Translating Web Search Queries into Natural Language Questions

Translating Web Search Queries into Natural Language Questions Adarsh Kumar, Sandipan Dandapat, Sushil Chordia AI & Research, Microsoft Hyderabad, India {adkuma, sadandap , sushilc}@microsoft.com Abstract Users often query a search engine with a specific question in mind and often these queries are keywords or sub-sentential fragments. In this paper, we are proposing a method to generate well-formed natural language question from a given keyword-based query, which has the same question intent as the query.Conversion of keyword based web query into a well formed question has lots of applications in search engines, Community Question Answering (CQA) website and bots communication. We found a synergy between query-to-question problem with standard machine translation (MT) task. We have used both Statistical MT (SMT) and Neural MT (NMT) models to generate the questions from query. We have observed that MT models performs well in terms of both automatic and human evaluation. Keywords: Natural Language Generation, Machine Translation, NLP 1. Introduction Search engines have improved a lot in last decade in all as- pects. Earlier, the primary task of a search engine was to extract most relevant links for the query and present them as results. Lately, instead of just giving relevant links re- lated to the query, search engines are trying to directly an- swer to any question asked. For example, for the query “japan’s capital” in modern search engines (eg. Bing and Google) directly answer “Tokyo”, instead of providing a link containing the answer. Thus, search engines are evolv- ing to save time for users and increase their productivity. To further enhance the user-experience and increase pro- ductivity, search engines apart from showing the answer for a particular question, are trying to show related ques- tions, to help users in their exploration. For example, for the query “fever symptoms”, user mostly wants answer to the question “What are the symptoms of fever?” and for the same query, questions like “How do you treat fever?”, “What causes high fever?” are highly related. To show re- lated questions, search engines need to have a well framed question corpus from which they can extract relevant ques- tions given a query. (White et al., 2015) have shown that more than 10% of queries issued on a search engine has question intent whereas only 3% of them are formulated as natural language questions. Most of these queries are pri- marily keywords or sentence fragments. Hence, a corpus of questions can not be created directly using the search queries with question intent due to the issue of grammatical correctness and incomplete sentence formation. To over- come this problem, we are proposing a technique to con- vert query with question intent, into a well-formed ques- tion. This technique can be used to generate well formed questions asked by the user, which can be used by search engines. Apart from the direct application in search en- gines, query keywords to question conversion has applica- tions in Question Answering (QA) systems, bots communi- cation, Community Question Answer (CQA) websites etc. In CQA websites, …

Content truncated to 500 words