Negative sentences in children with specific language impairment

Introduction

Children with specific language impairment (SLI) have difficulty with aspects of grammatical knowledge which is revealed in both their comprehension and production of language. There is considerable debate about the extent of the impairment, and the source of the impairment itself. The present study focuses on English, and is the first investigation of negative sentences in children with SLI. This research focus is important, since in English, the syntax of negation is intertwined with knowledge of tense, an area known to present a challenge to children with SLI. The present study joins a growing number of investigations about how specific language impairment is manifested cross-linguistically (Adani, Forgiarini, Guasti, & van der Lely, 2014; Clahsen, Bartke, & Goellner, 1997; Contemori & Garraffa, 2010; Friedmann & Novogrodsky, 2011; Friedmann, Yachini, & Szterman, 2015; Grinstead et al., 2013; Grinstead, Cantú-Sánchez, & Flores-Ávalos, 2008; Jakubowicz, 2011; Jakubowicz & Nash, 2001; Jakubowicz, Nash, Rigaut, & Gérard, 2009; Leonard & Dromi, 1994; Leonard, Hansson, Nettelbladt, & Deevy, 2004; Paradis & Crago, 2009; Petinou & Terzi, 2009; Rice, Ruff Noll, & Grimm, 1997; Roberts & Leonard, 1997; Varlokosta et al., 2015, among others).

Children’s knowledge of morphosyntax has been the subject of intense investigation since the 1990s. It is now well established that in English, grammatical morphemes associated with tense (that is, finiteness) are frequently, but not always, omitted from the productions of English-speaking children with SLI (e.g., Joseph, Serratrice, & Conti-Ramsden, 2002; Leonard, 2014; Rice & Wexler, 1996; Rice, Wexler, & Cleave, 1995; Rice, Wexler, & Hershberger, 1998). Children with SLI are reported to frequently produce sentences such as He walk to school rather than the adult versions He walks to school or He walked to school. Auxiliary verbs are also frequently omitted. Productions like He walking to school are characteristic of this group of children, as are omissions of do in questions such as You walk to school? instead of Do you walk to school? (Rice & Blossom, 2012). As compared to tense-related morphemes, grammatical morphemes not associated with tense, such as the plural -s marker, the aspectual -ing morpheme and prepositions in and on are reported to be supplied more consistently in these children’s productions (Rice et al., 1998). There are a number of accounts of the difficulty with grammatical morphemes that children with SLI encounter. Three accounts are introduced here. The first two accounts, the ‘Extended Optional Infinitive’ account and the ‘Extended Representations’ account, view the language impairment of children with SLI as having its source in the grammatical component of the language faculty. On the third account, children’s difficulties are rooted in the way in which they process linguistic input. We will introduce a version of the Input-driven account cited in Leonard, Fey, Deevy, and Bredin-Oja (2015).

Much recent work on language development in children with SLI has its origins in Rice, Wexler and colleagues’ important observation that the omissions of grammatical morphemes associated with tense that are observed in children with SLI mirror the pattern that is seen in young typically-developing children (Rice et al., 1995; Rice et al., 1998). This stage of language development in typically-developing children was termed the ‘Optional Infinitive’ stage by Wexler (cf. Rizzi, 1993; Wexler, 1992, 1994, 1998). According to Wexler (1994, 1998), this developmental stage is in place until the relevant linguistic knowledge matures, at about 3 years of age. On this account, the Optional Infinitive stage is a product of the human biological endowment; it is not a product of learning. Subsequent research has shown that the growth curves of the various tense morphemes in children with SLI closely track the growth curves of typically-developing children, but with a significant time delay (cf. Rice, 2013; Rice et al., 1998). Because the developmental pattern of children with SLI is the same but stretched out over time, Rice et al. (1995) proposed that children with SLI experience an ‘Extended Optional Infinitive’ (EOI) stage, and that the cluster of morphemes related to finiteness are a clinical marker of children with SLI. Although the focus of Rice and colleagues’ research has been to document difficulties with the cluster of morphemes associated with finiteness over time, it is also noted that vocabulary growth and utterance length (MLU), which is an indicator of children’s progression towards adult clause structure, are delayed (Rice, 2013; Rice & Hoffman, 2015). As Rice (2013) points out, these delays in early and middle childhood leave more complex grammatical structures vulnerable.

Research by van der Lely concurs that morphemes related to tense are problematic for children with SLI, and that this is a deficit that is specific to the language faculty (van der Lely, 2005). However, van der Lely has argued that, at least for a subset of children with SLI, the language impairment extends beyond morphosyntax to implicate syntactic structures as well. Van der Lely has named this subtype of SLI ‘Grammatical-SLI ‘(G-SLI) (van der Lely, 2005; van der Lely & Pinker, 2014). Experimental findings have revealed impairments in adult-like use of tense/agreement in the verbal system but also, for example, in wh-questions, reversible passives and the assignment of pronominal reference (van der Lely, 1998, 2005; van der Lely, Jones, & Marshall, 2011). The overview presented in van der Lely and Pinker (2014) proposes that those aspects of phonology, morphology and syntax that pose difficulty for children with SLI involve what are termed ‘Extended representations’, which can be contrasted with ‘Basic representations’. Extended representations in the syntax involve ‘hierarchical, composed phrase structures, abstract rules, and movement, whereas Basic syntax involves local phrases, lexical retrieval, and interpretation based on lexical semantics. Extended morphology involves multi-morphemic, composed, regular forms, whereas Basic morphology involves the retrieval of whole words comprising irregular, derived, or high-frequency regular forms’ (van der Lely & Pinker, 2014, p. 593).

In contrast to the two accounts that view the impairment as having its source in the language faculty, Leonard has argued that children with SLI have a more general language deficit due to a limitation in processing capacity (e.g., Leonard, 1998, 2014). Leonard has proposed that grammatical morphemes of short duration (not just tense-related morphemes) are particularly challenging for children with SLI. The idea is that children with SLI perceive the grammatical morphemes in question, but require considerably more linguistic experience with them in order to establish a morphological paradigm that is accessible by the syntactic component.

The importance of linguistic experience, that is, input, in forming children’s grammatical knowledge is further highlighted in Leonard et al.’s (2015) recent proposal about the Optional Infinitive stage in children with SLI. Following language acquisition researchers working within the constructivist framework (e.g., Ambridge & Lieven, 2011), Leonard et al. (2015) argue that characteristics of the linguistic input to children are responsible for children’s omissions of finiteness morphemes. They argue that children with SLI interpret the linguistic input such that undue emphasis is given to local subject-verb combinations at the end of utterances, much like the computational model MOSAIC (cf. Freudenthal, Pine, & Gobet, 2009, 2010). Leonard et al. (2015) state, ‘We suspect that the protracted use of optional infinitives after length constraints have relaxed is due to a combination of more general syntactic comprehension difficulties and the children’s reliance on communicatively functional subject-verb sequences that had their origins in utterance-final portions of input sentences’ (Leonard et al., 2015, p. 816). This means that for input sentences such as Watch the dolphin jump or Did the dolphin jump, children with SLI tend to focus on the portion the dolphin jump. This reliance on the end of the utterance invites them to cut out this latter portion of the sentence and to use it as a grammatical unit expressing a proposition. Hence children’s erroneous productions like The dolphin jump with a missing 3rd person -s.

