Ibrahim Bounhas

A hybrid approach for Arabic multi-word term extraction

Building a domain model from a specialized corpus requires identifying candidate terms. It also includes identifying semantic relations between terms. Once this model is constructed it can be used for many tasks of information retrieval. In this process, multi-word terms have a great importance. In the one hand they constitute domain relevant candidate terms. On the other hand syntactic relations that link their constituents can be used to infer semantic relations between terms. In this paper we propose to extract mutli-word terms from Arabic specialized corpora. The proposed approach uses linguistic rules based on morphological features and POS (Part Of Speech) tags to parse documents and retrieve candidate terms. Statistical measures are used to deal with ambiguities generated by the linguistic tools and to rank candidate terms according to their relevance. We present experiments on a corpus from the environment domain. We report high quality results that are confirm the targets set for the precision metric.

ArabOnto: experimenting a new distributional approach for building Arabic ontological resources

Ontologies are useful for modelling and retrieving knowledge in complex information systems. Ontology construction environments use statistical and linguistic information to extract knowledge from corpora. Within the great improvement in this field, there is a need to introduce the Arabic language in these environments. We present the ArabOnto architecture modelling the process of Arabic ontology extraction from corpora. ArabOnto focuses on linguistic issues related to Arabic term extraction and linking (i.e. from morphosyntactic parsing to clustering). We experiment our system by testing several alternatives on three domains. Besides, our ontologies are validated in the context of an information retrieval system.

A Possibilistic Approach for the Automatic Morphological Disambiguation of Arabic Texts

This paper presents a new approach for Arabic non-vocalized texts disambiguation based on a possibilistic classifier. A morphological analyzer provides all the possible solutions and the values of the morphological features of words. When texts are vocalized, the number of solutions is reduced and in many cases, we can identify the correct analysis of the input word. The main idea of this paper is to exploit this type of texts in order to learn contextual dependencies between the different values of morphological features modeled as a possibilistic network. This knowledge is used later to disambiguate non-vocalized texts. In order to evaluate our approach, we perform experiments on a corpus of arabic stories. In this paper, we present results concerning the Part-Of-Speech (POS) which is the main morphological feature. Our results are compared to the SVM-based system called MADA.

Arabic Cross-Language Information Retrieval: A Review

Cross-language information retrieval (CLIR) deals with retrieving relevant documents in one language using queries expressed in another language. As CLIR tools rely on translation techniques, they are challenged by the properties of highly derivational and flexional languages like Arabic. Much work has been done on CLIR for different languages including Arabic. In this article, we introduce the reader to the motivations for solving some problems related to Arabic CLIR approaches. The evaluation of these approaches is discussed starting from the 2001 and 2002 TREC Arabic CLIR tracks, which aim to objectively evaluate CLIR systems. We also study many other research works to highlight the unresolved problems or those that require further investigation. These works are discussed in the light of a deep study of the specificities and the tasks of Arabic information retrieval (IR). Particular attention is given to translation techniques and CLIR resources, which are key issues challenging Arabic CLIR. To push research in this field, we discuss how a new standard collection can improve Arabic IR and CLIR tracks.

Arabic morphological analysis and disambiguation using a possibilistic classifier

This paper proposes and experiments a new approach for morphological feature disambiguation of non-vocalized Arabic texts using a possibilistic classifier. The main idea is to learn contextual dependencies between features from vocalized texts and exploit this knowledge to disambiguate non-vocalized ones. We use possibility theory as a means to model imprecision in the training and testing steps, since the context is itself ambiguous. We also investigate the dependency between various features focusing on the Part-Of-Speech (POS).

Toward an Arabic Ontology for Arabic Word Sense Disambiguation Based on Normalized Dictionaries

In this paper, we propose an approach for constructing Arabic Ontology based on normalized dictionaries. This approach mainly consists in transforming non structured Arabic dictionaries into LMF Lexical Markup Framework based-normalized ones. We are basically exploiting Arabic dictionaries of Hadith for experimentation. Then, from an Arabic normalized dictionary of Hadith, an ontology will be constructed. It represents hidden knowledge in Hadith texts. It will be next integrated into an information retrieval and navigation system. We will take advantage of it to semantically disambiguate Arabic terms of both the formulated user query and/or the Arabic texts with application on texts of Hadith.

A hierarchical approach for semi-structured document indexing and terminology extraction

Many approaches of terminology extraction make use of contextual information to acquire relations between terms. The quality and the quantity of this information influence the accuracy of the terminology extractor. In this paper, we assume that logical structure of documents constitute a rich source of contextual information which can be used to infer semantic relations between terms and thus construct a termino-ontological resource. We propose a top-down indexing method which attributes greatest importance to terms that appear in the head nodes of the document. Terms are weighted according to their position in the hierarchical structure of the document. Once documents are indexed, logical relationships between their fragments are mined to build a contextual network of terms. Links of this network help deduce semantic relations useful for terminology organization. A so extracted knowledge can be exploited as a mapping scheme in a domain-specific information retrieval (IR) system. We experiment our approach by taking the example of an Arabic corpus talking about animals.

Semantic Information Retrieval: A Comparative Experimental Study of NLP Tools and Language Resources for Arabic

In this paper, we try to exploit the semantic richness of Arabic language for Information Retrieval (IR). The semantics of Arabic words may be extracted from dictionaries or corpora, which are analyzed and minded using Natural Language Processing (NLP) and text mining tools. This allows modeling the contextual dependencies between words, which help identify the meaning of queries in the search process. Thus, the queries are enriched by semantic knowledge, which enhances search performance. In this context, this paper describes a text mining-based approach for Arabic semantic IR, which considers senses of query terms. Experiments and results based on a standard Arabic Test collection are discussed through this communication. In the one hand, we compare dictionary versus corpus-based approaches for modeling semantics. On the other hand, we compare some Arabic NLP tools in the preprocessing step. Thus, we study the effect of Arabic morphology on the semantic interpretation of queries.