Automatic Analysis of Linguistic Complexity and Its Application in Language Learning Research

Abstract

The construct of complexity, together with accuracy and fluency have become the central foci of language learning research in recent years. This dissertation focuses on complexity, a multidimensional construct that has its own working mechanism, cognitive and psycholinguistic processes, and developmental dynamics. Six studies revolving around complexity, including its conceptualization, automatic measurement, and application in language acquisition research are reported.

The basis of these studies is the automatic multidimensional analysis of linguistic complexity, which was implemented into a Web platform called Common Text Analysis Platform by making use of state-of-the-art Natural Language Processing (NLP) technologies . The system provides a rich set of complexity measures that are easily accessible by normal users and supports collaborative development of complexity feature extractors.

An application study zooming into characterizing the text-level readability with the word-level feature of lexical frequency is reported next. It was found that the lexical complexity measure of word frequency was highly predictive of text readability. Another application study focuses on investigating the developmental interrelationship between complexity and accuracy, an issue that conflicting theories and research results have been reported. Our findings support the simultaneous development account.

The other few studies are about applying automatic complexity analysis to promote language development, which involves analyzing both learning input and learner production, as well as linking the two spaces. We first proposed and validated the approach to link input and production with complexity feature vector distances. Then the ICALL system SyB implementing the approach was developed and demonstrated. An effective test of the system was conducted with a randomized control experiment that tested the effects of different levels of input challenge on L2 development. Results of the experiment supported the comprehensible input hypothesis in Second Language Acquisition (SLA) and provided an automatizable operationalization of the theory.

The series of studies in this dissertation demonstrates how language learning research can benefit from NLP technologies. On the other hand, it also demonstrates how these technologies can be applied to build practical language learning systems based on solid theoretical and research foundations in SLA.