Denis Maurel

Finite-state transducer cascades to extract named entities in texts

A lot of Named Entity Extraction Systems were created in English thanks to the impulse of MUC conferences. This article describes a Finite-State Transducer Cascade for the extraction of named entities in French journalistic texts. Finite-State Cascades are widely used for Natural Language Processing: a cascade is a series of finite-state transducers applied to a text transforming it. Such transducer cascades allow implementation of syntactic analysis, translation memory and information extraction. We present our general system named CasSys: this system uses the INTEX natural language processing features to realize a transducer cascade. CasSys is not dedicated to the extraction of named entity; we use it for this task but thanks to Intex, it allows syntactic analyses, information extraction or other tasks.

Description of a Multilingual Database of Proper Names

This paper deals with a classification of proper names which has been chosen as a structure for an electronic multilingual dictionary. This dictionary is not only a list of word, but also a relational data base: we present the relations between proper names that justify this choice.

Direct Construction of Minimal Acyclic Subsequential Transducers

This paper presents an algorithm for direct building of minimal acyclic subsequential transducer, which represents a finite relation given as a sorted list of words with their outputs. The algorithm constructs the minimal transducer directly - without constructing intermediate tree-like or pseudo-minimal transducers. In NLP applications our algorithm provides significantly better efficiency than the other algorithms building minimal transducer for large-scale natural language dictionaries. Some experimental comparisons are presented at the end of the paper.

Psuedo-minimal transducer

The algorithm that we present here builds an acyclic deterministic finite state machine (automaton or transducer), as each word recognized has a proper element, i.e. a transition or a final state that belongs only to the recognizing path of this word.

The syntactic prediction with token automata: application to HandiAS system

This paper presents a finite-state machine to compute the probability of a word appearance when one knows the left syntactic context. We memorize the token number of words in a dictionary and the token number of syntactic categories on finite-state automata. We compute a word probability with these numbers. If we have not predicted the awaited word , we take into account the first letter for a new prediction, and so on. This system has been implemented on a prototype software for disabled communication aid, called HandiAS and it is a part of the Research project CNHL of the LI, the Computer Laboratory of the Tours University. Copyright 2001 Elsevier Science B.V.

Building automaton on schemata and acceptability tables

This paper presents a lexical finite states automaton to parse French Date Adverbials. To build this automaton, I have developed an original model of representation (Schemata and Acceptability Tables), the computation and use of which I will explain in this paper.

Dynamic Perfect Hashing with Finite-State Automata

Minimal perfect hashing provides a mapping between a set of n unique words and n consecutive numbers. When implemented with minimal finite-state automata, the mapping is determined only by the (usually alphabetical) order of words in the set. Addition of new words would change the order of words already in the language of the automaton, changing the whole mapping, and making it useless in many domains. Therefore, we call it static. Dynamic minimal perfect hashing assigns consecutive numbers to consecutive words as they are added to the language of the automaton. Dynamic perfect hashing is important in many domains, including text retrieval and databases. We investigate three methods for its implementation.

Incremental and semi-incremental construction of pseudo-minimal automata

Pseudo-minimal automata ([1],[2]) are minimal acyclic automata that have a proper element (a transition or a state) for each word belonging to the language of the automaton. That proper element is not shared with any other word, and it can be used for implementing a function on words belonging to the language. For instance, dynamic perfect hashing (e.g. a mapping from n unique words to n consecutive numbers, such that addition of new elements does not change the order of the previous elements) can be implemented using a pseudo-minimal automaton ([3]).

Towards an XML representation of proper names and their relationships

The presented work is a part of the Prolex project, whose aim is the design and implementation of a multi-lingual dictionary of proper names and their relationships. It focuses on the design of a standard XML representation for this kind of information. We first present the main lines of the conceptual model for proper names (a classical Entities / Relationships model), then we report on our experiment in designing an XML schema from this conceptual model. We describe the current resulting schema and discuss its main features.

``Beijing Frowns and Washington Pays Close Attention'' Computer Processing of Relations between Geographical Proper Names in Foreign Affairs

We present results of the project Prolex. The aim of the project is the automated analysis of proper names, especially a description of relations between different proper names in a text. The system currently works with geographical proper names (place names, derived adjectives and names of inhabitants) in French. It consists of a database containing specific types of proper names and relations between the different names. Using these names and relations, the program can group the proper names appearing in a text that may belong together (such as Beijing-Chinese-Pekinese-China; American-United States-Washington). This is done by constructing an association matrix between them and by computing the transitive closure of this Boolean matrix. The method is explained with an example.