** Job openings: PhD studentships on meaning variation in NLP **

Utrecht University, The Netherlands

The Natural Language Processing (NLP) group in the Computing and Information Sciences department of Utrecht University (UU) is offering PhD positions in AI / NLP.

Three four-year positions are available as part of the AiNed project “Dealing with Meaning Variation in NLP”, a collaboration between AI and Data Science and the Language Sciences Institute led by Prof. Massimo Poesio. The overall aim of the project is to allow NLP models to make better sense of variations in the ways that different speakers and readers interpret language. Positions are available to of the following projects:

PhD PROJECT 1: Variation in coreference and reference

Early research on learning from data with disagreement in Natural Language Processing (NLP) was often motivated by findings about anaaphoic reference - it turns out that often people disagree on what these pronouns mean, particularly in conversations. Methods for learning from data with disagreements (`learning from crowds’) have been successfully applied to other types of data containing disagreements, and substantial data sets containing multiple judgments on anaphoric reference now exist. But computational models of referring expression interpretation that can effectively learn from such data sets do not yet exist. Training co-reference models ‘from crowds’ has proven to be challenging, and there is no consensus over the question of how to test/evaluate interpretation models that take variation into account. This project will focus on addressing such challenges. It will also develop metrics that do justice to interpretative variation for co-reference, and use these metrics to test models. Ideally, the development of these metrics will be informed by cognitive and behavioural evidence on the processing of reference.

For this project, we are looking for a motivated researcher with a Master’s degree in Artificial Intelligence, Deep Learning, Computational Cognitive Science, Computer Science, Linguistics, or Statistics. A good mastery of deep learning and of NLP is essential. An understanding of coreference and discourse understanding would be a definite bonus.

PhD PROJECT 2: Subjectivity in the detection of problematic language

Variation in interpretation is particularly frequent with judgments that depend on an individual’s subjective biases, such as deciding whether a joke is funny or not. This PhD project focuses on NLP methods for subjective interpretive tasks with a high societal relevance, such as offensive/abusive language detection, used e.g., by social media platforms to identify cases of problematic use of language that can be harmful to people. Judgments on whether a given utterance is problematic are notoriously subjective, where differences between judges can have difficult cultural, ethnic, and racial overtones. The project will develop models for detecting problematic language that take into account the fact that the judgments involved can be controversial.

For this project, we are looking for a motivated researcher with a Master’s degree in Artificial Intelligence, Computational Cognitive Science, Computing Science, Computational Social Science, Linguistics, or Statistics. A good mastery of deep learning and of NLP is essential. An understanding of social science methodology would be a definite bonus.

PhD PROJECT 3: Conflicting interpretations in dialogue

In conversations, we produce language under time pressure. One of the effects of this time pressure is that less attention is paid ensuring that expressions can be interpreted univocally, resulting in misunderstandings that often go undetected. Such misunderstandings between dialogue partners cause problems for all aspects of NLP research. The first problem is that specifying that an expression was interpreted in one way by one participant and in another way by the other participant is difficult with present annotation methods. In turn, this makes it difficult to train models that can produce participant-specific interpretations and/or recognise disagreements in interpretation. In this project you will study misunderstandings in dialogue and how conversational agents can recognise and resolve them.

For this project, we are looking for a motivated researcher with Master’s degree in Artificial Intelligence, Computational Cognitive Science, Computing Science, Conversational Agents, Linguistics, or Statistics. A good mastery of deep learning, of dialogue, and of conversational agents is essential.

FOR MORE INFORMATION AND TO APPLY

Further information about these vacancies can be found at:

Project 1: https://www.uu.nl/en/organisation/working-at-utrecht-university/jobs/phd-pos... Project 2: https://www.uu.nl/en/organisation/working-at-utrecht-university/jobs/phd-pos... Project 3: https://www.uu.nl/en/organisation/working-at-utrecht-university/jobs/phd-pos...

The deadline for application is December 3rd 2023. We’re looking for someone to start as soon as possible after the recruitment process is concluded but we understand that it will normally take a few months before the candidate will be ready to start.

Applications should be made through the University's site (see links above).

CONTACTS

For further information, please contact: - Prof. Massimo Poesio (m.poesio AT uu.nl) (all projects) - Project 1: Prof. Yoad Winter (y.winter AT uu.nl) - Project 2: Dr. Dong Nguyen (d.p.nguyen AT uu.nl) or Prof. Antal van der Bosch (a.p.j.vandenbosch AT uu.nl) - Project 3: Prof. Albert Gatt (a.gatt AT uu.nl) or Dr. Denis Paperno (d.paperno AT uu.nl)