After the Null Subject Parameter: Acquisition of the Null-Overt Contrast in Spanish

ABSTRACT

In many so-called canonical null subject languages, null and overt subject pronouns have contrasting referential preferences: null subjects tend to maintain reference to the preceding subject while overt pronominal subjects do not. We propose that children acquire this contrast by initially restricting their attention to 1^st and 2^nd person pronouns, whose reference is simpler to infer compared to 3^rd person pronouns. We provide supporting evidence from spontaneous production and comprehension in Mexico City Spanish, showing that (i) the null/overt contrast is in principle acquirable from exclusively observing the referential preferences of 1^st and 2^nd person subject pronouns in caretaker speech; (ii) children themselves condition subject pronoun expression on pronoun reference in the 1^st and 2^nd persons before doing so in the 3^rd person; and (iii) children use the null/overt contrast in comprehension at a similar age when they begin making this distinction in production.

Acknowledgments

We would like to thank the members of the Michigan State University Language Acquisition Lab and the University of California Irvine Computation of Language Lab. Many thanks to our adult participants at MSU and to the teachers, parents, and children at Servicios Educativos del Desarrollo Infantil, Mexico City, D.F., as well as Beti López Juárez and Patricia de la Fuente Zuno. Thanks to the audiences of GALANA8, HLS2018, and BUCLD43 for insightful feedback. This research was funded in part by a Dissertation Completion Fellowship from the MSU Graduate School to Hannah Forsythe, by NSF Postdoctoral research grant #SPRF-1810159 to Hannah Forsythe, and by NSF grant #BCS-1656133 to Cristina Schmitt.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

¹ For concreteness we assume the variational model proposed by Yang (2004) for parameter setting. In this model children are endowed with a restricted hypothesis space, in the sense that the parametric options are given from the start. The child’s task is to eliminate parametric options that don’t fit the input by a Darwinian mechanism that punishes or rewards parametric options against every sentence the child parses. As the child is exposed to a higher number of sentences that fit one particular parametric option and not others, this setting will outweigh alternative settings and become stable. After setting parameters, the grammar is stable, but there is still learning to be done because the child also needs to learn exceptions to a rule (Yang, 2016) and the constraints (linguistic or extralinguistic) that determine the use of a particular grammatical option, and this will also depend to a certain extent on statistical learning. It is important to note that, although the use of statistical learning is common in usage-based/cognitive linguistic proposals, our approach does not assume that everything can be learned directly from the input without a very constrained hypothesis space (which we could call UG).

² See Barbosa (2011a, 2011b) for a discussion of canonical or “full” null subject languages versus partial null subject languages. In this paper we assume that in canonical null subject languages the null subject is a phonologically empty version of the overt subject, which we gloss here as pro (following Holmberg, 2005; Roberts, 2010, and others). Other analyses (most notably Alexiadou & Anagnostopoulou, 1998) argue that null subject clauses actually do not have a pro, but rather that the agreement marker on the verb itself performs the function of subject – it is an enclitic pronoun on the verb. On this latter analysis, the overt subject (las llaves in example (1a), ella in example (1b)) is not considered to be a true subject but rather a clitic-left-dislocated phrase adjoined to the clause. Regardless of which analysis turns out to be correct, the learning problem we are considering here remains largely unchanged. Children must still learn when to use which structure/form.

³ In canonical pro-drop languages, person and number features are in principle recoverable from the verb, so they are available even when the subject itself is null.

⁴ There are certain highly restricted cases such as (i) where 3^rd person plural NPs (ex. las dos “the two of us”) can appear with 1^st person plural morphology on the verb. In our sample of over 53,000 utterances we found 19 examples.

1. Vamos a bailar las dos juntas[JRC 5;11]

Go-1P to dance the two together

“Let’s dance the two of us together.”

⁵ This corpus was recorded in 2008 in Mexico City, Mexico and comprises approximately 1–2 hours of free-speech dialogue from 25 child-caretaker dyads. See section 5 for details.

⁶ The 2^nd person singular formal pronoun usted and 2^nd -person plural pronoun ustedes both trigger 3^rd-person agreement on the verb but refer to the addressee(s) and are therefore coded as semantically 2^nd person pronouns. Children produced a total of 15 (13 null, 2 overt), or 3.4% of 2^nd person subject pronouns. Adults produced 49 (46 null, 3 overt) or 3.1% of 2^nd person subject pronouns.

⁷ This is similar in spirit to the approach of Torres Cacoullos and Travis (2018), who argue that “human switch-reference” is the crucial switch-reference context for overt subject pronoun realization.

⁸ We also tested models including SES as an interaction with these main effects as well as a simple main effect. The interaction model failed to converge. The model including SES as a simple main effect failed to fit the data significantly better than the model without SES, as determined using R’s anova() function (χ²(1) = 0.60, p = 0.44). SES was therefore excluded.

⁹ A reviewer asks whether it is possible that this result is driven by children over 4 ½ simply using more 1^st and 2^nd person pronouns, for which the same/switch contrast is stronger. However, the frequency of 1^st and 2^nd person pronouns, as opposed to 3^rd person pronouns, does not correlate with age (β = 0.299, SE = 0.353, z = 0.299, p = .40). Moreover, three children (EAMR, PLG, and SLV) produce more 3^rd person pronouns than 1^st and 2^nd person combined, yet they still significantly distinguish between same- and switch-reference (see Appendix 1, Table II).

Additional information

Funding

This work was supported by the National Science Foundation [BCS-1656133,SPRF-1810159]; Michigan State University Graduate School [Dissertation Completion Fellowship].