Whoami Research interests Current projects Publications Reusable code Personal

 

 

Whoami

I am a post-doc at the Institute for Logic, Language and Computation, ILLC of the University of Amsterdam, where I have just completed my PhD with a dissertation entitled "The Neural Basis of Structure in Language". Before that, I obtained my masters degree in cognitive science at the same university, and long before that, almost forgotten, a masters degree in physics.
The easiest way to reach me is at gideonbor_AT_gmail_DOT_com. The address of the institute where I work is
Institute for Logic, Language and Computation (ILLC)
University of Amsterdam
Science Park 904
1098 XH AMSTERDAM The Netherlands

Research interests

My main interest is in the interface between cognitive science, logic, philosophy of language and linguistics. These different disciplines meet around the question of learning, which I view as the active construction of a mental world model. From this perspective I am trying to investigate the acquisition and neural representation of concepts, their gradual organization within a network structure, and the neural correllates of the processes of categorization and generalization, which are fundamental for learning. In particular I believe that the study of language and grammar acquisition offers a window on the mechanisms underlying categorization and abstraction, and helps understanding information processing in other cognitive domains.

Current projects

My research is aimed at the integration of cognitive science and computational linguistics. More specifically, I am part of the VICI project of Professor Rens Bod, where I hope to contribute by investigating into the cognitive foundations of the Data Oriented Parsing (DOP) model. This should eventually lead to a computational model of speech recognition. Apart from Professor Rens Bod I am supervised by Dr Jelle Zuidema.
My current work is inspired by a model of the neocortex, the so-called Memory Prediction Framework, which was proposed by Jeff Hawkins in his book On Intelligence. I have extended the framework to make it suitable for language processing, and for syntactic representations. This resulted in a connectionist model that gradually learns a topology of syntactic categories and rules from the distribution inherent in a corpus.
For my masters thesis I worked on unsupervised grammar induction, within the paradigm of Bayesian Model Merging (Stolcke and Omohundro, 1994), which I evaluated on realistic corpora. The best results were obtained in the semi-unsupervised case, with the task of unsupervised labelling. A thorough description of the formal model, and some philosophical reflections on language acquisition can be found in my masters thesis. A paper on the language philosophical aspects of the Talking Heads experiment of Luc Steels can be downloaded here: partI and part II

Publications

  • My Ph.D. dissertation: The Neural Basis of Structure in Language (ILLC Dissertation Series DS-2011-11). pdf (6 MB) / pdf with cover page (9 MB)
  • Gideon Borensztajn and Jelle Zuidema (2011), Episodic grammar: a computational model of the interaction between episodic and semantic memory in language processing. Proc. CogSci 2011 (pdf)
  • Gideon Borensztajn, Jelle Zuidema and Rens Bod (2009), The hierarchical prediction network: towards a neural theory of grammar acquisition. Proc. CogSci 2009 (pdf)
  • Gideon Borensztajn, Jelle Zuidema and Rens Bod (2009), Children’s grammars grow more abstract with age – Evidence from an automatic procedure for identifying the productive units of language, TopiCS 1 (1), pages 175-188 (pdf)
  • Gideon Borensztajn (2007), Bayesian Model Merging for Unsupervised Constituent Labeling and Grammar Induction
    Technical report, ILLC (pdf)

Talks

  • Grammar acquisition as the construction of a memory system. Talk for the First Amsterdam Workshop on the Neural Basis of Structure in Language, 2011. UvA, Amsterdam, The Netherlands (ppt / workshop poster (3MB))
  • The episodic hierarchical prediction network. Master course "Cognitive models of language and beyond", 2011. UvA, Amsterdam, The Netherlands (ppt)
  • A neural network that learns graded syntactic categories and rules. Summer school on Embodied Language and Construction Grammar, 2009. Cortona, Italy (ppt)
  • The hierarchical prediction network: towards a neural theory of grammar acquisition. Invited talk, 2009. Potsdam, Germany (ppt)
  • Children’s grammars grow more abstract with age. Cognitive Science Conference 2008. Washington, USA (ppt)
Popular press

  • Gideon Borensztajn, Jelle Zuidema and Rens Bod (2008), Taalverwerving. De wetenschap van de kleine wetenschapper
    Kunst en Wetenschap (dutch magazine on art and science) (pdf)
  • Linda Welther (2008), Taalrevolutie
    Didactief (dutch magazine for educational institutions) (pdf)
  • Maurits Martijn (2008), Taaltheorie onderuit
    Vrij Nederland (dutch weekly magazine) (pdf)

Reusable source code

The source code for the complete BMM algorithm can be downloaded here

The source code for the semi-supervised BMM labelling algorithm can be downloaded here

The source code for the DOP parser can be downloaded here

Personal

I used to like painting. Here are some of my selected works. (By the way, the background images are borrowed from my favourite painter, Jan Sluijters.) Since I am doing a PhD I stick with photography, which takes considerably less time. You can see my impressions of beautiful Amsterdam in the snow here.