To date, research on the development of morphosyntax, and in particular, the acquisition of finiteness by English-speaking children with SLI, has focused on affirmative sentences and is now well documented (see Joseph et al., 2002; Leonard, 2014; Rice & Wexler, 1996; Rice et al., 1995; Rice et al., 1998). Children’s productions of negative sentences are of interest because sentential negation bears a close connection with tense in languages like English. Yet there has only been a single study on negation in children with SLI, to our knowledge. A study by Davies (2002) investigated teenagers’ comprehension of negative sentences using a grammaticality judgment task. Davies found that the teenage participants had difficulty making grammaticality judgments, particularly when the negative marker in the test sentences was inside a wh-question. The study presented in this article focuses on children’s production, rather than comprehension, and focuses on younger children. An investigation of how children with SLI produce negative sentences provides an opportunity to better understand the full extent of their productions during the documented Extended Optional Infinitive stage of development.

At first look, negation seems to represent an even greater learnability challenge for children with SLI, as compared to the use of tense in affirmative sentences. Indeed, we do find that children with SLI experience an extended delay in acquiring the range of negative markers (not, and the clitic form n’t) used by adult English-speakers. However, a theoretical proposal by Zanuttini (1996) suggests that sentential negation and tense are intimately linked, such that the negative marker n’t cannot be used without representation of tense information in the sentence. This suggests that negative sentences could evoke more consistent use of tense than affirmative sentences. We provide some initial data to show this is the case.

Finally, using elicited production we wanted to further evaluate the proposal that children with SLI pass through a stage of acquisition that is similar to an early stage in the language development of younger typically-developing children. As well as a comparison with a group of children matched by MLU in our experiment, the present study compares the negative sentences produced by children with SLI with those from two recent studies of the use tense in negative sentences by even younger 2- to 3-year-old English-speaking children (Thornton & Tesan, 2013; Thornton & Rombough, 2015). Both the non-adult and adult-like uses of tense by these younger children offer a yardstick for comparison with SLI children who are several years older. This comparison is a significant component of the present study. As in previous studies, children with SLI produce sentences with omissions of grammatical morphemes for tense, but nevertheless, we will see that they provide these morphemes far more often than in affirmative sentences. We will demonstrate that part of the difficulty for children in acquiring negative sentences lies in adding the n’t negative marker to the grammar and acquiring the language-specific rules that combine verbs, negative markers and tense morphemes into adult-like negative sentences.

The next section summarizes the main details about sentential negation in linguistic theory. This is followed by a review of previous empirical findings from studies of negation in typically-developing children. We then turn to our experimental study of sentential negation in children with SLI, returning in the discussion to the three accounts of optional infinitives in children with SLI.

Tense and agreement in negative sentences

Our investigation focuses on ‘sentential negation’ in the language of children with SLI. This term is used to refer to sentences in which a proposition is negated. Sentences expressing sentential negation can contain the negative marker not or a negative auxiliary verb or modal verb combined with the n’t negative marker, as illustrated in (1). Sentential negation differs from anaphoric negation, which is the use of negation in response to a previous speaker, as shown in (2). Anaphoric negation in English always uses no. Our study is restricted to the acquisition of sentential negation as in (1).

(1) John does not/doesn’t like broccoli

(2) Q: Does John like broccoli?

A: No.

Our study focuses on children’s production of negative sentences with 3rd person subject noun phrases (NPs) elicited in present tense contexts, yielding sentences like (3) and (4). Third-person subject NPs (such as John, he, the girl, etc.) are targeted because these NPs must show agreement with a verb that is marked by an overt 3rd-person -s morpheme (e.g., compare I walk, you walk, He walks). We will refer to this morpheme throughout as the 3SGS grammatical morpheme. This particular morpheme ‘fuses’ together information about present tense as well as 3rd person singular agreement. In this sense, the 3SGS morpheme differs from the past tense -ed form which remains unchanged regardless of the number assigned to the subject NP, i.e., 1st, 2nd or 3rd person (e.g., I walked, You walked, He walked). Thus, the past tense -ed morpheme could be said to only mark information about tense. In sum, targeting 3rd person subject NPs permits us to probe how children incorporate tense/agreement into their productions of negative sentences.

To review how negation works in English, let us compare the affirmative and negative sentences in (3) to (6).

(3) Susie eats broccoli

(4) Susie never {sometimes, often, always…} eats broccoli

(5) Susie does not (ever) eat_ broccoli

(6) Susie doesn’t (ever) eat_ broccoli

In affirmative sentences like (3), the verb carries the 3SGS morpheme that carries tense and agreement information. Adding an adverb (never, sometimes, often, always) does not change anything; the main verb still carries the 3SGS morpheme, as illustrated with the negative adverb never in (4). However, several changes occur in the adult grammar when sentential negation is expressed by not or by n’t. The negative marker not is also often analyzed as a negative adverb, and we will follow this analysis. However, this particular negative adverb behaves differently from never. When not is introduced, the verb becomes a ‘bare’ form and the 3SGS morpheme is supported by do, as shown in (5). Similarly, when the negative marker is n’t, as shown in (6), the 3SGS morpheme resides between the auxiliary verb do, and negation n’t. It follows that the negative auxiliary verb doesn’t is decomposed into three parts: do + s + n’t. According to linguistic theory, as we will sketch out in this section, significant modifications are required in the phrase structure representation for negative sentences that contain n’t like (6), as compared to affirmative sentences like (3) and sentences with adverbial negation, such as (4) and (5). An additional projection is needed in the phrase structure representation (Zeijlstra, 2004). This additional projection serves as a host for the n’t morpheme that is internal to doesn’t. We turn to see how these sentences are represented in linguistic theory next.

In the negative sentence representation in (6), the sentence level functional projection is Tense Phrase, TP. The tense and agreement features that end up being pronounced as the 3SGS morpheme originate in the position Tense (T); this position is known as the ‘head’ of the Tense Phrase. These features must associate with the adjacent main verb, so that the 3SGS morpheme can be pronounced as a suffix on the main verb. The grammatical representation for Susie eats broccoli before pronunciation is shown in (7). ¹

(7)

Children’s negative sentences

There are studies investigating aspects of negation in typically-developing children (e.g., L. Bloom, 1970; Cameron-Faulkner, Lieven, & Theakston, 2007; Déprez & Pierce, 1993; Drozd, 1995) but relatively few studies investigating the development of the syntactic structures that children assign to negative sentences. To this day, the most detailed study of children’s longitudinal development of negation remains Bellugi’s (1967) seminal work, which investigated the development of negation in the Harvard children, Adam, Eve and Sarah (Brown, 1973).

