Abstract
We present a comprehensive evaluation framework for assessing interactive dialogues in second language (L2) communication, specifically focusing on English as a Second Language (ESL) and Chinese as a Second Language (CSL). The framework captures from dialogue overall quality, to dialogue-level interactivity labels (e.g., topic management) and micro-level linguistic features (e.g., backchannels, reference words), using 4 interactivity labels and 17 micro-level features. By applying to second language interactive conversations, our analysis revealed that certain micro-level features, such as reference words (e.g., she, her, he), strongly correlate with the overall quality of dialogue interactivity. This provides valuable insights into how lower-level linguistic signals influence higher-level dialogue outcomes. Our work opens new possibilities for automated, cross-linguistic dialogue evaluation and language learning assessment. These studies pave the way for expanding the possibilities of applying automated methods in speaking assessment for second language speaking tests.
