ECoh: Turn-level Coherence Evaluation for Multilingual Dialogues
This provides a lightweight, open-source solution for evaluating dialogue coherence across multiple languages, addressing a bottleneck in the field.
The paper tackles the lack of open-source multilingual dialogue evaluators by introducing GenResCoh, a dataset with over 130k responses, and ECoh, a family of evaluators trained on it. ECoh achieves superior multilingual coherence detection compared to GPT-3.5-Turbo and provides high-quality explanations.
Despite being heralded as the new standard for dialogue evaluation, the closed-source nature of GPT-4 poses challenges for the community. Motivated by the need for lightweight, open source, and multilingual dialogue evaluators, this paper introduces GenResCoh (Generated Responses targeting Coherence). GenResCoh is a novel LLM generated dataset comprising over 130k negative and positive responses and accompanying explanations seeded from XDailyDialog and XPersona covering English, French, German, Italian, and Chinese. Leveraging GenResCoh, we propose ECoh (Evaluation of Coherence), a family of evaluators trained to assess response coherence across multiple languages. Experimental results demonstrate that ECoh achieves multilingual detection capabilities superior to the teacher model (GPT-3.5-Turbo) on GenResCoh, despite being based on a much smaller architecture. Furthermore, the explanations provided by ECoh closely align in terms of quality with those generated by the teacher model.