Iason Demiros

A System for Recognition of Named Entities in Greek

In this paper, we describe work in progress for the development of a Greek named entity recognizer. The system aims at information extraction applications where large scale text processing is needed. Speed of analysis, system robustness, and results accuracy have been the basic guidelines for the systems design. Pattern matching techniques have been implemented on top of an existing automated pipeline for Greek text processing and the resulting system depends on non-recursive regular expressions in order to capture different types of named entities. For development and testing purposes, we collected a corpus of financial texts from several web sources and manually annotated part of it. Overall precision and recall are 86% and 81% respectively.

Integrating translation technologies towards a powerful translation Web service

Rapid changes in the global marketplace have given rise to new demands and have provided new opportunities for the translation industry. The need for multilinguality in the presentation and business logic layers of most modern systems, applications and services is a great challenge that the translation industry now faces. But even after many years of intense research and many commercial attempts of related products, translation systems of today still fail to completely meet the above needs. Within this framework, an architecture of a modern automatic translation system exploiting current infrastructure and covering today and future needs is proposed in this paper.

Generating bilingual lexical equivalences from parallel texts

This paper describes the application of artificial intelligence methods for the automatic extraction of translation equivalences from bilingual parallel text. In an attempt to implement as language independent a method as possible, the applications methodology features statistical inductive techniques coupled with symbolic processing techniques catering for the analysis of specific language phenomena. The method presupposes parallel texts and identifies translational equivalences at the word or multiword unit level for those cases that such an equivalence holds true. Parallel texts are first aligned at the sentence level and grammatically analyzed. Noun phrase grammars extract noun phrases, and statistical evaluation yields the most coherent multiword units on either side . Translation candidates of word or multiword units are evaluated by a similarity metric defined by the co-occurrence frequency and independent frequencies of the units . The method has been tested on an English - Greek corpus consistin...

Media Monitoring by Means of Speech and Language Indexing for Political Analysis

ABSTRACT In this article, we describe a media monitoring system that we have developed and implemented for the Secretariat General of Communication and Secretariat General of Information in Greece (SGC-SGI). The system applies emerging technologies for audiovisual recording, speech recognition, language processing, multimedia indexing, and retrieval, all integrated into a large video and audio library that covers broadcast news and current affairs in Greek and English. It assists SGC-SGI in compiling information; annotating and analyzing news; and monitoring national, political, social, economic, cultural, and environmental issues concerning Greece in general.

An analysis of linear weight updating algorithms for text classification

This paper addresses the problem of text classification in high dimensionality spaces by applying linear weight updating classifiers that have been highly studied in the domain of machine learning. Our experimental results are based on the Winnow family of algorithms that are simple to implement and efficient in terms of computation time and storage requirements. We applied an exponential multiplication function to weight updates and we experimentally calculated the optimal values of the learning rate and the separating surface parameters. Our results are at the level of the best results that were reported on the family of linear algorithms and perform nearly as well as the top performing methodologies in the literature.