The Status of Nonagreeing Don't and Theories of Root Infinitives

Abstract

This paper examines two issues concerning nonagreeing don't in child English, e.g., He don't fit. (1) Do children know that don't consists of auxiliary do plus sentential negation, or do they misanalyze it simply as negation? I argue that the former claim yields both empirical (distributional) and conceptual advantages, while the latter does not explain what it was designed to explain. (2) If it is not misanalyzed, why does this form fail to agree? I consider two accounts that assume it is part of the Root Infinitive stage—one based on a misset parameter involving how agreement is spelled out (Guasti & Rizzi 2002), and the other based on underspecification of Infl features in syntax (my alternative proposal)—and explore their divergent predictions. I argue that the underspecification approach requires fewer stipulations about how children differ from adults, particularly for capturing do-omissions in “medial neg” environments.

Notes

¹See note 7 for the status of this usage in the input to the children being studied here.

²The terms Root Infinitive and Optional Infinitive are both in common use to refer to this phenomenon. Each evokes an implication on which I do not wish to take a stand, namely that the non-adultlike nonfiniteness characteristics of this stage are limited to root contexts or optional, respectively (though in the latter case, Wexler in many works makes clear that no such claim is intended). Both terms also imply that infinitives are criterial, which is clearly false in the present context since don't could not be an infinitive under any reasonable definition. Nonetheless, in order to avoid the awkward use of a more accurate term, I have arbitrarily opted to use “Root Infinitive” as a descriptive label.

³In Eve's entire transcript there are at most two instances: Everybody don't like the fan (2;01,00) and She don't do it again (2;03,00). At 1;10,00 she is already producing doesn't.%

⁴Menyuk is discussing a rather heterogeneous set of data she collected. It is not completely clear what age range she intends to associate with “the earlier sentences”; my best guess from the surrounding discussion is that it is 2;10–3;00.

⁵Bloom is discussing transcripts from Stage VI of Gia's acquisition of English, when she was 2;03,01 with an MLU of 2.75 morphemes, i.e., toward the early end of the age ranges in Table 1.

⁶At this point Stromswold is discussing tense doubling errors (described in section 2.3), which seem to be her motivation for positing these misanalyses. Nonetheless, the quote is clearly intended to be a general statement about these children's grammars. The instances she is alluding to all come from files in the ranges listed in Table 1.

⁷While nonagreeing don't is ungrammatical in standard English, it is found in several nonstandard dialects, so we would like to know whether any of the children were hearing it. We can never completely exclude that possibility, because we have records of only a tiny fraction of the input they are exposed to, generally limited to speakers who were in the child's home at the time recordings were being made. Nonetheless, I carried out automated searches on the utterances by people other than the target child in the entire transcripts of the five children (not just the files listed in Table 1), looking for sequences of the form “{he/she/it}{don't/doesn't}” and filtering the results by hand. (For practical reasons I assumed that if people were using nonagreeing don't it would occur with pronominal subjects.) The totals of such utterances are as follows: Abe: 0 don't, 43 doesn't; Adam: 0 don't, 140 doesn't; Sarah: 38 don't, 102 doesn't; Nina: 0 don't, 94 doesn't; Ross: 0 don't, 18 doesn't. Thus, based on this limited sample, the only child whom we know to be hearing nonagreeing don't is Sarah, whose mother produces it about one quarter of the time, in apparent free variation with doesn't (the vast majority of these utterances from Sarah's files are her mother's). This was not noted by Guasti & Rizzi. (Even in the Sarah transcripts there are no adult instances of the inverted order “don't {he/she/it},” so I excluded the corresponding inverted agreeing forms from all baseline counts.) The question of how this observation might bear on the interpretation of the results will be dealt with as they are presented below. (Thanks to anonymous referees for encouraging me to look into this issue.)

⁸An anonymous referee suggests that this time frame is still too long, because the Misanalysis Hypothesis “is relevant only for the period before [the children] are productively producing modals.” This stance would seem to imply that if we find nonagreeing don't after modals are being productively used, it could not be treated under the Misanalysis Hypothesis, but would require some entirely different explanation. I reject such an approach as unparsimonious, absent any evidence that the nature of nonagreeing don't changes when modals become productive. I also am not certain what notion of productivity would be considered relevant, but if we use the first clear non-imitative non-formulaic use as the landmark, then the quoted suggestion would render the Misanalysis Hypothesis entirely inapplicable for two of the children under study here: as detailed in section 3.5, Sarah's and Adam's first such use of (nonnegated) can is not later than their first use of don't (for Adam it is two files earlier).

