Behavioral testing offers a crucial means of diagnosing linguistic errors and assessing capabilities of NLP models. However, applying behavioral testing to machine translation (MT) systems is challenging as it generally requires human efforts to craft references for evaluating the translation quality of such systems on newly generated test cases. Existing works in behavioral testing of MT systems circumvent this by evaluating translation quality without references, but this restricts diagnosis to specific types of errors, such as incorrect translation of single numeric or currency words. In order to diagnose general errors, this paper proposes a new Bilingual Translation Pair Generation based Behavior Testing (BTPGBT) framework for conducting behavioral testing of MT systems. The core idea of BTPGBT is to employ a novel bilingual translation pair generation (BTPG) approach that automates the construction of high-quality test cases and their pseudoreferences. Experimental results on various MT systems demonstrate that BTPGBT could provide comprehensive and accurate behavioral testing results for general error diagnosis, which further leads to several insightful findings. Our code and data are available at https: //github.com/wujunjie1998/BTPGBT.

Dialog topic management and background knowledge selection are essential factors for the success of knowledge-grounded open-domain conversations. However, existing models are primarily performed with symmetric knowledge bases or stylized with pre-defined roles between conversational partners, while people usually have their own knowledge before a real chit-chat. To address this problem, we propose a dynamic knowledge graph-based topical conversation model (DKGT). Given a dialog history context, our model first builds knowledge graphs from the context as an imitation of human’s ability to form logical relationships between known and unknown topics during a conversation. This logical information will be fed into a topic predictor to promote topic management, then facilitate background knowledge selection and response generation. To the best of our knowledge, this is the first attempt to dynamically form knowledge graphs between chatting topics to assist dialog topic management during a conversation. Experimental results manifest that our model can properly schedule conversational topics and pick suitable knowledge to generate informative responses comparing to several strong baselines.