English because English has the head form of negation n't in addition to its adverbial form. In a sentence like (9a), negation is licensed in the syntactic component.

1. (9)

   1. (a) John doesn't walk to school.

   2. (b) [_TP John does [_NegPOp⌝_[iNEG] [_Neg0n't_[uNEG]] [_vP walk to school]]]

As shown in (9b), the negative marker n't is positioned in the head of NegP and carries an uninterpretable negation feature [uNEG] that must be deleted before LF. This uninterpretable negation feature is checked off by a covert operator in SpecNegP that carries an interpretable negation feature ([iNEG]).

In negative concord varieties of English, there are two negative markers in the sentence representation, both carrying an uninterpretable negative feature ([uNEG]). These uninterpretable features are checked against the negative operator with the interpretable negation feature by multiple Agree (Ura Reference Ura1996; Hiraiwa Reference Hiraiwa2001, Reference Hiraiwa2005). Both uninterpretable negative features are thus eliminated before the derivation reaches the interface, yielding one semantic negation, and the interpretation that John did not do anything (but see Haegeman & Lohndal Reference Haegeman and Londahl2010). This is illustrated in (10).

1. (10)

   1. (a) John didn't do nothing.

   2. (b) [_TP John did [_NegPOp⌝_[iNEG] [_Neg0n't_[uNEG]] [_vP do nothing_[uNEG]]]

To summarize, cross-linguistic data reveal two types of languages. One set of languages has semantic negation, and sentential negation is expressed with a negative adverb. Another set of languages has a head form of negation. These languages have a NegP and negation is licensed in the syntactic component of the grammar. The two cross-linguistic options form the options realized by the Negative Concord Parameter (Zeijlstra Reference Zeijlstra2004, Reference Zeijlstra, Cate and Zeevat2007, Reference Zeijlstra and Biberauer2008a). The generalization is that a language with a negative head is always a negative concord language, but not vice versa. Finally, standard English is exceptional in that it has both semantic negation as well as syntactic negation, and, despite having a negative head, does not have negative concord.

In the next section, we turn to the predictions that Zeijlstra's theory makes for child English.

3.1 The Flexible Formal Feature Hypothesis

The Flexible Formal Feature Hypothesis (FFFH) is stated in Zeijlstra (Reference Zeijlstra, Cate and Zeevat2007) as follows:

1. (11)

   1. (a) Every feature [F] is first analysed as a semantic feature ([F]).

   2. (b) Only if there are doubling effects with respect to F in the language input, [F] has to be reanalyzed as a formal feature [i/uF].

The FFFH assumes that rich morphological structure in the positive input is what drives children to posit functional projections in their sentence representations. Returning to negation, the FFFH requires children to first assume that the language has semantic negation ([NEG)] and, only if there are doubling effects for negation in the input, will they reanalyze or add (in the case of English) negation as formal features ([iNEG] and [uNEG]) to their grammar. In the case of negation, the ‘doubling effects’ are sentences with negative concord since there are two negative markers that add up to one semantic negation. Negative concord sentences thus constitute the input that informs children that (at least) one of the negative markers carries an uninterpretable negation feature that must be checked in the syntactic component. This is what motivates a NegP functional projection. It is important for Zeijlstra's proposal that parameter values are not innately specified. Rather, they arise out of learnability requirements for simplicity.

Let us consider how the FFFH plays out in the grammars of children in different language communities. Children acquiring Phase V languages like Dutch or German, like all other children, initially hypothesize that their language has semantic negation. Since this is the correct option, and there is an appropriate lexical item in the positive input, these children converge speedily on the target grammar. What of children learning French or Italian? In principle, they too hypothesize semantic negation, but exposure to negative concord will soon cause them to reanalyze the local language as having formal features for negation, leading them to add a NegP projection to the grammar. Assuming that negative concord is abundant in the local language, adoption of formal features for negation could be very rapid, to the extent that the earlier, adverbial stage is never realized or detectable in the child's output (see Wexler's Reference Wexler1998 Very Early Parameter Setting).

Now consider children learning English. Once again, the FFFH should guide children to initially hypothesize semantic negation. The lexical items not (and/or no) are negative markers in the positive input that children can recruit for this purpose.Footnote ¹³ Children learning negative concord varieties of English can switch to formal features for negation once they recognize the ‘doubling’ evidence they need to converge on the adult grammar. Children learning standard English, however, lack exposure to the negative concord data that motivates formal features for negation. In our view, this means that children acquiring standard English refrain from projecting a NegP until alternative sources of linguistic data are forthcoming. This accounts for the protracted development of sentential negation observed in many children. Let us lay out the learning path for children acquiring standard English in more detail.

