Jeremy Jancsary

Researcher at the Language Technology (LT) Group
Austrian Research Institute for Artificial Intelligence (OFAI)

Freyung 6/3/1a
A-1010 Vienna
Austria

Phone:
Fax:
Email:
Web:

(+43-1)5324621-5
(+43-1)5324621-9
firstname.lastname@ofai.at
http://www.ofai.at/~jeremy.jancsary/

News

• I am the program chair of KONVENS 2012, to be held in Vienna, Austria from September 19-21. Please consider submitting you paper!
• On April 18^th, I will commence my internship at Microsoft Research Cambridge, working with Sebastian Nowozin and colleagues. Please note that I will be physically unavailable in Vienna until July 11^th; you can contact me via email meanwhile.
• Jointly with Friedrich Neubarth and Harald Trost, I organize the First Workshop on Algorithms and Resources for Modelling of Dialects and Language Varieties, to be held in conjunction with EMNLP 2011 in Edinburgh, Scotland. Please consider submitting your paper!

Research Interests

Natural language processing, graphical models, structured prediction, convex optimization.

About Me

I am currently a researcher at the Language Technology (LT) group of the Austrian Research Institute for Artificial Intelligence (Ofai). Previous work experience includes a position as an embedded software engineer at Inso and a spin-off thereof, Rise, where I contributed to firmware development for the Austrian electronic healthcare card reader (e-Card). Prior to that, I was a teaching assistant (Tutor) at the Compilers and Languages group of Vienna University of Technology.

I received my bachelor's degree in Software & Information Engineering from Vienna University of Technology in 2005 and completed a master's program in Software Engineering & Internet Computing in February of 2008. I am now enrolled in a PhD program. My advisors are Harald Trost and Gerald Matz.

The overarching goal of my research is to develop methodological advances that solve real-world problems in information processing and data analysis. A lot of these problems are popping up in natural language processing these days, so it is an ideal playground. As one of our principal means of communication, language is also a fascinating topic in itself. Turns out many tasks in natural language processing (and a variety of other areas, such as digital communications and protein design) can be described very accurately using graphical models. While extremely versatile and powerful conceptually, most computations in graphical models are NP-hard. For this reason, approximation algorithms are indispensable. Early algorithms in this field work empirically well, but are not yet completely understood and often have serious drawbacks such as non-convergence. So I currently try to improve my understanding of these issues.

Publications

Please see the dynamically created list of current publications, or check my profile on Mendeley.

You may also be interested in my master's thesis:

Recognizing Structure in Report Transcripts
An Approach Based on Conditional Random Fields (CRFs)

Projects

I'm currently involved in the following projects at OFAI:

• Machine Learning Methods for Modeling of Language Varieties (starting 01/2011, funded by WWTF)
• Magnificent - Multifaceted Analysis of News Articles for Intelligent User- and Context-Sensitive Presentation (funded by FIT-IT)

Previously, I was a researcher in the Inspiration project, the aim of which was to provide intelligent speech interpretation through text transformation. Prior to that, I participated in the Sparc project, where we worked on automatic reconstruction of dictations by applying semantic and phonetic knowledge. I was also involved in Also, which was concerned with automatic learning and specific adaptation of speech recognition.

Software

I am the author of VieCRF, a fast toolkit for Factorial Conditional Random Fields. The current version of VieCRF is 0.1.2.

UNIX users (or Linux, or Mac OS X):

• Get the source code to the core library: libviecrf-0.1.2.tar.gz;
• Get the source code to the Perl bindings and the tools: VieCRF-0.1.2.tar.gz;
• Follow the instructions.

Windows users:

• Be aware that VieCRF is mostly developed and tested on UNIX-like platforms;
• I prepared a package for ActiveState's Windows port of Perl;
• Point ActiveState's package manager to this directory and install the VieCRF package;
• Let me know if you encounter any problems.

Browse the documentation of the viecrf command line script to get started (UNIX users: man viecrf, Windows users: use ActiveState's documentation browser or perldoc viecrf on the command line).

Eventually, VieCRF will get its own web site (once I find some spare time).