Question Answering: A Survey of Research, Techniques and Issues

1. Introduction

The World Wide Web has become a popular place and information retrieval systems have become popular medium to access information throughout the world today. However, most of the modern IR systems only help in locating the relevant document and user himself has to investigate further within the ordered list of documents to extract the useful information. For e.g., the question “Who is the Chief Justice of India?” should get back the response “Rajendra Mal Lodha” but instead the user is presented with a list of relevant documents to explore the accurate answer. Thus, QA systems are recognized as a capability with great potential to present user with exact string of response instead of list of documents. A QA interface simulates man machine interaction and provides answers to user question in natural way.

Though automatic question answering will definitely be a significant advance in the state-of-art information retrieval technology in forthcoming years but still there are many challenging issues that are yet to be resolved. The task of understanding the natural language question and deducing its precise meaning for appropriate answer retrieval faces following issues:

• A.

  Question Classification: Since the advent of QA systems, question classification has remained a separate research problem in itself. Though, lot of efforts has been made for accurate question classification but its success is restricted mainly to factoid questions. When a user inputs a question to its interface, it has become necessary for the QA system to assess the question class accurately to further extract accurate answer, for instance the question, “Who has won Cricket World Cup in 1984 and 2011?” requires name of a country instead of a person name as an answer.

• B.

  Formulation of Right Query from Question: Actually this problem could be regarded as a common name to the issues like ambiguity resolution, semantic symmetry detection, identification of temporal relationships and co reference resolution. For example, the question “Who has won maximum gold in Olympics 2012?” has most obvious reformulation as the person who won maximum individual gold medals in Olympics 2012. But the question has one more interpretation as the country who won maximum gold in Olympics 2012.

• C.

  Knowledge Reuse: This problem could be regarded as a sub-problem of the above problem as it is related with issue of data (information resource) and the reuse of the knowledge that has already been extracted from the web. A QA system must have capability to identify that the questions “When did India attain independence?” and “In which year India got independence?” is associated with identical set of information.

• D.

  Answer Granularity: This is an issue related to the assessment of relevance for correct answer. For instance, consider a question “How many students qualify NET every year from BBAU?” The answer extraction module finds two answer segments, one as ‘More than 50 students qualify NET every year form BBAU’ and other as ‘around 100 students qualify NET every year from different departments of BBAU’, then how does the module be able to select one segment as correct answer.

• E.

  Answer Validation: This is a problem that specifically deals with authenticity of the information resource. E.g., for the question, “How many NASA scientists are Indians?” if answer is fetched from a web resource talking about persistent myths about Indians, then how QA system is going to predict the answer as incorrect.

• F.

  Personalization of Extracted Response: This problem is often associated with the type and presentation of information, since it is likely that people differ in the contents and presentation they prefer while interacting with the QA system. E.g., the answer of the question “How Columbia University could be reached?” expected by a local New York resident is merely different from the user in abroad.