Holistic Language Processing: Joint Models of Linguistic Structure

Опубликовано 17 августа 2016, 21:04

The natural language processing (NLP) applications which ultimately affect people's daily lives are high level, semantically-oriented ones: question answering, machine translation, machine reading, speech interfaces for robots and machines, and others that we haven't even thought of yet. Humans are very good at these types of tasks, in part because they naturally employ holistic language processing. They effortlessly keep track of many layers of low-level information, while simultaneously integrating in long distance information from elsewhere in the conversation or document. In contrast, much NLP research focuses on lower-level tasks, like parsing, named entity recognition, and part-of-speech tagging. Moreover, for the sake of efficiency, researchers modeling these phenomena make extremely strong independence assumptions, which completely decouple these tasks, and only look at local context when making decisions. This talk will cover multiple aspects of holistic language processing, and describe systems for joint parsing and named entity recognition; named entity recognition which incorporates long-distance information; and multi-task learning over multiple domains, and over multiple datasets with varying amounts of annotated information. These systems are designed to produce analyses which are more consistent, of higher quality, and generally more useful for doing the kinds of tasks that non-researchers actually care about.