Of course, we cannot exclude the possibility of a stage earlier than the transcripts cover when the Misanalysis Hypothesis would have captured the children's grammar. But with regard to Adam and Sarah at least, the hypothesis was proposed by Bellugi specifically to account for their data in files within the range I analyzed, so it is legitimate to test it on these files.

⁹An anonymous referee suggests that this targeted sampling procedure is improper, and children should have been selected based on some general measure of stage of development. But the goal of this study is to understand the reasons for the production of nonagreeing don't, not to document its prevalence in the population. We are unlikely to advance that understanding by studying children who produce it too infrequently for distributional generalizations to be drawn. Once we have a grasp of why the children who produce it frequently do so, then we could be in a position to ask why other children seem to do so relatively infrequently (so far as we can tell from existing samples). That issue is beyond the scope of this paper; see also section 3.5.

¹⁰MLU in words was calculated using the CLAN software from CHILDES. Abe's figures are much higher than the other children's because in his transcripts very long stretches of speech (apparently multiple sentences) are notated as a single utterance.

¹¹The don't of English negative imperatives arguably is not a composition of do + sentential negation (Beukema & Coopmans 1989), in light of facts like those in (i) (but see Flagg 2002 for an alternative explanation).

¹²12This could have been grammatical as constituent negation if the target was Mommy is hungry, not tired, but the child was not given the benefit of this doubt in the counts.

*A complication arises in Adam's data due to his well-known proclivity to use its apparently as a subject pronoun (Brown 1973), where adult English calls for it. For the purposes of this table, I counted his utterances as if they contained it in place of its as the subject when doing so rendered them grammatical (modulo agreement). Without this adjustment, there would be an additional 17 ungrammatical cases, most involving doesn't.

*The Ungrammatical don't cases for Ross exclude seven utterances of the form “Let's don't VP,” a locution that is possible colloquially in some dialects (though it was not produced by the adults in his transcripts). Since Let's do not … is impossible even colloquially, Let's don't plausibly contains an instance of imperative don't (cf. note 11), and hence is properly excluded from the table on grounds discussed in the text.

¹³As observed in note 7, Sarah sometimes hears nonagreeing don't from adults. I do not believe that fact makes her data irrelevant for the argument in this section, however. The hypotheses in the literature have not tied children's purported misanalysis of don't to any distributional property in the input to begin with. Furthermore, pre-theoretically one might expect that inconsistent use of don't in the input (i.e., hearing it in apparent free variation with doesn't, as Sarah does) should make it more likely to be misanalyzed, if anything, not less likely. As Table 4 shows, however, this is not the case: her word order errors are very similar in frequency to those of the other children, and the proportion of her uses of don't that occur with 3sg subjects is much lower than that of Nina. Even her ratio of don't to doesn't in 3sg contexts is not the highest among these children: Abe's and Nina's are higher. Thus, Sarah's apparently unique input among this group of children does not appear to have made her an outlier in terms of her own production, so I will not exclude her data from subsequent analyses.

¹⁴Indeed, Stromswold (1990: 229) reports that this analysis has been proposed in the literature, though I was unable to find it in the sources she mentions.

¹⁵ Table 7 excludes a substantial number of productions of not, those that occurred in “medial neg” environments, discussed in section 4.3. These are temporarily omitted because their classification as grammatical versus ungrammatical is not immediately obvious. They constitute the one environment where these children superficially appear to be using don't and not in free variation, but I will argue that this is not the correct characterization of the data.

¹⁶Anonymous reviewers have asked why the relevant comparison is not between uses of doesn't versus don't in 3sg subject contexts. The reason is that those numbers would tell us nothing about the children's overall ability to systematically relate verbal inflections to features of the subject. For that we must look beyond just 3sg contexts.

¹⁷Specifically, the portion of the non-agreeing don't stage during which doesn't is also attested is as follows (earliest and latest occurrence): Abe, files 43–79; Adam, files 9–55; Sarah, files 50–128; Nina, files 12–51; Ross, files 24–45.

¹⁸This proposal was initially proffered by Rizzi (1997).