The Bellugi study concluded that children pass through two non-adult stages before they have productive use of the full range of negative auxiliary verbs that are normally used to express sentential negation in English (Bellugi, 1967; Klima & Bellugi, 1966). At the first non-adult stage, according to Bellugi, children position the negative markers no or not at the beginning or end of words or phrases (or even sentences). Examples include No sit there, Not a teddy bear, No fall!, Wear mitten no. This first non-adult stage is often characterized as a period during which negation is external to the sentence (but cf. Déprez & Pierce, 1993; Drozd, 1995).

At the second non-adult stage, children continue to use no or not as negative markers, but negation is positioned inside the sentence, yielding utterances like He no bite you, I not want apple. Because the negative markers in these productions are sentence-internal, this stage is sometimes referred to as the ‘medial negation’ stage (see e.g., Harris & Wexler, 1996). Children also start to use don’t and can’t at this point (e.g., I can’t catch you, I don’t sit on Cromer coffee) but their negative auxiliary verbs are limited to just these two. Because the full range of auxiliary verbs is not in place, Bellugi (1967) proposed that the early uses of the negative auxiliaries don’t and can’t are unanalyzed (‘fixed’) forms of negation, on a par with no and not. This differs from the adult grammar of English in which don’t and can’t are decomposed into an auxiliary verb, do or can, plus the clitic form of negation, n’t. Once children achieved greater facility with the auxiliary verb system, they were considered to be able to produce adult-like negative sentences.

A later study by Harris and Wexler (1996) adopted the Optional Infinitive (OI) theory advanced by Wexler to make predictions about the form of young children’s negative sentences (Wexler, 1992, 1994, 1998). Harris and Wexler predicted that the optionality of tense found in children’s affirmative utterances should extend to their negative ones. These researchers predicted that a child in the Optional Infinitive stage would produce negative sentences like those in (11a–c), but not produce ones like (11d).

(11) a. This marker not work

b. This marker don’t work

c. This marker doesn’t work

d. *This marker not works

When Tense is omitted from the syntax in negative sentences, children are predicted to use the negative marker not, yielding sentences like (11a). This form is produced because not does not require an auxiliary verb (carrying tense) to serve as its host. When children optionally incorporate tense into their sentence representation, Harris and Wexler predict that children will produce sentences like the ones in (11b) and (11c). They assume that both don’t and doesn’t are composed using the separate morphemes do and n’t, such that children’s sentences with don’t and doesn’t are both specified for tense. Recall, this was not Bellugi’s position. Bellugi (1967) proposed that don’t is initially a fixed form in children’s lexicon, rather than being decomposed, at the earliest stage of development. We will also take Bellugi’s position for reasons that will become clear. Finally, according to Harris and Wexler, children are not expected to combine not with an inflected main verb, as in sentence (11d). This prediction follows from their assumption that not is a head form of negation, not an adverb as we have assumed in the previous section (see (9)). If not is a head, as Harris and Wexler (1996) assume, then sentences like This marker not works in (11d) should not be produced because they violate a putative universal principle known as the Head Movement Constraint (cf. Travis, 1984).

The predictions that Harris and Wexler (1996) made concerning the optional use of tense in negative sentences were largely confirmed by a detailed investigation of the transcripts of the spontaneous speech of 10 children (age 1;06–4;01) whose data are recorded in the CHILDES database (MacWhinney, 2000). They report 125 utterances that contained doesn’t, don’t, not or no used in combination with a 3rd person subject. Sentences with doesn’t or don’t constituted 57% (71 tokens) of the children’s negative sentences. Another 43% (54 tokens) of children’s negative sentences utterances contained not or no. Only 5 of the children’s 54 negative sentences with not or no failed to conform to the expected pattern. These 5 utterances were like (11d), in which not was combined with an inflected main verb. ³ Because these apparent counter-examples only constituted around 10% of children’s productions, Harris and Wexler treated them as exceptional (i.e., noise). They concluded that the pattern of children’s linguistic behavior supported the predictions of the Optional Infinitive model (see Wexler, 1992, 1994, 1998).

With only 125 examples of children’s spontaneous use of negation to assess, however, it would be easy to reach erroneous conclusions. ⁴ To provide a more thorough assessment of children’s knowledge of negation, Thornton and Rombough (2015) used an elicited production task to gather a larger sample of children’s negative sentences. This investigation found that children produce a variety of non-adult negative sentences, including several kinds of utterances that are rarely attested in the transcripts of children’s spontaneous speech. In addition, the Thornton and Rombough study found that adult-like negative utterances with doesn’t appear relatively late in child language.

The study by Thornton and Rombough (2015) was conducted with 25 typically-developing children between 2;05 and 3;04 years (mean 2;11), so these children were in the age range for the Optional Infinitive stage. The 25 children produced over 550 full negative sentences. Thirteen of the children produced adult-like negative sentences with the auxiliary verb doesn’t and produced very few non-adult negative sentence forms. The other 12 children produced negative sentences with no, not, or don’t as the negative marker. ⁵ These children omitted the 3SGS marker in their negative sentences as in (12a), or used the 3SGS marker ‘high’ in a position before not, as in (12b), or with an inflected main verb as in (12c) or sometimes with doubling of the 3SGS morpheme as in (12d). These 12 children did not show the pattern expected by Harris and Wexler (1996) because they produced many examples like (12c) and they did not produce any adult-like sentences with the negative auxiliary verb doesn’t.

(12) a. Ernie not/don’t fit in the bus

b. Ernie’s not fit in the bus

c. Ernie not/don’t fits in the bus

d. Ernie’s not/don’t fits

Thornton and Rombough (2015) assumed productions like (12a) are typical of the Optional Infinitive stage in which children frequently omit the 3SGS morpheme that carries tense. They assume that children sometimes produce ‘high’ inflection, as in (12b), because they have not yet acquired do to support the 3SGS morpheme. In the absence of the host auxiliary verb used by adults, children simply cliticize the 3SGS marker leftward, onto the subject NP. To explain the ‘low’ inflection seen in examples like (12c), Thornton and Rombough drew on a theoretical proposal advanced by Zeijlstra (2004, 2008). According to Zeijlstra (2004, 2008), learnability considerations dictate that language learners should initially analyze negation as an adverb because this implicates a more economical syntactic representation. Children are expected to formulate more complex functional projections, such as the Negation Phrase (NegP) projection, only in response to decisive input. This account predicts that children acquiring English initially analyze not as a negative adverb. If English-speaking children analyze not as a negative adverb, then it is hardly surprising that they treat it like never, and permit it to be used with an inflected main verb in sentences like Ernie not fits in the bus. Finally, the examples with a doubled inflection marker such as Ernie’s not fits in the bus can be taken to reflect children’s efforts to work out where to pronounce the 3SGS morpheme.

By initially analyzing not as an adverb, children initially avoid the additional complexity involved in constructing a NegP (see example (9)). In languages with only a head form of negation, or in languages like English with both head and adverbial forms of negation, there should be abundant input to inform children that their grammars need to accommodate a NegP projection. Following an earlier account by Thornton and Tesan (2013), Thornton and Rombough (2015) propose that English-speaking children take some time to sort out the evidence that English has a head form of negation as well as an adverbial form. More specifically, children initially only produce negative sentences with the adverb not (or possibly no). Once they encounter sufficient positive evidence, children extend their grammars to incorporate the head form of negation, as attested by their productions of sentences that contain a range of negative auxiliary verbs (can’t, won’t, hasn’t, etc.).

