Vivian Tsang & Suzanne Stevenson
University of Toronto
Theories of verb classification have elaborated a detailed mapping from underlying semantics to overt syntactic behavior (Pinker, 1989; Levin, 1993). This syntax/semantics mapping appears to aid language Acquisition, as the child uses syntactic cues to induce properties of verb semantics (Gleitman, 1990; Gillette et al., 1999). Recent experiments in Second Language Acquisition (SLA) reveal another fascinating role played by this mapping: second language (L2) learners assume that L2 verbs allow the same syntactic constructions as semantically similar verbs in the native language (L1) (Helms-Park, 2001; Inagaki, 1997). Thus, L2 learners appear to generalize their knowledge of the syntax/semantics mapping in L1 to learn the syntax of verbs in L2 --- an instance of "transfer effects" in SLA.
We have investigated these transfer effects in a set of computational experiments that explore the ability of L1 features to aid in the learning of L2 verb classes. Verb classes encapsulate the syntax/semantics mapping, and have thus been assumed to underlie the above SLA observations. We model the observed behavior in our experiments as the computational process of determining the appropriate class for an L2 verb, on the basis of semantic and syntactic similarities between verbs in L1 and L2.
Specifically, we select 16 verbs from each of two semantic classes in English --- the L2 for our study. We then select a set of verbs in Chinese --- the L1 for our study --- that are translations of the L2 verbs in a bilingual (English and Chinese) corpus. We determine syntactic features of the verbs related to the semantic class distinctions, and collect statistics over both the English verbs and their Chinese translations, from the bilingual corpus. We then use these features to train a system to classify the English verbs.
Earlier work has shown that such statistical syntactic features within English can be used to classify English verbs into semantic classes (Merlo & Stevenson 2001). Here we find that, analogously to the SLA observations, the syntactic behavior of L1 verbs (as captured in the Chinese statistical features) transfers over to L2 (English) verbs, aiding in their classification. In classifying the English verbs (a task with chance performance of 50%), we achieve an accuracy of 80% using a combination of English and Chinese features, significantly outperforming monolingual features on the same task.
We conclude that the syntax/semantics mapping for verbs, which plays a role in first language acquisition, may also be exploited crosslinguistically in SLA. Our computational experiments support the hypothesis that L2 learners use the mapping between the semantics and syntax of verbs in their L1, in acquiring properties of verbs in L2. Furthermore, our experiments elaborate a possible mechanism underlying this transfer of knowledge --- namely, the statistical analysis of verb behavior and its relation to semantic classes.
J. Gillette, L. Gleitman, H. Gleitman, & A. Lederer (1999). Human simulation of vocabulary learning. Cognition, 73(2): 135-176.
L. Gleitman (1990). Structural sources of verb meaning. Language Acquisition, 1(1): 3-55.
R. Helms-Park (2001). Evidence of lexical transfer in learner syntax. Studies in Second Language Acquisition, 23(1): 71-102.
S. Inagaki (1997). Japanese and Chinese learners' acquisition of the narrow-range rules for the dative alternation in English". Language Learning, 47(4): 637-669.
B. Levin, (1993). English Verb Classes and Alternations. University of Chicago Press.
P. Merlo, & S. Stevenson, 2001. Automatic verb classification based on statistical distribution of argument structure. Computational Linguistics, 27(3): 393-408.
S. Pinker (1989). Learnability and Cognition. MIT Press.