3.2 Early adverbial negation

The Flexible Formal Feature Hypothesis entreats children to initially posit semantic negation. Children therefore scout the adult input for sentences that contain one semantic negation and identify not and/or no as potential negative adverbs. The items not (and/or no) are assigned a semantic feature for negation [NEG], requiring negation to be interpreted in the semantic component. These items will be used in situations when an adult would use a negative auxiliary verb. Thus, children are predicted to produce sentences like This not fit or possibly This no fit instead of This doesn't fit.Footnote ¹⁴ The derivation would be as shown in (12). The verb and subject merge, the negative adverb takes vP as its complement, negating the proposition, after which the subject moves out to SpecTP (see Zeijlstra Reference Zeijlstra2004: 262).

1. (12) [_TP this_i [_T [_vPnot/no_[NEG] [_vP t_i fit]]]]

The fact that English-speaking children initially hypothesize only semantic negation means that their early English is a Phase V language, similar to many of the Germanic languages. This gives rise to some predictions about further properties of children's grammars at this stage of development.

The first prediction is that, in principle, a negative adverb does not block agreement/movement between the tense/agreement morpheme in T and the main verb. This can be illustrated in Phase V V2 languages like Swedish, in which the negative marker is a negative adverb. In V2 contexts, the verb moves over negation to V2 position, but in non-V2 contexts, such as when there is an initial adverb, inflection shows up on the main verb, as shown in (13a) (Vikner Reference Vikner1995). In English, this means that children could produce utterances like This not fits, in which the tense/agreement marker is realized on the main verb (see Bobaljik Reference Bobaljik1994).Footnote ¹⁵ This is illustrated in (13b).

1. (13)

Furthermore, if Bellugi's claim (see Section 1.1 above) that children's early uses of don't and can't at stage 2 are forms of negation rather than auxiliary verbs, these transitional forms can also be analysed as negative adverbs adjoined to vP, just like not. Quite possibly, children suspect that forms like don't and can't differ from not in their distribution (see Schütze Reference Schütze2010), but given that NegP has not yet been adopted into the grammar, they temporarily utilize these forms as vP adjuncts.Footnote ¹⁶ Children who take this path potentially could produce utterances like This don't fits, in which inflection associates with the main verb over the negative adverb don't. As a caveat, it has to be recognized, however, that adults do not permit utterances like It not fits, even though not is an adverb for adults also, so if such utterances occur in child grammars, we will need to consider how they are eliminated from the adult grammar.

A second prediction that follows from analyzing early English as a Phase V language is that children should take every negative marker as a form of semantic negation. Thus, early child language should permit double negation, as in sentences like John didn't tell nobody, where the intended meaning is that John told somebody. In practice, this may be difficult to test, as sentences with double negation are likely to be absent from two-year-old children's spontaneous speech for independent reasons. For one thing, double negation is infrequent even in adult languages that allow it (Zeijlstra Reference Zeijlstra2004). Conversely, English-speaking children should not use negative concord at this stage of development. There is nothing to prevent children from producing negative polarity items (henceforth NPIs) such as any in the adverbial stage, however, since the negative adverb can license the NPI.

A third prediction is that children will produce true negative imperatives. Affirmative imperatives such as Walk! will be produced by all children at all stages, but in the early development of negation, at the stage before do-support is acquired, the true negative imperative correlate Not run! is also predicted to occur, at least in some children's output. Given that there is no verb movement in English, and children know this early (see Wexler Reference Wexler, Lightfoot and Hornstein1994, Reference Wexler1998), the representation for a true negative imperative would be as in (14). An uninterpretable mood feature on the verb checks off its feature with an interpretable feature on the MoodP.Footnote ¹⁷

1. (14) [_MoodP[iMOOD] [_vPnot [_vP run_[uMOOD]]]]

It is worth considering what triggers the transition from the early adverbial negation stage of negation to the adult grammar. As we noted, for children learning standard English, there is no doubling evidence in the input to trigger addition of a NegP projection into the functional hierarchy, so some source of linguistic evidence must become available. Following Haegeman (Reference Haegeman1995), Zeijlstra (Reference Zeijlstra2004) observes that questions with raised negative auxiliaries are a potential catalyst to instigate change. Questions with raised negative auxiliaries must be formed by head movement of the negative auxiliary from I to C, which confirms the status of don't, can't, etc. as heads, and not as vP adjuncts. If children recognize these data in the positive input, they will potentially trigger reanalysis of the features on these lexical items.Footnote ¹⁸ Negative tags are another potential source of evidence, since these are analysed as reduced questions (e.g. Sailor Reference Sailor2009) and likely to be frequent in the input.Footnote ¹⁹

There may be more accessible sources of evidence to trigger adding the ‘negation as a head’ value for the Negative Concord Parameter. Negative auxiliary verbs, which are formed by adjunction of one head to another, should prove an abundant source of evidence. However, decomposing the morphology of negative auxiliary verbs may be problematic for children, since empirical evidence suggests that, at least for a while, don't seems to be treated as a negative adverb, rather than a negative auxiliary verb, at least by some children. The negative auxiliary doesn't may be more informative than most other auxiliary verb forms because the agreement morpheme is internal to the word. This provides overt evidence that the word is formed compositionally from several parts, including n't. We contend that doesn't, in particular, may be an important source of evidence for children to trigger change to the adult grammar.

