ABSTRACT
A system for scoring spontaneous speech is expected to assess global oral proficiency, which is widely regarded to be multi-componential. In the context of second language (L2) speaking, grammar, and vocabulary skills have been considered to be two core dimensions that strongly influence L2 learners’ oral proficiency, along with fluency and pronunciation. Not surprisingly, the strong impact of grammar and vocabulary on L2 proficiency has led to a growing interest in the development of quantitative features that can estimate the developmental level of L2 grammar and vocabulary skills. During the feature development stage, researcher considerations included the construct relevance of the features, feasibility of implementing the automated feature extraction process, and the features’ strength of association with oral proficiency scores. The automated assessment of lexical or grammatical accuracy from spontaneous speech requires very high accuracy in the automated speech recognition system, as well as multiple automated sub-processes such as syntactic parsing and automated error detection.
