Spoken Language Systems


L2F was formed in 2001, bringing together several research groups that could potentially add relevant contributions to the area of computational processing of spoken language for European Portuguese. Our long term goal is to create technology to bridge the gap between natural spoken language and the underlying semantic information.

Semantic processing of multimedia contents was one of the first interdisciplinary areas that integrated several core technologies in the group, with important milestones such as automatic captioning of broadcast news, meetings and other multimedia contents. This work led to the creation of our spin off VoiceInteraction, that now develops our in-house recognition and synthesis tools, AUDIMUS and DIXI. Another interdisciplinary area that was also one of our first priorities is spoken dialogue systems, with virtual agents acting as museum guides, robot companions, chatbot pals, and a myriad of other scenarios.

Speech-to-speech translation is the area which encompasses more core technologies in the group. Our current work in machine translation is mainly done in cooperation with Unbabel, a start-up company created by a former PhD student.

Our technologies are regularly tested in international challenges, reaching prominent results in topics such as speech/text analytics, and mining of social networks.

The two most relevant application areas from a societal point of view are e-learning and e-health. Our strong involvement in the Carnegie Mellon | Portugal Phd program has been particularly relevant for our progress in areas such as computer assisted language learning (for foreign students of Portuguese), and translation. In terms of e-health, our work now extends much beyond our initial involvement in the development of alternative and augmentative communication tools. We are strongly committed to the development of virtual therapists for aphasia (our award winning platform), tools for monitoring cognitive decline, and interacting with children with autism spectrum disorder, among many other applications. The potential of speech and language technologies for e-learning and e-health is, in fact, enormous.

Our work now extends to other signals besides speech, such as brain signals, music, and sign languages. It also extends to other languages, but Portuguese (in all its variants) remains the focus in which we continuously strive to create better linguistic tools and resources.

The group has been very active in the organisation of scientific events, being in the genesis of the PROPOR series of conferences (Computational Processing of Portuguese), and organising Interspeech 2005, EMNLP 2015, and IberSpeech 2016. Last but not least, we have been one of the organisers of the Lisbon Machine Learning Summer School that has successfully taken place every July at IST, Lisbon, since 2011.

Group Coordinator

Fernando Manuel Marques Batista

David Manuel Martins de Matos

Group Web Site


Administrative Support

Teresa Mafalda Mascarenhas Alves Mimoso