Report

Teaser, summary, work performed and final results

Periodic Reporting for period 2 - CLASS (Cross-Linguistic Acquisition of Sentence Structure: Integrating Experimental and Computational Approaches)

Teaser

How children acquire their native language remains one of the key unsolved problems in Cognitive Science. The aim of this project is to answer a question that lies at the heart of this problem: How do children acquire the abstract generalizations that allow them to produce...

Summary

How children acquire their native language remains one of the key unsolved problems in Cognitive Science. The aim of this project is to answer a question that lies at the heart of this problem: How do children acquire the abstract generalizations that allow them to produce novel sentences, while avoiding the ungrammatical utterances that result from across-the-board application of these generalizations? Although our primary objective is simply to advance scientific knowledge on this question, the findings could have potential implications for informing interventions for children with developmental language disorder, for second language learning, and for applications such as machine translation. The difficulty of this problem is highlighted by the fact that many children pass through a stage in which they produce these types of errors (e.g., *The clown laughed the man [for The clown made the man laugh]), before retreating from them, even in the absence of adult feedback (Bowerman, 1988; Marcus, 1993). Despite decades of research (since Braine, 1971), the only current attempted solutions to this problem are single-process theories – the entrenchment, preemption and semantic-verb-class accounts – none of which can explain all of the current data for English, let alone across other languages. This project tests a new account of the Cross-Linguistic Acquisition of Sentence Structure (CLASS). Our objectives are to explain how children develop exactly the right generalizations, eventually developing into adult speakers who do not produce such errors, and regard them as ungrammatical.

Work performed

So far, the majority of work completed has been on Work Package 1, which investigates children\'s knowledge of causative sentence structure across languages. Children learning to speak English (as a their mother tongue) sometimes produce errors such as “*The funny clown giggled the boy” or “*He falled over” (an asterisk indicates an ungrammatical sentence). Analogous errors have been observed for children learning other languages and – indeed – are presumably produced by children learning all the languages of the world. In WP1 we investigate how children learning five languages – English, Japanese, Hindi, Hebrew and K’iche’ Mayan – eventually come to stop producing such errors.

The hypotheses under investigation are (1) that such errors are gradually “pre-empted” or “blocked” as children learn the correct forms (e.g., “The funny clown made the boy giggle”; “He fell over”) and (2) that children form groups of verbs with similar meanings that behave in similar ways. For example, “*The funny clown laughed/chuckled/tittered the boy” all sound odd in exactly the same way that “*The funny clown giggled the boy” does, because the verbs “giggle”, “laugh”, “chuckle” and “titter” all have similar meanings, and so all behave in similar ways with regard to the sentences that they can and cannot appear in.

We have been testing these predictions by asking children and adults to (a) rate these errors using a 5-point “smiley face” scale and (b) describe animated cartoons (since they sometimes produce these errors when doing so). The aim to see whether children produce these errors more often – and rate them as less unacceptable - when the correct competing forms (e.g., fell vs *falled) are frequent in the language that children hear (estimated via publicly available corpora of speech to children). We also test the prediction that these errors will be rare – and rated as unacceptable – for verbs that describe “internally-caused” actions (which is why we can say “The clown pushed the boy” but not “*The clown giggled the boy”). To determine this, we have completed a rating task in which speakers of all five languages rate actions/verbs on this property.
We have almost completed the judgment tasks, with 704/720 participants (48 at ages 5-6, 9-10 and adults for each language) already tested. We have just begun work on the production tasks with 173/720 participants tested. The preliminary results from the almost-completed judgment task support the predictions of the new account under investigation: all three effects - semantics, preemption and entrenchment - are observed in the data, meaning that none of these single-process theories can work. This cross-linguistic work allows to investigate (a) how children develop exactly the right generalizations, (b) why they show effects of entrenchment in some cases and preemption or verb-semantics in others, (c) why the distribution of these effects is different across different structures and different languages. The next step (under Work Package 4) is to build a computer model of language learners that instantiates this new account, and to investigate whether it simulates the patterns observed in the data already collected. This will allow us to investigate (g) how children develop into adult speakers who do not produce such errors, and regard them as ungrammatical, in a way that is not possible with static statistical models.

The objective of Work Package 2 is to investigate the acquisition of language-specific constructions found in Hebrew, Japanese and Hindi; and whether these apparently-disparate phenomena can be explained by the new account under investigation. We have not yet begun work on the Hebrew and Hindi studies, but, thanks to the visit of a visiting researcher under the ERC Implementing Arrangements, we were able to make an early start on the design of the Japanese study. This investigates a phenomenon wher

Final results

Before the project, the current state of the art was studies conducted solely on English (which we summarized in a 2018 meta-analysis paper). We have begun to progress beyond the state of the art by testing both old theories and the new theory developed as part of the current work on four more languages (Hindi, Hebrew, K\'iche\' and Japanese), and artificial lab-created languages. This latter development also represents progress beyond the state of the art, as we have been able - for the first time - to create a novel language with semantic distinctions that are readily learnable by children.

Website & more info

More info: http://pcwww.liv.ac.uk/.