Language and Cognition Group
The Language and Cognition group carries out basic and applied research in natural language processing, computational linguistics and artificial intelligence. The holy grail of these disciplines is to build intelligent systems capable of processing and “understanding” large volumes of text and spoken data, and communicating in natural language.
Systems which can intelligently process natural languages are useful where the aim is to make sense of textual and spoken data or to judge the quality of this data. Applications include machine translation, sentiment analysis, information retrieval and extraction, speech recognition and synthesis, grammar checking and computer-assisted language learning. These applications are useful in a wide range of sectors including education, finance, market research, media and health.
We focus mainly on linguistically-informed statistical approaches to natural language processing. We specialise in machine translation, multi-modal information retrieval, natural language parsing with a focus on noisy web data, natural language generation, sentiment analysis, question-answering, computer-aided language learning and the World-Wide-Mind. We are a multilingual group, working on a wide range of languages including Arabic, Bengali, Chinese Mandarin, English, French, German, Hindi, Irish, Irish Sign Language, Japanese, Nawat, Spanish and Urdu.
Example Research Projects
- Quality Estimation for Machine Translation: given a translation produced by a Machine Translation system, how can we tell whether: a) it is good enough to publish without post-editing by a human translator; b) it requires post-editing by a human translator; or c) it is so bad that it needs to be translated from scratch by a human translator? The task of QE for MT is learning the quality of a translation, using cues from the source sentence, its translation and the MT system used to produce the translation.
- Aspect-based Sentiment Analysis: simple sentiment analysis engines that report whether the online opinion on a particular topic or entity is positive/negative/neutral, are ten-a-penny. A more useful sentiment engine reports on the sentiment of aspects of the topic/entity. For example, instead of reporting that sentiment on a new hotel is positive, an aspect-based engine might report that people like the price, are unhappy with the service and have no strong opinions about the facilities.
- Robust Syntactic Parsing: syntactic parsing is the process of analysing the syntactic structure of a sentence. The input to a natural language parser is a sentence and the output is a tree or graph structure which depicts how the words in the sentence relate to each other. Parsing is a core task in NLP because knowledge about syntactic structure is extremely useful in higher-level tasks such as machine translation, sentiment analysis and grammar checking. A particular challenge for existing natural language parsing systems is to cope with the anything-goes approach to language found in the online world.
A list of the projects that are currently funded or have recently finished can be found here.