🧑‍🔬 Gülşen Eryiğit

Gülşen Eryiğit works in natural language processing. Their papers here, spanning 2006–2026, cover syntactic parsing, multiword expressions and, more recently, multilingual and multimodal language understanding. A recurring question is how systems should treat structures that resist plain word-by-word analysis: dependencies that cross one another, expressions whose meaning is not the sum of their parts, and references to entities that a sentence leaves unstated. The methods range from transition-based parsers driven by statistical classifiers to the construction of shared benchmarks and the prompting or instruction-tuning of large language models.

The two most cited works come from different strands of this research. Labeled Pseudo-Projective Dependency Parsing with Support Vector Machines was a CoNLL-X shared task entry. It is a deterministic MaltParser system that encodes non-projective arcs in the labels and splits morphological features into atomic attributes. Multiword Expression Processing: A Survey, written within the PARSEME network, proposes a framework that divides the field into two tasks. One is discovery, which collects expressions from corpora into lexicons. The other is identification, which locates their occurrences in running text.

The newest papers bring idioms into the vision-and-language setting and adapt large language models to older tasks. A Parallel Cross-Lingual Benchmark for Multimodal Idiomaticity Understanding extends an English–Portuguese idiom-and-image design to many languages, with expert-produced counterparts for each one. A related shared task entry ranks candidate images with a training-free, chain-of-thought pipeline built on vision–language models. CorefInst instruction-tunes decoder-only models for multilingual coreference resolution, including omitted pronouns in pro-drop languages.

Eryiğit is based at Istanbul Technical University. Co-authors on the papers here include Joakim Nivre, Johan Hall and Jens Nilsson on dependency parsing, and Mathieu Constant, Carlos Ramisch and Aline Villavicencio on multiword expression research.

Gülşen Eryiğit has 5 papers in our library, in 2006–2026, most of them on Idioms & multiword expressions, Multilingual & cross-lingual NLP, Vision & language; often with Adriana Pagano, Aline Villavicencio, Carlos Ramisch.