Automatic extraction of subordinate clauses and its application in second language acquisition research

Abstract

Clause subordination is an important linguistic phenomenon that is relevant to research in psycholinguistics, cognitive and behavioral sciences, language acquisition, and computational information retrieval. The paper presents a comprehensive tool called AutoSubClause, which is specifically designed for extracting subordinate clause (SC) information from natural English production. Using dependency parsing, AutoSubClause is able to extract not only information characterizing the three main types of SCs—complement, adverbial, and relative clauses—but also information regarding the internal structure of different clause types and their semantic and structural relations with elements of the main clause. Robustness testing of the system and its underlying dependency parser Stanford CoreNLP showed satisfactory results. To demonstrate the usefulness of AutoSubClause, we used it to analyze a large-scale learner corpus and investigate the effects of first language (L1) on the acquisition of subordination in second language (L2) English. Our analysis shows that learners from an L1 that is typologically different from the L2 in clause subordination tend to have different developmental trajectories from those whose L1 is typologically similar to the L2. Furthermore, the developmental patterns for different types of SCs also vary. This finding suggests the need to approach clausal subordination as a multi-componential construct rather than a unitary one, as is the case in most previous research. Finally, we demonstrate how NLP technology can support research questions that rely on linguistic analysis across various disciplines and help gain new insights with the increasing opportunities for up-scaled analysis.