Acquiring wanna: Beyond Universal Grammar

ABSTRACT

The wanna facts are a classic Poverty of Stimulus (PoS) problem: Wanna is grammatical in certain contexts (Who do you want PRO to play with?) but not others (Who do you want who to play with you?). On a standard analysis, “contraction” to wanna is blocked by some empty constituents (WH-copies) but not others (PRO). All empty constituents are inaudible, so it has been unclear how restrictions on them could be learned. Children’s reported knowledge of the wanna facts (Crain & Thornton, 1998) has therefore been attributed to a principle of Universal Grammar (UG). In two experiments, we demonstrate that children’s use of wanna is not in fact adultlike and that error rates are modulated by the frequency of the embedded verb (play). These results suggest that if there is a UG principle, children appear not to know that it is relevant, raising important questions about what learning mechanisms enable children to circumvent the input’s apparent poverty.

Acknowledgments

Many thanks to Elissa Newport and David Lightfoot for invaluable discussion and comments throughout this project. I am also grateful to Jeffrey Lidz and Misha Becker for thoughtful comments provided during the editorial process; to Tiffany Yang, Gabriella Iskin and Sarah Furlong for assistance with data collection; and to the staff, parents, and children at the Washington International School and the Arlington YMCA.

Notes

¹ The status of PRO is controversial, and some have argued that the empty constituent in (1a) is actually a movement trace (e.g., Boeckx & Hornstein 2004). Our use of the label should be taken as a notational convenience, not a commitment to any particular syntactic view.

² In previous research, the contexts in which wanna is grammatical and ungrammatical have been referred to in ways that center a particular view of the restriction on wanna (e.g., “object extraction” for (1a) and “subject extraction” for (1b) reflect a movement-based analysis). Here we adopt the theory-neutral notation ✓wanna and *wanna to refer to the form when it is used in a grammatical or ungrammatical context respectively. We also differentiate between the phonological form wanna and the lexeme want.

³ I thank David Lightfoot for bringing this illustrative fact to my attention.

⁴ Major changes included omitting the object in a transitive construction (“Who do you want to kick the football?” → “Who do you want to kick?”), appending a preposition (“Which one do you want to ride to space?” → “Which one do you want to ride to space with?”), or changing the main and/or embedded verb (“Who do you want to put on the chest?” → “Who do you wanna protect the treasure?”).

⁵ Embedded verbs in the two conditions were matched overall for mean and median frequency; however, we did not a priori divide them into “lower” and “medium” frequency. A post hoc cutoff of 360 resulted in even numbers of verbs in each group (six in each category for each condition).

⁶ Several reviewers wonder if the frequency effect is actually an item effect (e.g., because several lower-frequency embedded verbs were intransitive, and the sentences could have been misparsed). It is not. Use of *wanna was attested for every single embedded verb, regardless of argument structure. The rate of *wanna for individual items ranged from 14% (catch) to 67% (fly); see section 4.3.1 for further discussion.

⁷ One child (C004) produced only one target question in the *wanna condition (with a low-frequency embedded verb) and therefore had zero observations for medium-frequency embedded verbs. He is excluded from the statistical analysis reported in this paragraph, but we include him in Table 2 and Figure 3.

⁸ It was not possible to include WH-item as a term in the repeated measures ANOVA in Section 2.2.3 because there were too few observations per cell. However, we performed a separate repeated measures ANOVA with WH-item and syntactic context (but not frequency or age group) as within-subjects variables and confirmed that the rate of *wanna did not differ significantly by WH-item (p > .4).

⁹ Argument structure errors are not the only possible source of misanalysis. It is also possible that children misanalyzed the sentences in some other way. For example, a reviewer suggests that a child might misanalyze to as a complementizer and the WH-copy in canonical embedded subject position, such that the WH-copy does not intervene between want and to. This story, while not implausible, illustrates the puzzles that remain even for a child with substantial innate knowledge. She must not only determine whether it is PRO or a WH-copy that intervenes between any given sequence of want and to; she must also know the precise position that each empty constituent occupies in the sentence.

¹⁰ A reviewer points out recent work suggesting that older children’s difficulty with promise has to do with structural intervention effects (Mateu Martin, 2016). This may be true, but it does not change the nature of the learning required. Children must learn that the word promise but not the word tell is licensed in a particular structural configuration. The only way to do this is by analyzing the distribution of the form promise.

¹¹ The term “abstract” is used differently in UG and non-UG approaches, so it is worthwhile to be explicit about what we mean here. The representations in Figure 7 are “abstract” in that they reflect generalizations over experience (specific words and sentences). In UG approaches, “abstract” is sometimes used to refer to representations that exist independently of any experience, as innate mental representations. The representations in Figure 7 are not abstract in that sense, though their components (e.g., grammatical categories) could be.

¹² This included the following corpora: Bates, Bernstein, Bliss, Bloom70, Bloom73, Bohannon, Brent, Brown, Clark, Cornell, Demetras1, Demetras2, ErvinTripp, EllisWeismer, Feldman, Gleason, Kuczaj, MacWhinney, McCune, Morisset, Nelson, NewEngland, NH, Normal, Peters, Post, Providence, Rollins, Sachs, Snow, Soderstrom, Trdif, Valian, Vanhouten, Vankleeck, Warren, and Weist.

Additional information

Funding

This research was supported by NIH grants R01 HD037082 and K18 DC014558, by the Feldstein Veron Fund for Cognitive Science, and by the Center for Brain Plasticity and Recovery at Georgetown.