Mining search query logs for spoken language understanding

In a spoken dialog system that can handle natural conversation between a human and a machine, spoken language understanding (SLU) is a crucial component aiming at capturing the key semantic components of utterances. Building a robust SLU system is a challenging task due to variability in the usage of language, need for labeled data, and requirements to expand to new domains (movies, travel, finance, etc.). In this paper, we survey recent research on bootstrapping or improving SLU systems by using information mined or extracted from web search query logs, which include (natural language) queries entered by users as well as the links (web sites) they click on. We focus on learning methods that help unveiling hidden information in search query logs via implicit crowd-sourcing.