Once children use negative auxiliary verbs productively (and doesn't, in particular) we can infer that negation is no longer just semantic negation in children's grammars. We can assume that they have added formal features for negation and incorporated a NegP phrase into their sentence representations.

3.3 Negation as a head

Once children have added n't as a lexical item into their lexicon, they require a NegP projection to license it in the syntactic component of the grammar. The head form of negation, n't, generates an uninterpretable [uNEG] feature in the head position of NegP. In order for the uninterpretable feature on n't to be checked off, the child needs to posit a covert negative operator with an interpretable [iNEG] feature in the SpecNeg position. The negation and auxiliary verb can then merge (Bobaljik Reference Bobaljik1994). The now adult-like representation for the sentence This doesn't fit is shown in (15).

1. (15) [_TP this_i [_T does [_NegPOp⌝_[iNEG] [_Neg0n't_[uNEG]] [_vP t_i fit]]]]

Once children adopt the second value of the Negative Concord Parameter, they have moved through one phase of Jespersen's Cycle, from a Phase V to a Phase VI grammar.

The consequence of adding another lexical entry for negation is that children should now be able to produce adult-like sentential negation, sentences like It doesn't fit. The question is whether they retain the form It not fits (if it was used in their earlier Phase V grammar) alongside It doesn't fit. The same question could be asked of negative imperatives. If children produce Not run! at the adverbial stage, do they simply add Don't run! once the head form of negation is in the lexicon, or do they eliminate the earlier form? Clearly the forms It not fits and Not run! are not grammatical in adult English, even though a negative adverb continues to be part of the adult grammar, so they must be eliminated from the grammar at some point. The question for acquisition, though, is when these forms are purged, and on what grounds.

One possibility is that a learnability principle, such as the Uniqueness Principle, kicks in to exclude the possibility that is not supported by the input (Wexler & Culicover Reference Wexler and Culicover1980, Pinker Reference Pinker1984). The Uniqueness Principle does not tolerate two syntactic forms being used for the same meaning, so it could force the non-adult optional form from children's grammar. Another possibility is that the acquisition of do-support has the effect of eliminating the non-adult form. This seems likely, although the mechanism that causes inflection to be blocked from appearing on the main verb is not clear. All we know is that sentences with the adverb not require do-support in the adult grammar (i.e. It does not fit).

The third prediction that follows from Zeijlstra's parameter is that once n't is added to the lexicon and a NegP functional projection is added to the grammar, negative concord is licensed, suggesting that children could, in principle, produce it. Of course, negative concord is not witnessed in the environmental input of children learning standard English, so this prediction may just be an ‘in principle’ prediction that fails to eventuate due to lack of linguistic evidence. At the same time, NPIs such as any are present in the input for children acquiring standard English, so nothing precludes children from producing them, either at the adverbial stage, or once they have adopted a head form of negation, since NPIs are licensed by either type of negative marker in adult English.

It is interesting to speculate about the timing of children's progression to adult-like sentential negation across dialects of English. Children acquiring negative concord dialects are exposed to the doubling of negative markers in the input, so, all things being equal, these children should adopt the ‘negation as a head’ value of the Negative Concord Parameter early. The caveat is that children acquiring any variety of English still have to acquire do-support before they can use the head form of negation to its full extent, so early use of negative concord could be hampered by this fact.

The fourth prediction is that a number of structures should become available to children as part of their syntactic repertoire once they have incorporated the head form of negation. The presence of negative questions with a raised negative auxiliary verb offers uncontroversial evidence that children have adopted n't as a negative head. In adult English, negative questions can be formed with a negative auxiliary verb or with not, but the unmarked way is to use a negative auxiliary verb. As noted, children can only produce this structure once they can move the negative auxiliary verb to the left periphery via head movement, as illustrated in (16).

1. (16)

   1. (a) What don't you like?

   2. (b) [_CP what_j [_C don't_i [_TP you [_T t_i [_vP like t_j]]]]]

Tag questions are another construction that should not be present at the earlier stage, when negation is restricted to an adverb. Tags can be of two types; invariant, as in (17a), or a form of the auxiliary verb that is dependent on the main clause, as in (17b). The dependent type has long been considered to be some kind of reduced question. An analysis by Sailor (Reference Sailor2009) proposes that dependent tags are derived by I-to-C movement with subsequent ellipsis of the VP. The movement of the negative auxiliary verb to C should mean that children are not able to produce negative tag questions until negation has been analysed as a head. The structure is illustrated in (17c).