According to Thornton and Tesan (2013) and Thornton and Rombough (2015), the auxiliary verb doesn’t is the most informative negative auxiliary verb. It provides children with the most salient information that English has a negative head. The auxiliary verb doesn’t is informative because the 3SGS morpheme is sandwiched word-internally, between do and n’t. The proposal is that this auxiliary verb is likely to alert children to the independent contributions of the three morphemes, do + 3SGS + n’t, which combine to form doesn’t. Only heads of syntactic categories could join together in this way. Provided that this knowledge about how morphemes combine is part of children’s inborn linguistic knowledge, the negative auxiliary verb doesn’t is evidence for them that the negative marker n’t is a head. Once children have incorporated the head form of negation into their grammars, they are expected to exclusively use doesn’t thereafter. At this point, children’s non-adult forms of negative sentences should rapidly disappear from their productions.

We now turn to the study of negative sentences in children with SLI. Given that previous studies of negative sentences in typically-developing children have been couched in the Optional Infinitive theory, we will use this to frame our research questions.

Research questions

The experimental study of negative sentences in children with SLI was designed to address four fundamental questions.

Experiment

Participants

Twenty-one children with SLI participated in the elicited production study of negative sentences. All of the children with SLI were drawn from monolingual English-speaking homes. The child participants with SLI had a mean age of 5;5 (ranging in age from 5;1 to 6;2) and included 15 boys and 6 girls. There were two control groups of children, an Age Equivalent group and a Language Equivalent group. The 21 Age Equivalent (AE) children had a mean age of 5;6 (ranging in age from 5;0 to 6;1). To ensure that the control groups were evenly matched, all of the children in the AE group were determined to fall within 1 month of age of at least one child in the SLI group. The 21 Language Equivalent (LE) children had a mean age of 3;6 (ranging in age from 3;0 to 4;3) and were matched by Mean Length of Utterance (MLU). These children had MLU values that were within ±1 SD of the mean expected for their age, based on Rice et al. (2010) norms, where M = 100 and SD = 15. In order to ensure equivalent MLU levels across the SLI and LE groups, the MLU level of each subject in the LE group was within 0.10 morphemes of at least one child in the SLI group. MLU was chosen as the criterion to match the SLI and Language Equivalent group because it is a reliable measure of syntactic complexity, at least up to MLU values of around 4.5 (Blake, Quartaro, & Onorati, 1993; Rice et al., 2010; Scarborough, Rescorla, Tager-Flusberg, Fowler, & Sudhalter, 1991).

The children with SLI were recruited from early intervention centers for students with language impairment, in Perth, Western Australia. ⁶ These children were diagnosed as language-impaired as preschoolers and were recommended by speech language pathologists for enrolment in the intervention school program. The children in the LE and AE groups were recruited from preschool centers in the Sydney metropolitan area or were recruited through the affiliated center’s paid participant subject pool at the university. Although some phonological and lexical differences exist in the English spoken across Australia, the syntax of the English spoken is generally reported to be homogeneous (Newbrook, 2001). For this reason, we assumed that there were no dialectal differences in the syntax of negation in the children with SLI and the children in the control groups.

For inclusion in the study, all of the children were required to have normal or above-normal intellectual functioning, as defined by a standardized score of 85 or above on the Kaufman Brief Intelligence Test (KBIT-2) (Kaufman & Kaufman, 2004). None of the children had a diagnosis for autism, intellectual, behavioral or social impairments. Children with multiple severe articulation problems were not included in the study, due to the ambiguities that speech errors may have posed in the analyses of the relevant grammatical morphemes in children’s negative sentences. Participants in the SLI group of children were required to have a passing score on the phonological probe of the Test of Early Grammatical Impairment (TEGI) (Rice & Wexler, 2001), which assesses final -t, -d, -s and -z. The children with SLI all had receptive and expressive language performance that fell below age expectations. The children in the SLI group all scored one or more standard deviations below the mean on the Clinical Evaluation of Language Fundamentals Preschool-2 (CELF-P2) (Wiig, Secord, & Semel, 2006). Children in the LE and AE groups were in the normal to normal-to-high range on the CELF-P2. Three children in the AE group were tested on the Clinical Evaluation of Language Fundamentals Screening Test (CELF-4) (Semel, Wiig, & Secord, 2006) as they had turned 6 years of age; their data are reported as raw scores that are compared with the criterion score of 12 for their age range.

Spontaneous language samples of at least 100 complete and intelligible utterances were collected from all children for calculation of MLU. The same collection of toys was used to evoke the spontaneous language sample for all groups of children. It consisted of a range of toy people and animals, a car for them to ride in, a picnic set with food, and various other props designed to capture the children’s interest and evoke comment. The data from the spontaneous language sample were entered into the SALT program for calculation of MLU values (Miller & Chapman, 1985). The descriptive information for the three groups of children is summarized in Table 1.

OPEN IN VIEWER

Table 1.

Descriptive information for children in the SLI group, the Language Equivalent group (LE) and the Age Equivalent (AE) groups. ^a

Group	N	Mean age (years;months)	MLU in morphemes	CELF-P2	KBIT-2	TEGI
SLI	21	5;5 (range 5;1–6;2)	3.48 (0.58)	71.09 (9.72)	98.28 (8.45)	20.82 (20.01)
LE	21	3;7 (range 3;0–4;3)	3.69 (0.60)	116.29 (9.28)	111 (7.31)	67.38 (13.98)
AE	21	5;6 (range 5;0–6;1)	4.87 (0.93)	110.82 (11.38)	106 (10.83)	84.66 (6.57)
				*17 (2.06)

a

Information is means (and SDs) for Mean Length of Utterance (MLU), the Clinical Evaluation of Language Fundamentals 2 (CELF-P2) or the Clinical Evaluation of Language Fundamentals 4 Screener (*CELF-4 Screener), Kaufman Brief Intelligence Test 2 (KBIT-2) and the Test of Early Grammatical Impairment (TEGI). Scores for the CELF-P2 and KBIT-2 are standard scores; scores for the CELF-4 screener are raw scores; scores on the TEGI are percent correct on the 3SGS subtest.

Results

The experimental task succeeded in eliciting a robust set of data from the children. The three groups of children, the SLI group, the Age Equivalent (AE) group and the MLU-matched Language Equivalent (LE) group each produced over 650 negative sentences for investigation. ¹⁰ The groups all produced full negative sentences, ones with null subjects, as well as some with verb phrase ellipsis. A summary of the breakdown of each type of children’s negative sentence is given in Table 2, to indicate how much data the elicitation task successfully evoked from children.

OPEN IN VIEWER

Table 2.

Percent and number of negative sentence types by group.

