Explainable and Discourse Topic-aware Neural Language Understanding
This work addresses the need for more explainable and context-aware language understanding in NLP, though it appears incremental by extending existing topic-language model integrations.
The authors tackled the problem of limited document-level context in language models by incorporating both latent and explainable topic representations along with sentence-level topical discourse, resulting in improved performance across tasks like language modeling, word sense disambiguation, and document classification.
Marrying topic models and language models exposes language understanding to a broader source of document-level context beyond sentences via topics. While introducing topical semantics in language models, existing approaches incorporate latent document topic proportions and ignore topical discourse in sentences of the document. This work extends the line of research by additionally introducing an explainable topic representation in language understanding, obtained from a set of key terms correspondingly for each latent topic of the proportion. Moreover, we retain sentence-topic associations along with document-topic association by modeling topical discourse for every sentence in the document. We present a novel neural composite language model that exploits both the latent and explainable topics along with topical discourse at sentence-level in a joint learning framework of topic and language models. Experiments over a range of tasks such as language modeling, word sense disambiguation, document classification, retrieval and text generation demonstrate ability of the proposed model in improving language understanding.