¹⁹It is not clear to me why they eschew characterizing this as a parametric choice, saying rather that in situations where checking has not happened overtly, “the morphological expression of agreement is a matter of a language specific morphological rule, one that must be learned under no special UG guidance. We thus expect the observed fluctuation …for a fairly long period” (183); this rule “is not enforced by the core system of UG principles and parameters” (189). I do not see what the difficulty for the child is in this forced choice situation, nor why the available options cannot be part of UG, therefore it does not seem to me that the prediction of an extended period of uncertainty is justified. I shall put this issue aside, and for expository convenience I will discuss their approach in terms of parameter setting.

²⁰The exact identity of the functional categories involved is not relevant, what matters is which features are checked there.

²¹See section 4.4 for a minor exception.

²²In this subsection I follow Guasti & Rizzi in assuming the older version of Minimalist clause structure (Chomsky 1993) in which AgrS heads a separate projection from Tense, since this is crucial to their account.

²³Guasti & Rizzi apparently assume that the subject does not move through Spec-TP on the way from its VP-internal base position to Spec-AgrSP. For the sake of discussion I grant that this can be ensured by some technical means.

²⁴The same word order is of course possible for other auxiliaries, in particular be: He probably is not sick. This would seem to predict that nonagreement should be possible with be, contra most of the RI literature and Guasti & Rizzi's own counts. They escape this incorrect implication by positing that in finite clauses be must always raise to AgrS overtly, due to its rich morphological agreement paradigm (which marks more distinctions than do), extending Vikner's (1995) crosslinguistic parameterization to language-internal differences among verbs. This forces them to treat the word order where probably precedes is by assuming that the subject has raised higher than Spec-AgrSP, so that the adverb can be adjoined to AgrSP. But availing themselves of this option obviously undermines the relevance of (9) as evidence for the lower position of does. For a different approach to the absence of nonfinite be in the RI stage within the general framework assumed here, see Schütze (2004b).

²⁵This account does not rely crucially on an adjacency-based account of Infl-to-V lowering in English (like that hinted at in the main text, inspired by Bobaljik 1994). As long as Infl does not get affixed to a base-generated verb or auxiliary, for whatever reason, it will be a “stray” affix in need of a host.

²⁶Other M heads in English will include the mandative subjunctive (It is vital that she [_M ⊘ _Subjunc] leave now) and the traditional modal auxiliaries (can, must, should, etc.), encoding the fact that these are all in complementary distribution. Thus, my approach could be characterized as a “dummy modal” approach to do.

²⁷See Zwicky & Pullum (1983) for arguments in a traditional framework as to whether n't is a verbal affix as opposed to simply a clitic counterpart of not. This choice is immaterial for my analysis.

²⁸This is close to Thornton's (1993) treatment of tense-doubled questions like Did he didn't go, where the second instance of did is a realization of Aspect, inserted to support n't. It is also close to Tesan & Thornton's (2005) assumption that the do portion of nonagreeing don't “does not represent the tense category, it is only a morphological host” (270).

²⁹I assume that English modals always bear an Infl suffix, in keeping with their distributional restriction to finite clauses. They select a zero allomorph of the 3sg present suffix, and their past tense forms often have idiosyncratic meanings, but could and would can be simply the past forms of can and will, confirming that the modal+past combination is possible.

³⁰This string is grammatical only with emphasis on does, in which case not is most likely constituent VP negation.

³¹Despite expressing these caveats, Guasti & Rizzi (184) suggest that the data in Table 9 constitute an argument against the kind of analysis I propose in this paper: “These facts suggest that do cannot be simply a host for negation. If it were, it would have such function both in uninverted and in inverted positions, and we would expect to find examples like [Why don't he go].” But if we break Table 9 down by which children contribute to which cells, as in Table I, we can see that no individual child demonstrates the relevant pattern, namely, production of both agreeing and nonagreeing d-words and both inverted and uninverted questions. Therefore, the data do not make the point they are supposed to make, and the empirical issue is open.

³²In fact they tentatively propose a Truncation analysis of those cases. I return to this proposal in section 4.4.

³³It is well-known that some English-speaking children can use dummy do without emphatic prosody and apparently without intended emphatic meaning (Hollebrandse & Roeper 1996; Zukowski 1996; Bohnacker 1999; Thornton 2010; i.a.). For simplicity of exposition, in the main text “emphatic do” is meant to encompass such cases.

³⁴Their fifth child, Peter (Bloom 1970), was excluded here because the number of instances counted by Guasti & Rizzi was already small.