Negative productions	SLI (N = 21)	LE (N = 21)	AE (N = 21)
	% (No.)	% (No.)	% (No.)
Full sentence (e.g., It not work)	44.5 (384)	51.1 (425)	51.0 (342)
VP ellipsis (e.g., It doesn’t)	6.8 (59)	15.6 (130)	6.6 (44)
Null subjects (e.g., Not fit)	48.7 (420)	33.3 (277)	42.4 (284)
Total no.	863	832	670

Our main focus is the productions that were full sentences, so these will be discussed in detail. These full sentences are the most informative about children’s knowledge of tense and agreement. Full negative sentences can be divided into adult and non-adult negative sentences. The range of full sentence productions is summarized in Table 3. For each child, we first calculated what proportion each type of negative sentence (adult, bare verbs, non-agreeing don’t, modal can’t, etc.) contributed to their total productions. Then we calculated the overall mean percent of each type of negative sentence across the group of children.

OPEN IN VIEWER

Table 3.

Mean percent (and SD) and raw numbers of children’s full sentence productions by group.

Sentence type	SLI % (SD)	No.	LE % (SD)	No.	AE % (SD)	No.
Adult
Adult doesn’t (It doesn’t work)	17.4 (31.0)	56	80.8 (32.6)	368	88.2 (21.9)	310
Non-target
Bare verbs (It not work)	22.1 (32.0)	66	1.2 (5.5)	1	0	0
Non-agreeing don’t (It don’t work)	5.2 (13.4)	23	2.2 (10.1)	6	0.3 (1.4)	1
Non-target 3SGS + Verb (It not works, It’s not work)	28.5 (26.8)	132	3.2 (9.3)	9	0	0
Progressive -ing (It’s not working)	1.5 (3.8)	5	2.1 (5.8)	4	0	0
Past tense didn’t (It didn’t work)	4.9 (19.2)	15	2.4 (10.9)	11	0.2 (0.8)	1
Modal can’t (It can’t work)	13.5 (25.6)	47	8.1 (16.6)	26	10.6 (21.9)	27
Modal won’t (It won’t work)	6.9 (22.2)	40	0		0.8 (2.0)	3
Total number		384		425		342

In response to the present elicited production task, the most natural negative sentence is one with the negative auxiliary doesn’t. Many of the children produced adult-like sentences with this form, including some of the children with SLI. However, some children also produced alternative types of negative sentences, including sentences with modals, e.g., can’t or won’t, sentences with past tense, didn’t, and sentences with the progressive form of the verb, -ing. The children with SLI produced more of these alternatives than the children in the control groups did. These particular alternatives do not inform our research questions about how tense/agreement interact with negation, so they were not investigated further.

The focus of analysis, from this point forward, will be on children’s negative sentences that contained or could be contrasted with ones where adult English-speakers would have used the negative auxiliary verb doesn’t. We will call these ‘full negative sentences’. Once we removed children’s negative utterances that were not informative for our research questions because they contained modal verbs, past tense (didn’t), verbs in the progressive (-ing) and so on, the remaining data set of full negative sentences contained 277 negative sentences. The details of these negative sentences are summarized in Table 4. This shows a subset of the entire array of children’s negative sentences that is summarized in Table 3. To generate the summary statistics in Table 4, we first calculated the percentage of each subtype of 3SGS negative sentence for each child, followed by a calculation of the mean percentage for each sentence type, across children.

OPEN IN VIEWER

Table 4.

Mean percent (and SD) of 3SGS full negative sentence types by group.

Sentence type	SLI	No.	LE	No.	AE	No.
	%		%		%
	(SD)		(SD)		(SD)
Adult doesn’t (It doesn’t work)	22.8 (38.2)	56	88.7 (30.0)	368	99.7 (1.4)	310
Bare verbs (It not work)	25.0 (34.0)	66	4.8 (21.8)	1	0	0
Non-agreeing don’t (It don’t work)	9.7 (24.3)	23	2.6 (11.9)	6	0.3 (1.4)	1
Non-target 3SGS + Verb (It not works, It’s not work, It’s not works)	42.5 (37.2)	132	4.0 (11.1)	9	0	0
Total no.		277		384		311

The research findings show that children’s full negative sentences (i.e., ones corresponding to adult-like sentences with doesn’t) reveal a different profile for the children with SLI as compared to children in the two control groups. First and foremost, the children in the SLI group produced far fewer full negative sentences with the negative auxiliary doesn’t, as compared to children in the LE and AE control groups. The children in the SLI group produced a mean percentage of 22.8% adult-like negative sentences with doesn’t. The mean percentage of these sentences for children in the LE group was 88.7%, and it was 99.7% for children in the AE group. Further statistical analyses were conducted using the non-parametric Kruskal–Wallis test. This analysis confirmed highly significant between-group differences in the number of full negative sentences with doesn’t (N = 63, χ² = 25.095, df = 2, p < .001). It was also the case that for the 5-year-old children in the SLI group, there were as many as 25% of negative sentences with bare verbs expressing no tense information, whereas these constituted less than 5% both for children in the LE control group and for children in the AE control group. This between-group difference was highly significant (Kruskal–Wallis, N = 63, χ² = 20.24, df = 2, p < .001). The children with SLI also produced some negative sentences with the non-agreeing form don’t (It don’t fit), though not in large numbers. However, 42.5% of the full negative sentences produced by children in the SLI group contained a 3SGS morpheme that was non-adult in one position or another (e.g., It not works, It’s not work). In contrast to the SLI group, children in the LE control group produced only 4% non-adult utterances of this kind, and children in the AE group did not produce any non-adult utterances. Again, this between-group difference was highly significant (N = 63, χ² = 37.428, df = 2, p < .001). The data are summarized in Figure 1. This figure shows that the children with SLI clearly exhibited a different profile from the two control groups of children.

OPEN IN VIEWER

Figure 1.

Mean percent of children’s full negative sentences by group and by type (N = 63).

In view of the scarcity of non-adult negative sentences by children in both control groups, we can conclude that the task did not artificially induce non-adult responses. Nevertheless, children in the SLI group produced non-adult negative sentence with the 3SGS in a non-target position as much as 42.5% of the time. There were three basic variants of the non-adult 3SGS negative sentences produced by children in the SLI group. Raw numbers for these variants are summarized in Table 5. The numbers suggest that there are too many non-adult productions for these to be classified as performance errors. In some of children’s productions, the negative marker not or don’t was combined with an inflected main verb (i.e., It not works, It don’t works). The children with SLI also produced negative sentences in which the 3SGS appeared ‘high’ and preceded not, such as It’s not work. Finally, children sometimes doubled the 3SGS morpheme. Some examples of this were of the form It’s not fits, whereas others were of the form It doesn’t fits. ¹¹ These variants are shown in Table 5.

OPEN IN VIEWER

Table 5.

Number of full negative sentences with non-target 3SGS by group.

Non-target 3SGS negative sentences	SLI (N = 21)	LE (N = 21)	AE (N = 21)
	No.	No.	No.
It not work s	10	0	0
It don’t work s	17	6	0
It’ s not work	56	0	0
It’ s not work s	28	0	0
It doe s n’t work s	21	3	0
Total no.	132	9	0

Few non-target 3SGS productions of any kind surfaced in the productions of negative sentences by children in the LE control group, and non-adult negative sentences were absent entirely in the AE control group.

