From corpus to target data as steps for automatic assessment of L2 speech: L2 French phonological lexicon of Japanese learners

在2021年至2022年期间，主要遵循三个轨道：1)探索CLIJAF语料库中的语音注释；2)在CLIJAF语料库上测试多级自动评估工具；3)准备试点实验方案（包括与C. Barcat和S. Kim起草法英同源词列表）。首先，对CLIJAF的语音标注进行了分析，并综合了一个自由会话任务中不同熟练程度的主要错误。为了做到这一点，我们建立了一个预先评估实验，招募了法语专家教师来评估CLIJAF发言者在口语生产方面的水平。最终，这项研究发表在《日本语音学会杂志》（Detey et al. 2022）上。其次，与法国图卢兹合作团队（L. Fontan， V. de Fino和IRIT研究人员）使用CLIJAF语料库测试不同语言水平（流利度，词汇丰富度，句法复杂性等）的自动评估。最终，该研究在不同的会议上发表，包括INTERSPEECH2022 （De Fino et al. 2022）。第三，为2022年秋季在日本进行现场定向记录和实验方案准备材料和研究计划。

Between 2021 and 2022, three main tracks have been followed: 1) exploration of the phonetic annotations in the CLIJAF corpus, 2) testing the multilevel automatic assessment tools on the CLIJAF corpus, 3) prepare the pilot experimental protocol (including the drafting of a list of French-English cognate words with C. Barcat & S. Kim). First, the phonetic annotations of CLIJAF were analyzed and resulted in a synthesis of the main errors to be found at different levels of proficiency in a free conversation task. To do this a prior evaluation experiment was set up with expert French teachers recruited to assess the level of the CLIJAF speakers in spoken production. Ultimately, this was published in (Detey et al. 2022) in the Journal of the Phonetic Society of Japan. Second, with the French partner team in Toulouse (L. Fontan, V. de Fino and IRIT researchers) the CLIJAF corpus was used to test the automatic assessment of different linguistic levels (fluency, lexical richness, syntactic complexity, etc.). Ultimately, this was presented and published in different conferences, including INTERSPEECH2022 (De Fino et al. 2022). Third, the material and research plan were prepared to run the on-site targeted recording and experimental protocol in Japan in the Fall 2022.