³⁵My counts for Abe do not cover the same range of files as Sano's, because his criteria for determining start and stop points are different from Guasti & Rizzi's, and I follow the latter. Also, as can be seen by comparing Table 1 and Table 11, Abe is the only child for whom nonagreement errors were found in later files than the last word order error involving negation. This results from a single instance of nonagreeing don't in file 94, after a stretch of no such instances in files 78–93.

³⁶We can see from Tables 3–6 that the exclusion of tense doubling utterances from Guasti & Rizzi's counts had almost no impact, because there were extremely few of these in their children's transcripts. However, Table 2 shows that Abe, whom Guasti & Rizzi did not examine, has 11 instances of don't in tense doubling sentences with 3sg subjects. (These turn out all to be in the age range covered by Abe's data in Table 12.) None of them are inverted, and all are followed by an inflected main verb; an example is This don't works. I suggest that these involve only one set of Infl features, which are expressed on the main verb; then do is again occurring entirely to support n't, and such cases are correctly captured by the Underspecification Account.

³⁷Once more we should ask, given the aforementioned presence of 3sg don't in Sarah's input, whether her data in Table 12 should be treated specially. If it had turned out that Sarah never produced nonagreeing do in inversion or “Other” contexts, one might have attributed that to the idea that in her grammar don't is fully inflected, as it apparently is in her mother's grammar, and thus there would be no evidence that she is in a stage of Infl underspecification at all. The fact that she produces a few nonagreeing do forms, along with the fact that her ratio of 3sg don't to doesn't is much higher than what is found in her input, suggests instead that nonagreeing don't may have a different analysis in her grammar from what it has in her mother's.

*The cells where the column labeled “do” intersects a row labeled “Before n't” refer to productions of the word don't.

³⁸ Thornton (1999) reports a single child's data (age 2;05) from elicitation sessions that focused on VP-ellipsis with 3sg subjects. In non-negated VP-ellipsis utterances the child used does 27 times, do 0 times; with contracted negation she used an idiosyncratic inflected form don'ts 11 times, doesn't 11 times, and nonagreeing don't 12 times. This is as predicted by the Underspecification Account but not by the Spell-Out Account.

³⁹Guasti's description of the mechanism underlying forms that lack finiteness for purely morphological/spell-out reasons appears to be different from that in Guasti & Rizzi (2002), which was written earlier. Guasti (2000) says that nonagreeing don't and the relevant uninflected root main verbs have a -⊘ suffix instead of -s, but she does not explain why the “wrong” suffix is chosen. This is not a trivial re-statement, because -⊘ Infl triggers do-support when it lacks a host (for adults, at least). My discussion in the main text assumes that Guasti & Rizzi's (2002) formulation is compatible with Guasti's (2000) other claims.

⁴⁰The Ungrammatical category includes “inflected medial neg” examples (Tesan & Thornton 2005), where the main verb following not is inflected with 3sg -s or past tense. These are claimed by Harris & Wexler (1996) to be virtually unattested in corpus data because they would violate the Head Movement Constraint, and indeed, there turned out to be only two such examples, Abe's It not works Mom and Ross's My tummy not hurts. But see Tesan & Thornton (2005) for elicited data where inflected medial neg is more frequent. Also, I observed in note 36 the existence of 11 utterances like This don't works in Abe's speech, which are just as unexpected for Harris & Wexler.

⁴¹ Stromswold (1997) independently analyzed the distribution of not in many of these same transcripts, along with those of several other children. Details aside, she reached the same conclusion as I do: word order errors involving the placement of not are exceedingly rare.

*As before, I counted an utterance of Adam's as if it contained it as the subject in place of its when doing so rendered the utterance (closer to) grammatical. Without this adjustment, there would be one additional ungrammatical case in this table.

⁴²As is evident from the raw numbers in the tables, this mean belies a large range across children, potentially an artifact of very small sample sizes. A common way to attempt to deal with this problem is to pool the number of tokens across children, assuming for the sake of calculation that each child provides a sample representing the same developing grammar (given that they are all producing a common set of utterance types), and that larger samples are more informative. Pooling these four children's data yields a do-omission rate of 71%, which I take to further support the claim that this is not a negligible phenomenon.

⁴³It is unclear what should allow the subject to surface in Spec-TP, which is otherwise not an available position in English (cf. *There have some people crashed the party).