Having reviewed the full negative sentences produced by the three groups of children, we turn to the negative sentences with null subjects. The point of interest is that the null subject productions of children in the SLI group reveal a similar pattern to that observed in their full negative sentences. The child participants produced negative sentences in which null subjects appeared with bare verbs, such as Not work and Don’t work. They also produced sentences in which negation was combined with the 3SGS morpheme, such as Don’t works, Do not works, Does not works, Not works, No works. And, there were few adult-like negative sentences with null subjects (Doesn’t work, Does not work). The production of non-adult negative sentences with null subjects proved to be highly significant across the three groups (N = 63, χ² = 28.8, df = 2, p < .001). The breakdown of the total null subject negative sentences, where the target utterance would have contained the negative auxiliary doesn’t, is summarized in Table 6.

OPEN IN VIEWER

Table 6.

Mean percent (and number) of null subject negative sentences types by group.

Null subject negative sentences	SLI (N = 18 a )	LE (N = 18 a )	AE (N = 20 a )
	% (No.)	% (No.)	% (No.)
Adult doesn’t/does not	24.2 (75)	91.5 (235)	95.0 (257)
Non-agreeing don’t	6.8 (42)	4.1 (6)	5.0 (2)
Non-target 3SGS (e.g., Not works)	31.0 (105)	2.6 (5)	0
Bare verb (e.g., Not work)	38.0 (128)	1.8 (5)	0
Total no.	(350)	(250)	(259)

a

Some children did not produce null subject negative sentences.

Discussion

The first research question was designed to establish the range of negative sentences produced by 5-year-old children with SLI in an elicited production experiment. The experimental results revealed that children with SLI were able to produce a variety of negative sentences, although many of them were non-adult in form. The fact that children in the SLI group produced non-adult negative sentences is not surprising, given that negation and tense are closely connected in English. Thus the impairment that children with SLI demonstrate in producing affirmative sentences is likely to emerge in their negative sentences, to one degree or another. Moreover, the elicited production task used in the present study was specifically designed to probe children’s grammatical hypotheses about negation, a structure they might otherwise avoid, as attested by the scarcity of negative sentences in the transcripts of children’s spontaneous speech (cf. Harris & Wexler, 1996). Given that children with SLI were attempting to produce linguistic structures with which they may have had little experience, the likelihood of finding a considerable number of non-adult negative sentences was high.

In our view, children have two choices when they are confronted with a task that requires them to compute a linguistic structure that they find challenging. Children can opt to produce sentences that are pragmatically felicitous but ungrammatical, or they can sacrifice the pragmatics in favor of producing a structure that they know is grammatical. The negative sentences produced by children in the SLI group revealed that they adopted both strategies. Some children chose to ignore pragmatic norms, in favor of constructing sentences that were grammatical. Examples of children adopting this strategy included negative sentences with past tense didn’t, or with the verb in the progressive (-ing) or with a modal (e.g., won’t), even though the lead-in sentence in the experiment was designed to elicit doesn’t. These forms were not entirely felicitous in the experimental context. On the other hand, some children’s negative sentences contained a verb in the present tense, as appropriate for the context, but were non-adult in the sense that the 3SGS morpheme may have been dropped or appeared in a position that is not licensed for adult speakers.

Before we turn to our discussion of the remaining research questions, we deviate temporarily to ask whether SLI children’s non-target sentences truly represent their grammatical knowledge, or whether children’s non-adult productions could have occurred due to effects of the input.

Negative sentences with null subjects

The fourth research question asked whether children with SLI consistently produce full negative sentences, or whether they tend to drop the subject NP. The data show that children with SLI produce a variety of sentences with null subjects that nevertheless have the 3SGS morpheme in the sentence – sentences like Don’t works, Do not works, Not works, No works. A comparison of the SLI children’s full negative sentences and their negative sentences with null subjects is summarized in Figure 2. The main difference is that in the negative sentences with null subjects, the mean percent of non-target 3SGS sentences dropped by about 10%. This is not surprising, since one type of non-target 3SGS production is not available with a null subject. That is, the 3SGS morpheme could not be positioned ‘high’ in negative sentences with null subjects, since there is no subject NP to its left to act as a host. Therefore, null subject negative sentences such as ’s not work are not predicted to occur (and in fact, do not occur). The unavailability of this ‘high inflection’ option may have increased the number of null subject productions with bare verbs.

OPEN IN VIEWER

Figure 2.

Mean percent of full negative sentences produced by children with SLI (N = 21) and their null subject 3SGS negative sentences (N = 18^a).

The fact that children with SLI produce null subject utterances with 3SGS morphology is important because it indicates that, even in negative sentences with null subjects, children are generating syntactic representations for a full sentence, as illustrated in (16). That is, in negative sentences like Not works, the presence of the 3SGS morpheme requires agreement with the ‘empty’ 3rd person subject NP, as indicated by the ‘3SG AGR’ feature which is shared by the null subject NP (indicated by the triangle) and Tense (T). We conclude, therefore, that when children produced a sentence with a null subject and an inflected main verb, the syntactic representation nevertheless contains a Tense Phrase (TP) syntactic category; the only difference is that the subject was not pronounced. ¹⁷

(16)

If children with SLI generate complete syntactic representations for their null subject productions with 3SGS morphology, then why is the subject NP frequently omitted in their output? A comparison of the frequency of null subjects in the SLI group and in the 2- to 3-year-olds that participated in the Thornton and Rombough (2015) study shows that although both groups of children reveal a similar breakdown of negative sentence forms (both producing non-target 3SGS negative sentences), the proportion of null subject negative sentences differs across the two groups. Of the total number of negative sentences produced by the SLI group of children, 48.7% (420/863) had null subjects. The 2- to 3-year old group of children, by contrast, produced only 16.8% null subject negative sentences (161/957). If the 2- to 3-year-old group is taken as a baseline for the proportion of null subjects expected in a group of children who are in the Optional Infinitive stage, then it is clear that the children with SLI are using null subjects over and above this level. This finding suggests that the null subjects may emerge more often in the productions of children with SLI due to more limited processing resources (P. Bloom, 1990). In this respect, we agree with Leonard (1998, 2014) that processing limitations can be a factor for children with SLI.

This ends our discussion of the research questions. There are two further findings worth comment. The first was that the children with SLI manifested a distinct profile in linguistic behavior, one that was not mirrored by the 3-year-old children in the Language Equivalent MLU-matched control group. The second finding was that the proportion of use of the 3SGS morpheme differed widely in assessments of children’s affirmative and negative sentences. We address these two issues next.

Profile of delay

Our experimental findings showed that the children with SLI produced non-adult negative sentence forms not seen in the LE typically-developing children. Therefore, it is not the case that the language of children with SLI exhibits the same language characteristics as younger children with the same MLU. The data show what Leonard (2014) terms an ‘uneven profile’ of delay. The children with SLI show a lower rate of use of the negative auxiliary verb doesn’t than would be predicted by their MLU. The LE group of children who are matched by MLU use doesn’t at a mean percentage rate of 80.8% while the children with SLI were using doesn’t in only 17.4% of their overall productions.

