This is the home page of the LCG TMR network.Learning Computational Grammars (LCG) is a project funded by the EC Training and Mobility of Researchers programme. The project began 1 April 1998 and will run for three years.
The goals of the network are to explore machine learning techniques for natural language. These will include symbolic learning, statistical training and methods of neuro-computing. Their will be a focus on comparison of results based on attempts to learn noun phrase structure in English. Work on French and German is also possible. See the project proposal for a more detailed sketch.
The programme will apply different machine learning techniques to a set of shared tasks, viz. learning noun phrase (NP) structure from annotated textual input. The task involves recognizing NP's in text, especially their boundaries; recognizing the hierarchical tree structure within NP's; recognizing the essential grammatical roles such as head noun, modifiers and determiners (if any); recognizing the semantic roles of constituents.
The following machine learning paradigms are interesting to the LCG group: memory-based learning, data-oriented parsing and data-oriented structure processing, connectionist learning, inductive logic programming, stochastic feature grammars (and the problem of estimated feature values), explanation-based learning, transformation-based error-driven learning (based on Brill's "string transformations), genetic algorithms, semantic-driven learning, probably approximately correct learning. Postdocs may make further suggestions.
LCG is particularly interested in the degree to which the different learning techniques benefit from initial knowledge.
Please note that not all of the machine learning approaches can be tackled immediately, and that several sites have preferences among the machine learning approaches. All postdocs will be expected to contribute to common needs such as utility programs for data annotation, etc.
For further information about any of the member labs and the range of
activities there, please contact the local coordinators.
There are local preferences for some topics over others, but there is also some choice.
If you are interested in applying, send your CV, the names and addresses of two referees, and a 2-3 pp. sketch of your interest in LCG to the address below. You can do this by email if you wish. Indicate which approaches (see the detail document) and which labs interest you, and when you expect to be available. In the sketch please explain how this is related to work you have done and what special expertise you bring to the problem.
SRI, Cambridge has described its job more exactly in a separate note.)