1. (17)

   1. (a) You like linguistics, eh/huh/right?

   2. (b) You like linguistics, don't you?

   3. (c) You like linguistics [_CP [_C don't_i [_TP you t_i [_VPlike linguistics]]]]?

The predictions that follow from Zeijlstra's theory are summarized in Table 1.

Table 1

Summary of predictions.

4.1 Prediction 1

At the adverbial negation stage, children should, in principle, be able to produce negative sentences in which the main verb is inflected, such as This not fits or This don't fits.

The interaction of inflection and negation was studied longitudinally in four two-year-old children (Tesan Reference Tesan2005, Thornton & Tesan Reference Thornton and Tesan2007). The children visited our language acquisition laboratory every two weeks for about a year. Details of the children's participation are listed in Appendix B. The elicited production study targeted sentences with third person singular subjects, since these sentences provide the most information about children's knowledge of negation as it interacts with inflection.

As is usual for children in the optional infinitive stage of development, the four children all frequently omitted the morphology in their utterances with 3rd person subjects (see Rizzi Reference Rizzi1994; Wexler Reference Wexler, Lightfoot and Hornstein1994, Reference Wexler1998). Our focus, however, is those productions in which inflection was present in children's negative sentences. The important finding is that among those utterances with overt inflection were ones that conformed to Prediction 1. That is, three of the four children produced some number of utterances in which inflection was positioned on the main verb, such as He not fits in there. Utterances of this kind have been previously mentioned only in passing, and have usually been interpreted as performance errors due to their infrequent appearance.Footnote ²¹ The fourth child from the group produced fewer negative sentences, but nevertheless did produce one utterance of the form This don't fits in here.

The children's data also revealed another type of utterance that has not been well documented in the literature. The children produced some utterances with what we term ‘misplaced’ morphology, ones like He -s fit in there or negative ones like He -s not fit in there.Footnote ²² The utterances with ‘misplaced morphology’ are not informative for the present inquiry and are therefore excluded from the data we present, but for completeness, they are included in the summary of the data in Appendix C.Footnote ²³

The four children's data that are represented in the graphs in this paper include only sentences with 3rd person singular subjects, since these data reveal the position of the tense/agreement morpheme, if present. The first graph for each child shows the variety of utterances that express sentential negation in combination with a lexical verb, excluding the copula.Footnote ²⁴ Following this, we present a second graph for each child that collapses the data according to whether their negative sentences incorporate the lexical item not or n't. These graphs all show a sharp transition between use of not being exchanged for n't. The data show that when children add the head value of the Negative Concord Parameter, they cease to use the adverb, at least for their productions of lexical verbs with third person subjects.

Turning to the first of the four children, this child, Kristen, produced a variety of negative structures, among them sentences with negation and an inflected main verb. The utterances of the form This not fits were soon replaced by don't and then doesn't, as shown in Figure 1.

Figure 1

Kristen's negative sentences with 3rd person subjects by age in months.

The next child, Georgia, was a conservative learner and produced few negative sentences. This child did not produce not with inflected main verbs, but she produced one instance of don't used with an inflected main verb among her utterances. Therefore, although the support is limited for Prediction 1, her data are compatible with the prediction. Georgia's data are shown in Figure 2.

Figure 2

Georgia's negative sentences with 3rd person subjects by age in months.

The data for Caitlyn are shown in Figure 3. Caitlyn was the fastest to develop linguistically, rapidly becoming adult-like at just over two-and-a-half years of age, a good year earlier than Curtis, the child whose data follow, and considerably earlier than the majority of the children whose data are recorded in the CHILDES database. Caitlyn began using sentential negation by using not plus an inflected main verb, so her data support the proposal that children begin with a negative adverb.

Figure 3

Caitlyn's negative sentences with 3rd person subjects by age in months.

The final child, Curtis, exhibited the most chaotic path of development. There were many cases of utterances with not and an inflected main verb in his early productions. He vacillated between alternative structures for negative sentences for many months. When he finally acquired the form doesn't at age 3;2, adult-like negative sentences dominated the other forms of negation, which gradually disappeared, but not before he was about three-and-a-half years old. Curtis's development of negation is shown in Figure 4.

Figure 4

Curtis's negative sentences with 3rd person subjects by age in months.

Although each child showed a different profile when their negative sentences were charted, the following graphs from each of the children show that once the data are organized by negative marker, not or n't, each child's pattern of development looks remarkably similar, with not being prominent at first, and gradually being replaced by the much more frequent n't marker. The data for the four children are shown in Figures 5–8.