The grammatical profile with respect to use of the 3SGS morpheme and form of negative marker is similar to the younger 2- to 3-year-old children (mean age: 2;11) who participated in the Thornton and Rombough (2015) study using the same methodology, however. These children produced similar sentence types. Figure 3 compares the attempts at doesn’t by the two groups of children. Both groups of children produced utterances with not and a bare verb, don’t and productions with a non-target 3SGS morpheme.

OPEN IN VIEWER

Figure 3.

Mean percent of full negative sentences for children with SLI (N = 21) and 2- to 3-year-old children in Thornton and Rombough (2015) by type (N = 25).

The difference between the SLI group and the 2- to 3-year-old group is that the 2- to 3-year-old children in the Thornton and Rombough (2015) study produced a higher mean proportion of adult-like negative sentences with doesn’t (58%) as compared to the majority of children with SLI (23%). This difference is simply because 13 of the 25 2- to 3-year-old children had already progressed to negating with doesn’t, and would be placed in an Advanced group by the criterion we have established. The uneven profile of delay highlights the challenge of mastering the adult system of negation, in particular acquiring the negative auxiliary verb doesn’t. Acquisition of this form signals the addition of NegP into the sentence structure, as instantiated by use of the morpheme n’t as well as obligatory use of tense/agreement in these sentences.

Elevated use of the 3SGS morpheme in negative sentences

The next issue involves a comparison between children’s proportion of tense/agreement in their affirmative versus their negative sentences. Overall, the children with SLI produced a 3SGS morpheme in 65.3% of their negative sentences. ¹⁸ This percentage of negative sentences with a 3SGS tense/agreement morpheme is considerably higher than the proportion of such morphemes in children’s affirmative sentences, as measured by their performance on the affirmative TEGI probe (Rice & Wexler, 2001) and the spontaneous MLU language sample. In the 3SGS subtest of the TEGI, children with SLI produced the 3SGS tense/agreement morpheme only 24.9% of the time. The spontaneous language sample elicited a higher proportion of 3SGS morphemes at 37.4%, although some children produced few sentences with a 3rd person subject and 3 children produced no relevant examples. For this reason, we used the TEGI 3SGS subtest as our measure for comparison. ¹⁹ Because we cannot compare children’s affirmative and negative sentences directly using the same methodology, the comparison is suggestive only. It will be important to compare provision of tense morphemes in affirmative and negative sentences using similar methodology in future research.

The comparison data from the TEGI and elicited production of negative sentences are summarized in Table 7.

OPEN IN VIEWER

Table 7.

Mean percent (and SD) of 3SGS morpheme in affirmative TEGI 3SGS subtest and in negative sentences (target and non-target uses of 3SGS) for SLI group.

Source of 3SGS	SLI all (N = 21)	SLI Less Advanced group (N = 16)	SLI Advanced group (N = 5)
	% (SD)	% (SD)	% (SD)
TEGI 3SGS subtest	24.9 (28.5)	18.7 (25.1)	44.6 (32.5)
Negative sentences	65.3 (37.8)	55.4 (38.1)	97.1 (6.3)

As Table 7 shows, there is a large discrepancy in the use of the 3SGS morpheme in children’s affirmative and negative sentences. For the Less Advanced SLI group, the 3SGS morpheme was used three times more often in negative sentences. The children in the Advanced group were almost at ceiling in their use of the 3SGS morpheme in negative sentences (97.1%) whereas the 3SGS appeared only 44.6% of the time in the TEGI subtest. The finding that children use the 3SGS morpheme more frequently in negative sentences may be counterintuitive, as one might expect the added complexity of negation in sentences to depress the level of tense/agreement, not to elevate it (L. Bloom, 1970).

The incorporation of negative markers into the sentence representation apparently encourages the child to build additional syntactic structure, including a TP projection that carries the 3SGS feature information. As noted, according to Zanuttini (1996), a NegP in the phrase structure requires the projection of a functional category for Tense Phrase. This suggests that, once children build a NegP phrase to host the head form of negation, n’t, they should invariably use tense in their negative sentences. Although this is true for the children in the Advanced group, as predicted, children in the Less Advanced group who just used the adverb form of negation, not or don’t, also used inflection at a much higher rate in their negative sentences. The data from the experiment suggest that any form of negation in the phrase structure has the effect of elevating children’s proportion of use of the 3SGS morpheme in negative sentences. Note that in affirmative sentences, there is no requirement that a functional projection for TP be projected, at least, not for children in the (Extended) Optional Infinitive stage, according to Wexler (1994, 1998). Therefore even though children use a finiteness marker quite consistently in negative sentences, from a theoretical perspective, there is no expectation that their affirmative sentences will demonstrate the same consistent use of the 3SGS morpheme.

Conclusion

The present study is the first investigation of children’s productions of sentential negation in English-speaking children with SLI. An elicited production task was used to elicit sentences from children with SLI, and from two control groups – Age-Equivalent (AE) children and Language-Equivalent children (LE, as matched by MLU). All of these children performed a task that required them to form the negation of an affirmative proposition. The most natural negative sentence (i.e., the preferred adult form) incorporates the negative auxiliary verb doesn’t. By targeting this structure, a robust sample of over 2300 negative productions was obtained from the three groups of children, including more than 850 negative productions from children with SLI. This permitted us to execute an in-depth study of the more than 800 full sentence forms and null subject productions used by the children.

The study found that, by 5 years of age, children with SLI were able to express sentential negation, but many children with SLI produced non-adult forms. The same pattern of performance was not seen in the AE control group or in the LE control group. The productions of children in the two control groups were by and large adult-like. In short, the elicited production task was successful in evoking the target sentences, and did not induce experimental artifacts.

Children’s early negative sentences were found to only use the adverbial form of negation, not. Among the non-target negative sentences elicited from the group of children with SLI were ones with bare verb forms (It not work). These children also produced negative sentences with the 3rd person morpheme either positioned ‘high’ (It’s not work) or positioned ‘low’ (It not works). Sometimes the 3rd person morpheme was repeated (It’s not works) in the sentence structure. Individual children tended to either put the 3SGS morpheme ‘high’ or ‘low’, as well as producing ones with a bare verb (It not work). However, there were also a few children with SLI, 5 of the 21 children, who used the adult form, doesn’t, in their negative sentences. These children had added the head form of negation, n’t, to their grammars, and were able to combine it with do and the 3SGS morpheme into this complex form. Interestingly, these children with SLI did not also produce negative sentences with the non-adult alternations between finite and non-finite forms. The findings clearly demonstrate that once children with SLI are able to produce negative sentences with doesn’t, the non-adult forms of negative sentences are eliminated from the grammars of these children. Thereafter, these children with SLI produce the same negative sentences as typically-developing children and adults. Although the non-target negative sentences with the 3SGS morpheme ‘high’ or ‘low’ appeared only occasionally in the sentences produced by children in the LE control group, and never appeared in the sentences produced by children in the AE control group, these same forms were observed in greater quantities in a previous experimental study of younger 2- to 3-year-old children (Thornton & Rombough, 2015). So, overall our findings reveal that children with SLI are delayed by about 3 years in acquiring simple negative sentences with the adult form.

