| Title | : | Incorporating Domain Knowledge in Multilingual, Goal-Oriented Neural Dialogue Models |
| Speaker | : | Suman Banerjee (IITM) |
| Details | : | Tue, 29 Jan, 2019 2:00 PM @ A M Turing Hall |
| Abstract: | : | There is an increasing demand for goal-oriented conversation systems which can assist users in various day-to-day activities such as booking tickets, restaurant reservations, shopping, etc. A typical knowledge base-driven goal-oriented conversation contains three phases. In the first phase, the user expresses his/her preferences such as cuisine, location, price range, etc in a restaurant reservation task. In the second phase, the bot refers to a knowledge base (KB) and fetches the triples which match the user's preferences. Finally, in the third phase, based on the initial preferences and information from the knowledge base the bot helps the user to arrive at the desired goal (say, reserve a table at a restaurant). The entire conversation thus contains three components: (i) PRE-KB utterances, (ii) KB triples and (iii) POST-KB utterances. Curre nt state-of-the-art models treat these three components as one unified sequence and learn to compute attention over this entire long sequence. Instead, we propose a Sequential Attention Network (SeAN) which separately computes attention scores over these components and then combines them to feed a better context representation to the decoder. Most of the existing datasets for building such conversation systems focus on monolingual conversations and there is hardly any work on multilingual and/or code-mixed conversations. Such datasets and systems thus do not cater to the multilingual regions of the world, such as India, where it is very common for people to speak more than one language and seamlessly switch between them resulting in code-mixed conversations. For example, a Hindi speaking user looking to book a restaurant would typically ask, Kya tum is restaurant mein ek table book karne mein meri help karoge? (Can you help me in booking a table at this restaurant?). To facilitate the development of such code-mixed conversation models, we build a goal-oriented dialogue dataset containing code-mixed conversations in Hindi-English, Bengali-English, Gujarati-English and Tamil-English. While modeling the inputs to a goal-oriented dialogue system, current state-of-the-art models typically ignore the rich structure inherent in the knowledge graph and the sentences in the conversation context. Inspired by the recent success of Graph Convolutional Networks (GCNs) for various NLP tasks such as machine translation, semantic role labeling and document dating, we propose a memory augmented GCN with sequential attention for goal-oriented dialogues. Our model exploits (i) the entity relation graph in a knowledge-base and (ii) the dependency graph associated with an utterance to compute richer representations for words and entities. Further, we take cognizance of the fact that in the case of code-mixed conversations, dependency parsers may not be available . We show that in such situations we could use the global word co-occurrence graph to enrich the representations of utterances. We experiment with the modified DSTC2 dataset and its code-mixed versions and show that our method outperforms existing state-of-the-art methods, using a wide range of evaluation metrics. |