Figure 5

Negative markers n't versus not in Kirsten's grammar.

Figure 6

Negative markers n't versus not in Georgia's grammar.

Figure 7

Negative markers n't versus not in Caitlyn's grammar.

Figure 8

Negative markers n't versus not in Curtis's grammar.

As the graphs illustrate, once children adopted n't as a head and were using the negative auxiliary doesn't productively, utterances like This not fits in here disappeared from children's productions.Footnote ²⁵ There is considerable variation in the age that this takes place, but once it occurs, it is quite rapid. At this point, children have incorporated syntactic negation into their grammar, and added a NegP. The empirical data suggest that acquisition of do-support causes utterances with an inflected main verb to be eliminated, but exactly why this is so is not clear. We simply note here that the ungrammaticality of utterances with negation and an inflected main verb in adult English is also not well understood, especially given that such utterances are perfectly grammatical with the negative adverb never, as in It never fits.

Another potential reason why children might hesitate to use not once doesn't enters the grammar is because they are uncertain about its status as a negative adverb or a head. The positive evidence showing do-support in lieu of sentences with inflected main verbs such as He not fits may lead them to question their former analysis of not as a negative adverb. Consequently they may refrain from its use while they reevaluate its status. Although Zeijlstra (Reference Zeijlstra2004, Reference Zeijlstra and Biberauer2008a) assumes that not is an adverb in the adult grammar, and this is the assumption we follow in this paper, its status as adverb or head has been debated by theoretical linguists (e.g. Chomsky Reference Chomsky1995; Lindstad Reference Lindstad2006; van Gelderen Reference Gelderen2008, Reference Gelderen2011), so its status in child grammar is also worthy of consideration.

Although these issues cannot be resolved, the child data are nevertheless compatible with Prediction 1. The children's data show a rapid change from semantic negation as linguistic evidence causes children to adopt the syntactic negation value of the Negative Concord Parameter. Once children start to use doesn't productively, utterances like This not fits disappear rapidly from the grammar.

4.2 Prediction 2

Children in the adverbial negation stage should produce true negative imperatives. Recall Zeijlstra's (Reference Zeijlstra2004) proposal that Phase V languages, in which negation is a negative adverb, have true negative imperatives, whereas Phase VI languages do not. While there is some uncertainty as to whether negative imperatives with do-support such as Don't run! are true negative imperatives or not, it is clear that imperatives of the form Not run! fit the criterion – such examples show negation simply preceding the affirmative imperative form with no syntactic alternative required for the negative imperative. Evidence that children produce such forms would lend further support for the proposal that children's early grammars are limited to a negative adverb.

Prediction 2 is investigated in diary data from one child, Laura (Thornton, unpublished corpus). Diary data are the source of data here because there are a sizeable number of imperatives, including negative ones, produced at an early age in this corpus. The diary data are comprised of about 4500 utterances, collected from age 1;6 to 2;6.Footnote ²⁶ The first step is to illustrate that this child was in the early semantic negation stage under discussion. Her negative utterances with inflected main verbs at around age 2 shown in (18) confirm she was using not as a negative adverb (age format is shown in years; months.days).

1. (18)

   1. (a) Daddy, this is not works. (2;0.4)

   2. (b) This pencil not works. (2;0.4)

   3. (c) This is not goes here. (2;0.6)

   4. (d) This not goes here. (2;0.23)

   5. (e) This not fits. (2;0.25)

   6. (f) But he don't fits. (2;3.8)

About a month earlier, Laura produced negative imperatives, with the form Not + V, as illustrated in (19). The examples in (19d, h) are particularly clear examples of negative imperatives since they are followed directly by a non-negative imperative. Imperatives of this form were frequent at around age two, but then came to an abrupt halt, presumably replaced by Don't + V imperatives (in which don't could still be a transitional adverb, but we cannot tell from its surface form).

1. (19)

   1. (a) Not come in, Boomer! (1;11.0) [Boomer is the dog]

   2. (b) Not go here! (1;11.0)

   3. (c) Not sing in the car! (1;11.12)

   4. (d) Not do that! Walk! (1;11.12)

   5. (e) Not sing that song! (1;11.13)

   6. (f) Not play your flower pot! (1;11.27)

   7. (g) Not walk! Dance! (2;0.15)

These data support Prediction 2. True negative imperatives appear in the early grammar when the Negative Concord Parameter defaults to semantic negation. The true negative imperatives are short-lived, however, disappearing well before negation in its head form is added to the grammar. Bellugi (Reference Bellugi1967), for example, gives examples of Don't V imperatives in the Brown (Reference Brown1973) children's grammars at stage 2, so these forms with do-support are documented before children have acquired negation as a head. As we have noted, there is no way to tell whether children use don't in these imperatives as a transitional negative adverb, or whether it is being analysed as a negative auxiliary verb with the negative head n't. Our theory favors the former option, since children are not expected to analyse n't as a head until considerably later in the course of acquisition.