It is worth considering how our experimental findings fit with the theories of SLI that were sketched in the introduction. All three proposals, the Extended Optional Infinitive account (Rice, 2013; Rice & Wexler, 1996; Rice et al., 1995; Rice et al., 1998), the Extended Representations account (van der Lely & Pinker, 2014) and the Input-driven account based on MOSAIC (Freudenthal et al., 2009, 2010; Leonard et al., 2015), were designed to explain children’s omissions of tense-related morphemes in affirmative sentences. All three accounts correctly predict that finiteness morphemes are not always present in negative sentences, but they do not predict the appearance of the 3SGS morpheme in non-target positions, and an extended period of adverbial negation as seen by a delay in use of the auxiliary verb do and the head form of negation n’t in the form doesn’t. The three accounts all need to be augmented in order to handle the facts about sentential negation.

On the Input account of optional infinitives, children focus on utterance-final subject-verb sequences, and harness these as propositions (Freudenthal et al., 2009, 2010; Leonard et al., 2015). Thus Did the dolphin jump? makes available The dolphin jump as a proposition. Turning to negative questions, the most frequent form uses a negative auxiliary verb in inverted position (Didn’t the dolphin jump? Doesn’t the dolphin jump?). Since negation is not in utterance-final position, this account says nothing about children’s negative sentence forms. Less frequent questions with not, such as Did the dolphin not jump? could potentially encourage children to produce utterances like The dolphin not jump, which do, indeed, occur. However, the relevant negative questions are infrequent in colloquial speech, so it seems unlikely that they are responsible for children’s use of not. Other structures using not such as subjunctives like I demand that he not smoke are even less frequent in the input to children, and less likely to be the cause of children’s sole use of not at the early stages of acquisition. Crucially, far more frequent sentences negated with doesn’t, such as The dolphin doesn’t jump, appear to have no influence on the form of children’s early productions. Thus, the Input account does not lend itself easily to an account of children’s negative sentences in the optional infinitive stage of grammatical development.

Generative accounts of language acquisition appeal to adverbial negation as being the initial default option made available by Universal Grammar (Thornton & Tesan, 2013; Zeijlstra, 2004). The head form of negation, n’t, must be added later, once instantiated by the language-specific (i.e., English) input. As generative linguistic accounts, both the Extended Optional Infinitive account and the Extended Representation account of SLI can easily accommodate this assumption. The question for these accounts is why children with SLI take so long to transition from a grammar that has only adverbial negation to one that also incorporates the clitic (head) form of negation (cf. Thornton & Rombough, 2015; Thornton & Tesan, 2013).

This additional delay in grammatical development invites the inference that the language impairment in SLI is not confined to tense, but extends to negation as well. However, it may be that the extended period of adverbial negation is parasitic on the impairment with tense. The triggering linguistic evidence for the head form of negation that we have proposed (i.e., doesn’t) resides in the auxiliary verb system in English, which is, of course, an expression of finiteness. Therefore, it is likely that children with SLI will encounter greater difficulty decomposing the negative form doesn’t into its component morphemes than typically-developing children, and until they are aware that it is a multi-morphemic form, they will not be able to add the clitic form of negation into the grammar. This interpretation, again, is compatible with either of the grammatical accounts, but is particularly pertinent for van der Lely and Pinker’s (2014) Extended Representations account. Since doesn’t is a complex form composed of three different morphemes, it involves Extended morphology, and there is a clear prediction that children with SLI would find analysis of this data point challenging.

There is another finding, admittedly somewhat speculative, that does not easily fall out of any of the proposals about optional infinitives. This is the difference that was found in the proportion of affirmative versus negative sentences with the 3SGS morpheme present. Recall that children with SLI produced the 3SGS morpheme in 65% of their negative sentences (including the non-target 3SGS productions such as It not works, It’s not work) as compared with 24.9% for affirmatives in the TEGI and 37.4% in their spontaneous production of affirmatives.

We have already mentioned one way to account for the elevated rate of the 3SGS morpheme in negative sentences produced by children with SLI which can also be used alongside either the Extended Optional Infinitives account or the Extended Representations account. As we saw, Zanuttini (1996) contends that a Negation Phrase (NegP) requires a Tense Phrase (TP) in the sentence representation. On this account, the grammatical contingency between TP and NegP forces the construction of a TP functional projection for the 3SGS morpheme in the syntactic representation of negative sentences. In affirmative sentences, on the other hand, there is no such contingency between functional projections, so children may not represent TP in the phrase structure (cf. Harris & Wexler, 1996). Therefore, the prediction is that tense-related morphemes will be optional in affirmatives.

There is another interpretation of the finding that children use the 3SGS morpheme more often with negative than affirmative sentences that is consistent with the Extended Representations account of van der Lely and Pinker (2014). Suppose that children with SLI learn doesn’t as a ‘holistic’ item and store it as such in their mental lexicon. If doesn’t is stored as a lexical item in the lexicon, it could be retrieved and inserted directly into the syntactic representation without the need to compose the three contributing morphemes (or features) in the sentence derivation. Analyzing doesn’t as a Basic morphological unit, rather than an Extended representation, would provide children with SLI a foolproof route to producing adult-like negative sentences. This route is not possible in affirmative sentences, because affirmative sentences require the composition of a main verb with the 3SGS morpheme by rule. Therefore, pursing this Basic morphology account of doesn’t affords us with another way to explain why children with SLI produce more negative sentences than affirmative sentences with (what appears to be) 3SGS morphology. However, if children were to analyze doesn’t as a whole, then it cannot serve as the trigger for the head form of negation for these children. As a consequence, its introduction into the grammar would not cause the disappearance of the non-target negative sentences with ‘high’ and ‘low’ inflection, so this proposal fails to fit with our observations about children’s longitudinal development.

One question for the future is whether negative sentences in the past tense also evoke non-target uses of the past tense affix in ‘high’ and ‘low’ positions in sentences such as Minnie Mouse’d not jump or Minnie Mouse not jumped. But the most important area for future research is the nexus between affirmative and negative sentences. First, it will be useful to elicit both affirmative and negative sentences in similar experimental contexts, in order to facilitate a direct comparison of the rate of 3SGS usage in these two linguistic contexts. From a clinical perspective, it is too early to know whether or not eliciting negative sentences with doesn’t could encourage higher proportions of use of tense morphemes overall in the grammars of children with SLI. Following Zanuttini’s (1996) proposal, we have suggested that once children acquire the head form of negation, they must obligatorily project a Tense Phrase (TP) as well as a NegP in the hierarchical phrase structure. This means that they will unfailingly use a morpheme for tense in negative sentences, presumably the 3SGS inside doesn’t. However, there is no grammatical principle that forces children to project a TP in affirmative sentences. What is not clear at present is whether obligatory use of tense in negative sentences could have the added value of accelerating use of tense in affirmative sentences. This will require a longitudinal study. But, if this were the case, the clinical benefit could be significant.