Within the context of the ERC Consolidator Grant project TENACITY (Tensors and Neural Networks for Computational Creativity), funded by the European Research Council, the Centre for Computational Linguistics (CCL), part of the ComForT research unit at KU Leuven, seeks to hire a postdoctoral researcher to work on the integration of latent factorization models and neural language models for creative language generation.
Creativity pervades human language use, yet the mechanisms behind it are poorly understood, and current language technology shows little true creativity. The ERC project TENACITY develops computational models of language that exhibit creativity, along two complementary lines: interpretable tensor factorization models that induce latent semantic structure from multi-way word co-occurrences, and constraint-based neural language models whose generation process is steered towards creative output.
This postdoctoral position aims to bring both lines together into one integrated framework. The key research question is how the interpretable latent structure induced by factorization models can inform the latent representations of neural language models, for instance through variational inference methods in which the network’s latent variables are tied to the latent distribution of the factorization approach. A proper control and understanding of these latent variables would allow them to be exploited for creative generation, such as the generation of text within a particular style. This is a challenging, high-risk/high-gain topic, and the successful candidate will have considerable freedom to shape the research according to their own expertise and vision.
Website unit
Responsibilities
- Develop variational inference methods for transformer-based language models, informed by the latent structure of tensor factorization models.
- Investigate the interpretability of the resulting latent space, and its exploitation for creative language generation (metaphor, style).
- Collaborate closely with, and provide day-to-day guidance to, the project’s PhD students (working on tensor factorization and constraint-based neural generation).
- Publish in top-level natural language processing and machine learning venues, and present the project’s results at international conferences.
Profile
- You hold a PhD in natural language processing, machine learning, computational linguistics, or a related field, with peer-reviewed publications in the domain.
- You have a strong background in deep learning for natural language processing; experience with latent variable models and/or variational inference is a strong plus.
- You have excellent programming skills (Python, PyTorch) and experience with large-scale experimentation on GPU infrastructure.
- You have experience in guiding junior researchers, or the ambition to develop such experience.
- You exhibit excellent proficiency in English and good oral and written communication skills.
Offer
- We offer a full-time postdoctoral position for 2 years.
- You will be able to conduct scientific research within a high-level, ERC-funded research environment, with access to ample computing infrastructure and funding for conference travel.
- You will have considerable research freedom to orient the research based on your own background and vision.
- The position is research-only, without teaching obligations.
Interested?
For more information please contact Prof. dr. Tim Van de Cruys, mail: [email protected].
KU Leuven strives for an inclusive, respectful and socially safe environment. We embrace diversity among individuals and groups as an asset. Open dialogue and differences in perspective are essential for an ambitious research and educational environment. In our commitment to equal opportunity, we recognize the consequences of historical inequalities. We do not accept any form of discrimination based on, but not limited to, gender identity and expression, sexual orientation, age, ethnic or national background, skin colour, religious and philosophical diversity, neurodivergence, employment disability, health, or socioeconomic status. For questions about accessibility or support offered, we are happy to assist you at this email address.
