Sara Stymne

I am a Postdoctoral Research Fellow at Computational Linguistics and Language Technology, Department of Linguistics and Philology, Uppsala University since 2012.

I was a researcher at the Department of computer and information science at Linköping University until 2012. I received a PhD in Computational Linguistics from Linköping University in 2012, with the thesis Text Harmonization Strategies for Phrase-Based Statistical Machine Translation. I received a Licentiate degree in Computational Linguistics in 2009, and a Master's degree in Cognitive science in 2006, both from Linköping University.

I spent the autumn 2010 and spring 2009 at Xerox Research Centre Europe in Grenoble, France.

Research interests

My main research interest is machine translation. I am interested mainly in statistical translation, but also in hybrid methods.

My current project is called Efficient Algorithms for Natural Language Processing Beyond Sentence Boundaries, with the main focus on statistical machine translation. Contrary to standard SMT techniques we try to incorporate information from full documents rather than from single sentences into the translation process, with the goal of handling discourse phenomena such as lexical cohesion and pronominal anaphora. The project is funded by eSSENCE - The e-Science Collaboration.

For my PhD I mainly investiged how text harmonization can improve phrase-based statistical machine translation (PBSMT). The main idea is to transform one of the two lanugages to look more similar to the other, for instance with regard to compounds, word order, or definiteness.

I have also worked with error analysis and eye tracking for evaluating MT.

Software and resources

Blast is a tool for error analysis of machine translation output.

In my earlier work I investigated verb frame divergences between Swedish and English and how to handle these in an MT-system. In order to do this I developed a bilingual HPSG grammar for Swedish and English which can be used both for parsing and generation.