Weizmann Institute

Research Projects


home | cv | projects | resources

2013–Present: Semantic Parsing for Scenario-Based Programming
with: Prof. David Harel.

In this project we view the programming task as automatically generating a system based on a verbal description of its behavior. Our task is to parse natural language requirements into executable scenarios, using advanced statistical models for structure prediction.

2009–Present: Parsing Morphologically Rich Languages.
with: Djame Saddeh, Sandra Kubler, Prof. Joakim Nivre.

Most NLP models are developed with English in mind. What happens when applying them to a different language? In this project we investigate, develop and evaluate resources and models for parsing morphologically rich languages (e.g., Hebrew, Arabic, Turkish and more, which are known to be notoriously hard to parse), thus significantly broadening the empirical reach of statistical parsers.

2010–2013: Evaluation Algorithms for Structure Prediction.
with: Prof. Joakim Nivre.

Statistical NLP systems predict complex graph structures that represent different notions of meaning. How can we quantitatively evaluate the correctness of these structures? How can we faithfully compare the resulting structures across frameworks? In this project we develop distance-based algorithms that solve complex evaluation tasks.

2005–2009: Joint Morphological and Syntactic Disambiguation
Supervisors: Prof. Remko Scha, Dr. Khalil Sima'an.

Standard approaches to NLP separate word-level (morphology) and sentence-level (syntax) processing. There is ample evidence that this separation is empirically problematic. In this project we develop statistical morphosyntactic parsing architectures (lattice parsing, RR parsing) and show their superiority with respect to standard pipeline NLP approaches.

2003–2004: Formal Semantics of Tense and Aspect
Supervisor: Prof. Michiel van Lambalgen.

In this project we view the Semitic morphological templates (‘binyanim’) as formal semantic operators, and develop a formal logic that allows us to calculate the aspectual meaning of verbs by applying the operator realized by the template to the verbal class of the root. The project involved empirical evaluation in the realm of language acquisition (ages 3-30).