4.3 Prediction 3

In the semantic negation stage, children should permit double negation, and negative concord should not be licensed. Negative polarity items such as any occur in children's input data and children are expected to use them, as long as they are licensed by a negative adverb (i.e. not). Once children have syntactic negation, children's grammars should, in principle, license negative concord. In practice, children may not take up this option but either way, use of NPIs is expected to continue. Children learning negative concord dialects should adopt negative concord early, unless acquiring do-support is a complicating factor.

The first part of Prediction 3 states that at the early semantic negation stage, double negation should be permitted in the grammar. In such Phase V grammars, each instance of negation is semantic negation. Given the cognitive complexity of sentences with double negation, it is unlikely examples will show up in toddlers' speech, and indeed, no such examples have been discussed or observed in the literature. Bellugi (Reference Bellugi1967) explicitly notes that children do not use double negation in their spontaneous production and that it is a late development in English. Experiments will be required to test when children are able to interpret double negation.

Second, at the semantic negation stage, NPIs should be possible, if licensed by a negative adverb. To assess this, Laura's diary data, and Adam and Sarah's transcript data (Brown Reference Brown1973) were searched for NPI any (as well as anymore and anything). In Laura's diary data, the finding is that NPIs emerge early, simultaneously with inflected main verbs and true negative imperatives (as discussed for Prediction 1). As expected, in all of the examples below other than the imperative utterance in (20b), the NPI is licensed by no or not.Footnote ²⁷

1. (20)

   1. (a) Boomer's not hungry anymore. (2;0.3)

   2. (b) Don't touch anything! (2;0.7)

   3. (c) I not a dog anymore. (2;0.15)

   4. (d) I'm not scared anymore. (2;0.18)

   5. (e) There's no anymore. (2;0.24) [=no cookies left]

   6. (f) Girl's not sad anymore. (2;0.25)

According to Bellugi (Reference Bellugi1967), Adam and Sarah entered stage 3 and started to use a range of positive and negative auxiliaries productively at quite different ages: Adam was 3;2, and Sarah was 3;8. If we take the children's productive use of negative auxiliary verbs to mean that they have adopted syntactic negation, then we could expect that the children might use NPI any/anymore/anything before this age. Adam has a couple of examples at age 3, but Sarah does not use these items as NPIs in the early period. This is somewhat surprising, given Laura's early uses. Adam's examples are given in (21).

1. (21)

   1. (a) I can't wear that anymore. (3;0.2) file 24

   2. (b) He won't have anything to… (3;4.01) file 28

It follows from Zeijlstra's theory of negation that once children have added syntactic negation, sentences with negative concord can be licensed, at least ‘in principle’. The intriguing fact is that Bellugi (Reference Bellugi1967) reported that both Adam and Sarah used negative concord, despite the fact that their parents were not speakers of negative concord dialects.Footnote ²⁸ Crucially, the sentences with negative concord only occurred late in acquisition, after the children had made the transition to stage 3. Some examples from the children are given from Adam in (22) and Sarah in (23).

1. (22)

   1. (a) She is not having no picnic. (3;11) file 40

   2. (b) I just don't want nothing in there. (4;0) file 42

   3. (c) I don't want to share none of my books. (4;6) file 49

   4. (d) I'm not scared of nothing. (4;7) file 51

1. (23)

   1. (a) I didn't do nothing. (3;5) file 63

   2. (b) I didn't call him nothing. (3;8) file 72

   3. (c) Because nobody didn't broke it. (4;5) file 107

   4. (d) I don't think I can do this no more. (4;8) file 121

A closer look at Adam and Sarah's data shows that the children both used NPIs and negative concord after syntactic negation has been adopted, but their profiles differed. Recall, Bellugi (Reference Bellugi1967) identified Adam as moving to stage 3 at 3;2. A search of the children's transcripts (Brown Reference Brown1973) revealed that Adam began by using NPIs, but once he started to use negative concord, the NPIs temporarily dropped out of his grammar, appearing again at 4;4. Adam's data, arranged in three-monthly periods, are illustrated in Table 2.

Table 2

Adam's use of negative polarity items and negative concord by age.

