2018 Sat Poster 6559

Saturday, November 3, 2018 | Poster Session II, Metcalf Small | 3:15pm

Privileged computations for closed-class items
H. Getz, E. Newport

Across languages, closed-class items (e.g. determiners, tense) are perceptually distinct due to their high frequency, short duration, and light syllable structure.1 Previous research2,3,4,5 suggests that ‘anchoring’ distributional analysis to these items helps learners acquire grammatical structure. However, we lack a mechanistic understanding of what learners do with closed-class items. Are closed-class items merely “flags”, indicating other elements to analyze? Or are closed-class items the elements learners analyze?

In natural languages, closed-class items predict open-class items, but not the reverse. For example, the always occurs with a noun, but nouns can occur without the.6 In two miniature language experiments, we asked whether learners analyze closed-class items as predictive of open-class items. “Open” and “closed” classes were distinguishable based on class size and phonological properties. The closed class contained a single short, monosyllabic item (e.g., ka). Open classes contained multiple longer words (e.g., tombur, lapal).

In Experiment 1, children age 7-9 were exposed to a miniature language Noun-Inflection-Verb. (Words were meaningless; these labels are used for clarity.) All words were optional, but if Infl was present, V was also present; V could occur with or without Infl (“if Infl then V”). Two conditions contrasted whether V or Infl was closed class (Table 1). In the CLOSED-CLASS INFL condition, Infl was closed and V was open. In the OPEN-CLASS INFL condition, Infl was open and V was closed. If learners analyze closed- class items as predictive, they should learn the Closed-Class Infl condition best. Children listened to an alien speaking one of these languages for 20 min and then took a 2AFC test measuring (i) preference for Infl V over other grammatical sequences and (ii) knowledge that Infl predicts V and not vice versa (Figure 1). Only learners in the Closed-Class Infl condition performed above chance, suggesting that learners analyze closed-class items as predictive of open-class items but not the reverse.

In Experiment 2, we asked whether learners could use this mechanism to acquire “verb movement”, where the verb’s position correlates with its inflection (e.g. as in German). In a new miniature language either N or V could occur first (both were obligatory). V could occur with or without an Inflection prefix luh. In one condition, verbs in first position were always inflected, but inflected verbs could be first or second (V1àINFL). In a contrasting condition, inflected verbs always occurred first, but verbs in first position could be inflected or uninflected (INFLàV1). These exposure conditions are extremely similar (2/3 of the sentences overlap) but computationally distinct (Table 2). Children age 7-9 listened to the language 20 min/day for 2 days, then rated novel sentences as grammatical or ungrammatical. Learning was strong in the InflàV1 condition but at chance in the V1àInfl condition (Figures 2 and 3).

These results show that learners analyze closed-class items as predictive of the presence (Experiment 1) and position (Experiment 2) of open-class items. Modern syntactic theory posits that language patterns are organized around a small set of closed-class items.7 This computational mechanism may underlie the acquisition of such patterns.

References

  1. Shi, Morgan, & Allopena (1998) Journal of Child Language
  2. Valian and Coulson (1988) Journal of Memory and Language
  3. Morgan, Meier & Newport (1987) Cognitive Psychology
  4. Braine (1963) Psychological Review
  5. Shi, Werker, & Cutler (2006) Infancy
  6. Saffran (2001) Journal of Memory and Language
  7. Rizzi & Cinque (2016) Annual Review of Linguistics