Gabriel Pereira Lopes


Multimodal Systems

Country: Portugal

Affiliation: Faculdade de Ciências e Tecnologia, UNL



I am a member of the MultiModal Systems group, where I lead the Human Language Technologies team. In an earlier stage of my research, Parsing, Reasoning and Knowledge Representation were at the center of my interests. Lately, the areas of Machine Translation, Machine Aided Translation, Information Retrieval and Machine Learning became my main concern. These include: parallel text alignment at sentence and subsentence levels; single- and multi-word term translation extraction; machine classification of extracted word and phrase translations, prior to human evaluation; discovery of bi-lingual stem and suffix translations; suffix translation clustering for generating Out-Of-Vocabulary translation entries; generation of translation engines for several language pairs (including Portuguese, Spanish, English, French, German, Czech, Arab and Mandarin) that distinguish from statistical and hierarchical translating engines and gave rise to the Request of a Provisional National (Portuguese) Patent in August 2015 which was not assigned - our engines compare positively with MOSES, the state-of-art system for statistical machine translation, because they produce better quality translations in several language pairs, where learning has been deeper; word senses clustering, using the context where word occurs and its translation as a semantic tag in parallel sub-sentence aligned corpora; selection of word translations in the context where it appears, using what was learned during word sense clustering; research on data structures for full text indexing and compression and their application to translation retrieval activities. Complementary areas of research include: unsupervised, language independent, multi-words extraction (giving rise to my top 6 most cited articles), clustering and classification of Languages and documents, Monolingual Word Sense Disambiguation (also appearing in the top 5 most cited articles) and automatic thesauri construction. Learned information, classified as correct or incorrect, is used in multiple applications. I assume that a machine must be corrected and learn also from humanly identified errors. I have supervised 16 PhD thesis and co-supervised 1 PhD Thesis, all finished, of Portuguese, Brazilian and Indian PhD students. I also supervised 15 Masters thesis, in Portugal, already finished. And additionally I co-supervised 4 Masters students abroad and in Portugal . I have also supervised 4 Post-doc students from Brazil, France, Bulgary, Spain and worked with Veronica Dahl from Canada while she was staying for her Sabatical with a Gulbenkian Professorship. I have coordinated 31 projects: 5 European projects, 3 National projects of cooperation with industry, 14 National projects and 9 projects of international cooperation. In July 2013, I have launched a start-up ISTRION BOX -TRANSLATION AND REVISION, LDA. at Parkurbis, Covilhã.