Sarah, on the other hand, used both NPIs and negative concord alongside each other from age 3;5 through to 5;0, so these two structures appeared to be optional in her grammar. As shown in Table 3, Sarah's first use of negative concord is at 3;5 (file 63), so this is three months before it is predicted, according to Bellugi's (Reference Bellugi1967) criterion for moving into stage 3. In this context, it is worth noting that Clark & Clark (Reference Clark and Clark1977) estimate Sarah's transition to stage 3 to be at 3;5, rather than 3;8. If true, this would bring her data in line with the prediction. Notice that Sarah appears to have a strict negative concord dialect, like Czech (recall (1) above), as (23c) above shows that she allows an n-word in subject position in negative concord sentences. Sarah's data are shown in Table 3.

Table 3

Sarah's use of negative polarity items and negative concord by age.

Although Adam and Sarah take up the negative concord option made available once the grammar has instantiated a NegP functional projection, presumably other children choose not to. The child whose diary data we examined, Laura, for example, did not use negative concord productively. The key point is that Universal Grammar does not prohibit children acquiring any dialect of English from using it once they have added a NegP. This raises the question of how children acquiring standard English get rid of the unwanted negative concord utterances. Since the grammar licenses negative concord, it is quite likely that it does not dissipate until it is stamped out by prescriptive grammar in the school years.

All things being equal, the expectation is that children learning negative concord varieties of English should use negative concord earlier than Adam and Sarah, since they are exposed to the relevant doubling input with two negative markers. There are no sources of African American English available on the CHILDES database that can be used to investigate this prediction. Although there is a limited literature on negative concord in African American English (e.g. Green Reference Green2011), it does not address the age of emergence. There are data from children acquiring Belfast English (Henry Reference Henry1995, Wilson & Henry Reference Wilson and Henry1998) and Bristol English (Wells Reference Wells1981) in the CHILDES database, both of which are varieties of English in which negative concord is optional. Here, the Belfast corpus is used to investigate age of emergence.

According to Henry et al. (Reference Henry, McLaren, Wilson and Finlay1997), negative concord first appears very late in Belfast English children, with the first instance not appearing until 4;6. This late use is compared with children acquiring Bristol English, who reportedly have command of negative concord by 3;3. Henry et al. (Reference Henry, McLaren, Wilson and Finlay1997) propose that this difference cannot be attributed to frequency of input or to sociolinguistic variables, and suggest that it has its source in distributional facts about negative concord in the two languages. Our own searches of the Belfast English transcripts on CHILDES show earlier use of negative concord than reported by Henry et al. (Reference Henry, McLaren, Wilson and Finlay1997), with one child's first use of negative concord at 3;7, bringing it closer to the reports of children acquiring Bristol English.Footnote ²⁹ The Belfast corpus transcripts reveal that the children do not use negative concord very frequently, with uses of NPIs outnumbering instances of negative concord. These data from the Belfast children are summarized in Table 4.

Table 4

Negative polarity items and negative concord in the Belfast corpus.

The data from the Belfast children suggest that exposure to negative concord in the input does not speed acquisition of syntactic negation. Both the Brown children and the children in the Belfast corpus start to use negative concord at three-and-a-half years of age or later. It would be useful to see production data from an English variety in which negative concord is the norm, and not optional, as it is in Belfast English, to see whether this has any effect on the emergence of negative concord and the use of NPIs.

There is another interacting variable that may delay use of negative concord no matter what the dialect, or the frequency of its use. It is possible that even if children were to adopt a NegP phrase early due an abundance of negative concord input, in practice, its use could be severely curtailed until do-support has been acquired, typically a fairly late acquisition for English-speaking children. This issue remains for future research.

4.4 Prediction 4

At the semantic negation stage, children cannot produce constructions that require head movement of the negative auxiliary verb. As a consequence, children will not be able to form adult-like negative questions with an auxiliary verb raised to C such as What don't you like? or (dependent) negative tags such as You like linguistics, don't you?. These structures should become available once the child has added a head form of negation to their grammar.

To test Prediction 4, Adam and Sarah's transcripts (Brown Reference Brown1973) were searched to see when these structures emerge. We have seen that Adam entered stage 3 at age 3;2, so his negative questions with raised n't and tags should emerge after this point. The data for his negative questions are summarized in Table 5. The column labeled ‘Negative Aux raised to C’ shows Adam's adult-like questions, and the column showing ‘Negation in IP’ shows any alternative form of negative question in which the negation remained in the IP.

Table 5

Adam's negative question forms by age.

The transcript data show Adam's first negative questions at 3;4, after he entered Bellugi's stage 3. These first four questions are all why-questions, which, for some children, are optionally formed with an unraised Aux for a considerable period of time, even when other wh-words always appear with a raised Aux (see Thornton Reference Thornton2008). By age 3;5, Adam was producing some yes/no questions with a raised Aux, as well as some with negation remaining in the IP. The main point is that, as predicted, adult-like questions with the negative auxiliary verb raised to C only appear after he has adopted a head form of negation.

