Selection Announcement Topics

There are two topics in this year's Language, Interaction and Computation Selection Announcement:

1. Compositionality in Distributional Semantics (3)

2. The encoding of event knowledge in language resources as a support for the diagnosis and rehabilitation of language disorders (1)

-------------------------------

1. Compositionality in Distributional Semantics

Distributional semantic models that approximate the meaning of words with vectors that record the patterns of co-occurrence of words in large amounts of text have been shown to provide an excellent proxy to meaning in both cognitive and computational tasks (Turney and Pantel, 2010). In recent years, there has been much interest in extending these models to account for the meaning of not only words but phrases and even sentences (Mitchell and Lapata, 2010, Baroni, 2012). This project will tackle one of a set of possible aspects of the design, implementation and evaluation of compositional distributional semantic models. Examples of possible focus areas include the following:

- development of models of composition functions to be applied in distributional space;

- methods to improve the automated estimation of such functions;

- modeling compositional phenomena described in theoretical linguistics in distributional terms;

- creation of evaluation sets for compositional distributional semantic models, and psychological testing of such models.

References:

Jeff Mitchell, Mirella Lapata: Composition in Distributional Models of Semantics. Cognitive Science 34(8): 1388-1429 (2010)

P. D. Turney and P. Pantel (2010) "From Frequency to Meaning: Vector Space Models of Semantics", Volume 37, pages 141-188.

M. Baroni. 2012. Compositionality in distributional semantics. EACL 2012 tutorial. Slides available at:

http://clic.cimec.unitn.it/composes/materials/eacl-2012-compositionality-tutorial-ho.pdf

Contact Marco Baroni: marco[DOT]baroni[AT]unitn[DOT]it

2. The encoding of event knowledge in language resources as a support for the diagnosis and rehabilitation of language disorders

Recent studies have shown that language resources widely used for Natural Language Processing purposes (such as WordNet) can be extended and exploited to support speech therapists in the treatment of patients affected by language disorders (see STaRS.sys and related publications).

Efforts in this direction have so far concentrated on concrete nouns, also because  the treatment of anomic patients, that is patients affected by the inability to retrieve intended words during conversation or in structured language tasks such as naming pictures, are currently focused on nouns.  However a number of studies have insisted on the opportunity of including verbs, and full sentence production tasks, as an important ingredient of the treatment of anomic patients (see  Rymer and Kohen, 2006). This trend goes in parallel with a growing interest of the Human Language Technologies community for the processing of event information, that is the kind of  information typically expressed by verbs.

The candidate will try to give and answer to the following research issues: What kind of event information is relevant for the treatment of  anomia? How is event information represented in existing language resources (e.g. WordNet and FrameNet)? Can this information be exploited to support speech therapists?  Is there a need for extending such information from both a quantitative and qualitative point of view? Can additional information be acquired by extraction from corpora, or through crowd-sourcing techniques?

The candidate will have to interact with language disorder therapists, computational linguists, lexicographers, and possibly patients/subjects. He/she will have a strong background both in cognitive science and computational linguistics. At least some knowledge of the Italian languages is required.

All through the project, the candidate will closely collaborate with members of the Center for Mind/Brain Sciences (CIMEC) of the University of Trento, where active cognitive and computational research on related areas is currently being conducted, and with computational linguists and lexicographers of the Human Language Technology group at Fondazione Bruno Kessler (FBK) in Trento.

REFERENCES

Raymer, A. & Kohen, F. (2006). Word-retrieval treatment in aphasia: Effects of sentence context. The Journal of Rehabilitation Research and Development, 43(3), 367. doi:10.1682/JRRD.2005.01.0028

Vinson, D. P. & Vigliocco, G. (2008). Semantic feature production norms for a large set of objects and events. Behavior Research Methods, 40(1), 183-190. doi:10.3758/BRM.40.1.183

Andrews, M., Vigliocco, G. & Vinson, D. (2009). Integrating experiential and distributional data to learn semantic representations. Psychological review, 116(3), 463-98. doi:10.1037/a0016261

Contact Emanuele Pianta: pianta-at-FBK-dot-eu