Cross-Language Information Retrieval on the Web

Introduction

The development of the semantic Web requires great economic and human effort. Consequently, it is very useful to create mechanisms and tools that facilitate its expansion. From the standpoint of information retrieval, access to the contents of the semantic Web can be favoured by the use of natural language, as it is much simpler and faster for the user to engage in his habitual form of expression.

The discipline known as Natural Language Processing (NLP)—a subarea of Artificial Intelligence and Computational Linguistics—proves particularly useful in this context. NLP looks at the problems deriving from the automatic comprehension of natural language. It also focuses on the design of systems and efficient mechanisms that allow for communication between people and machines. Among the diverse spheres of application of NLP we have automatic translating and information retrieval, two of the specific areas that later gave rise to cross-language information retrieval.

The growing popularity of Internet and the wide availability of Webinformative resources for general audiences are a fairly recent phenomenon, although man´s need to hurdle the language barrier and communicate with others is as old as the history of mankind. The World Wide Web, together with the growing globalization of companies and organizations, and the increase of the non-English speaking audience, entails the demand for tools allowing users to secure information from a wide range of resources. Yet the underlying linguistic restrictions are often overlooked by researchers and designers. Against this background, a key characteristic to be evaluated in terms of the efficiency of IR systems is its capacity to allow users to look up a corpus of documents in different languages, and to facilitate the relevant information despite limited linguistic competence regarding the target language. This may call for resorting to translations of the texts involved.

More generally, information retrieval has been known as the automated process by which a user makes a query expressing information needs, and the system responds by providing a specific list of the most relevant documents related to the query. The traditional model assumes that the process of information retrieval always goes through the same series of tasks: (1) the user has an information need, (2) he/she enters a query into an IR system, (3) the system compares the specific query with the representations of the documents stored in the databases, (4) it presents the users the text or those texts of possible interest, and (5) finally, the user examines them and determines the relevance of the retrieved results. Ideally, some or all of the retrieved documents will solve —totally or partially— the need for information. A good IR system should retrieve all the relevant documents (meaning complex coverage), and only the documents that are relevant (precision).

This information retrieval model has some implicit restrictions, however, such as the notion that query and document are written in the same language. In a multilingual environment like the Web, most IR systems or Websearch engines can only locate documents written in the same language as the query word(s), although there are some exceptions. Sometimes these systems feature machine translation (MT) devices that go to work when the documents have already been located, but they do not break the language barrier in the searching process.

While the traditional model is rooted in the approach of monolingual retrieval, it can be broadened to encompass other perspectives: the terms “cross-language information retrieval” or “cross-lingual information retrieval” draw in (at least) two languages. Both entail specific systems that carry out bilingual information retrieval, also known as translingual information retrieval. In them, the query is made in a source language that is different than the language of the corpus of documents, written in a single target language. Moreover, CLIR can be used to refer to multilingual information retrieval where the system, from a query expressed in a specific language and some documents belonging to a multilingual corpus, offers a ranking or a unique list of documents written in different languages, shown depending on their relevance (López-Ostenero, Gonzalo & Verdejo, 2004).