Turning to Sarah, there are no negative questions with a raised Aux before 3;8, the age Bellugi states that she entered stage 3. The first negative question with the Aux raised to C is at 4;0. Overall, Sarah asks few negative questions, however. Sarah's data are shown in Table 6.

Table 6

Sarah's negative question forms by age.

As Sarah's data illustrate, children acquiring English tend to ask few negative questions in their spontaneous speech (see Harris & Wexler Reference Harris, Wexler and Clahsen1996). When the situation calls for a negative question, however, children can oblige. An elicited production experiment by Guasti et al. (Reference Guasti Maria, Thornton and Wexler1996) elicited positive and negative questions from 10 four- and five-year-old children. In their positive questions, children raised the Aux to C 96% of the time in yes/no questions and in 88% of their wh-questions. The children's negative questions took a wide range of forms, however, as illustrated in (24). A raised negative auxiliary verb was used in only 1% of yes/no questions and in 23% of wh-questions (see Stromswold Reference Stromswold1990).

1. (24)

   1. (a) What don't you like? Adult form

   2. (b) What do you don't like? Aux-doubling structure

   3. (c) What don't you don't like? Neg/Aux-doubling structure

   4. (d) What you don't like? Noninverted structure

   5. (e) What do you not like? Not-structure

The most common alternative to the adult negative question structure for both wh-questions and yes/no questions was the Aux-doubling structure, shown in (24b) above. A few such Aux-doubling structures were also noted by Stromswold (Reference Stromswold1990) in Adam's questions. Occasionally the negative Aux doubled was doubled, as in (24c). The other structures used by the children in Guasti et al.'s (Reference Guasti Maria, Thornton and Wexler1996) experiment included questions with an unraised auxiliary verb (in contrast to their positive questions) and use of not rather than n't. The figures taken from Guasti et al. (Reference Guasti Maria, Thornton and Wexler1996) are shown in Table 7.

Table 7

Number and percentage of different negative question structures used by question type for children (N = 10).

The children's non-adult question structures all have in common the fact that negation remains in the IP domain, and is not raised by head movement to C. In all of these alternative structures that children produce, negation could be analysed as an adverb. In Aux-doubling structures such as What do you don't like? and in non-inversion questions such as What you don't like?, don't could be a transitional negative adverb, as discussed earlier. Children's questions with not are, of course, grammatical in the adult grammar, but they are not adults' preferred form, so they can be interpreted as a strategy to avoid raising a negative Aux to C. In sum, all of children's non-adult question structures with negation in the IP are expected before children add the new value of the Negative Concord Parameter.

There is one fact that does not line up with our predictions. Our investigation suggests that English-speaking children add a head negative marker to their grammar, somewhere between age three and age four, so by age four, children should produce adult-like negative questions. The children in the Guasti et al. (Reference Guasti Maria, Thornton and Wexler1996) experiment were aged four and five, so it is surprising that they are producing this assortment of non-adult structures considerably after syntactic negation should have been incorporated into their grammars. The children are apparently quite conservative, reluctant to implement the consequences of syntactic negation until they are certain.

Tag questions, too, are late in acquisition, as noted originally by Bellugi (Reference Bellugi1967) and predicted by our theory. Recent linguistic theory analyses negative tags as negative questions ‘reduced’ to the tag by VP ellipsis, so the prediction is that the timing should be very similar to negative questions, all things being equal. This is confirmed in Adam and Sarah's data in Tables 8 and 9.

Table 8

Number of negative questions and negative tags by age for Adam.

Table 9

Number of negative questions and negative tags by age for Sarah.

Other than two suspect examples well before age three, Adam's first negative dependent tag in the transcripts is at 3;5, exactly the same age that he starts to use negative questions with a raised negative Aux. Before he started to use dependent tags, according to Bellugi (Reference Bellugi1967), he often formed tags with ‘huh?’ Some examples at 4;0 are given in (25).

1. (25)

   1. (a) It's turned on red [?] # huh? (4;0) file 42

   2. (b) That's like you doing thread # huh? (4;0) file 42

   3. (c) I have to get black color # huh? (4;0) file 42

As Table 8 shows, neither negative questions nor negative tags appear often before about 4;5 so productivity of these two structures lags considerably after adoption of the head form of negation into the grammar.

Sarah also produced tags at about the same time as she produced negative questions, again confirming our prediction that these structures follow incorporating the syntactic negation value of the Negative Concord Parameter. Sarah adopts the head form of negation at around 3;8, and the data in the table suggest the two structures become productive between 4;1 and 4;4.

We can conclude that Prediction 4 is supported by Adam and Sarah's data. Syntactic structures that require movement of the negative auxiliary to C are not produced before the grammar has incorporated a head form of negation.