2018 Sun Session B 0930

Sunday, November 4, 2018 | Session B, Conference Auditorium | 9:30am

Bootstrapping syntactic acquisition with a semantic seed
M. Babineau, A. de Carvalho, J. Trueswell, A. Christophe

Several studies suggest that, when learning the meaning of a new word, infants can base their inferences on the other words in the sentence (the syntactic context), a mechanism called syntactic bootstrapping [1]. For instance, when hearing, “It’s a pratch!” 18-month-olds infer that pratch refers to an object, while upon hearing, “It’s pratching!”, they infer that pratching refers  to  an  action  (e.g.,  [2]). However, the question of  how infants learn which syntactic contexts correspond to which semantic features is still unresolved. Infants could solve this problem using their knowledge of a handful of word meanings (frequent, concrete words, like “ball” and “eat”, [3]), which are grouped into basic semantic categories (e.g., object, action, [4]). These known words are then used as a seed  for future syntactic categories [5]. Infants would learn the syntactic contexts in which these familiar words occur (e.g., the car; the ball), then use this information to infer some semantic properties of new words they encounter in the same syntactic contexts: the dax ⇒ dax refers to an object. Critically, this should hold true also for new function words that are encountered with known words of a  particular  syntactic class; e.g, hearing, “ko shoe” and “ko ball” could lead to the expectation that “ko dase” refers to an object, whereas hearing, “ko is pushing” and “ko will play” should lead to the expectation that “ko dase” instead refers to an action.

To test this experimentally, we taught children a novel function word (“ko”) embedded in sentences which contained words they already know (the semantic seed). In a between- subject design, French-speaking 3-4-year-olds were assigned to one of two conditions (Noun vs Verb). They watched a training video showing a woman speaking in child-directed speech while playing with stuffed animals and toys. Each video contained the same sentences/script, except that articles were replaced by “ko” in the Noun Condition (“ko turtle”, “ko pig”), while pronouns were replaced by “ko” in the Verb Condition (“ko plays”, “ko rolls”; 60 occurrences in each, see Fig.1). During the test phase, two videos were presented on the screen, one showing a novel object and the other a novel action, along with sentences which contained “ko”, followed by a new content word (e.g., Oh look! Ko dase!). Children’s pointing responses and eye gaze were collected. The results showed that in the Verb Condition children pointed more toward the action video than in the Noun Condition (Fig.2), suggesting that in the Noun Condition they learned that “ko” was preceding words referring to objects (noun contexts), while in the Verb Condition “ko” preceded words referring to actions (verb contexts).

Our results show that three-year-olds can rapidly encode a new function word (“ko”), and later exploit it to infer the probable meaning of novel content words. More generally, children can rely on the meaning of a few known words (the semantic seed) to learn their syntactic contexts, which in turn can help them to uncover the meaning of new words appearing in the same syntactic contexts.

References

  1. Gleitman, L. (1990). The Structural Sources of Verb Meanings. Language Acquisition, 1, 3–55.
  2. He, A. X., & Lidz, J. (2017). Verb learning in 14-and 18-month-old English-learning infants. Language Learning and Development, 13(3), 335-356.
  3. Bergelson, E., & Swingley, D. (2012). At 6-9 months, human infants know the meanings of many common nouns. PNAS, 109(9), 3253–8.
  4. Carey, S. (2009). The origin of concepts: Oxford University Press.
  5. Christophe, A., Dautriche, I., de Carvalho, A., & Brusini, P. (2016). Bootstrapping the Syntactic Bootstrapper. In Proceedings of the 40th annual Boston University Conference on Language Development. Boston, MA: Cascadilla Press, 75-88.