Improving Question Analysis for Arabic Question Answering in the Medical Domain

Abstract

Question analysis is a basic module in a question answering (QA) system, and its quality affects the performance of QA system. In this paper, we address the problem of Arabic question analysis in the medical domain where several specific challenges are met. The major challenging issue in processing Arabic medical question is the need for ambiguity resolution. Nevertheless, this issue has not been well studied in related works. Our question analysis uses dictionaries and transducers to analyze any medical question, factoid or complex. This module detects important elements of the question, including: different words in the question that identify what the user wants to ask for, and the nature of the expected answer. To identify well these elements a step of disambiguation is applied. Then, the words used in the question will be extended by adding new words that connect semantically to those in the question. Experimentation of the question analysis module of our Arabic medical question answering system show interesting results.