Postdoctoral Position in Speech Language Modelling

Max Planck Institute; Nijmegen; Remote; EUR 64,000 - 78,000; Part time

Benefits offered by this job

0.8 FTE position

TVöD E13 salary

30 holidays per year

Enrolment in pension scheme

Job summary

The Max Planck Institute for Psycholinguistics invites applications for a part-time Postdoctoral Position in Speech Language Modelling within the Language and Predictive Computation group led by Dr. Micha Heilbron.

The successful candidate will take a leading role and may supervise master students. The 0.8 FTE, funded for 1 year, offers a German TVöD E13 salary (EUR 5.709,87–EUR 7.025,87 gross per month) plus 30 holidays and enrolment in a personal pension scheme.

Qualifications

• PhD in a relevant discipline; applicants may have submitted thesis or expect to defend before start date.
• Demonstrable experience developing neural models of speech from scratch (not only fine-tuning).
• Strong programming in Python; proficient with PyTorch; experience on GPU clusters.

Responsibilities

• Lead the research line within the LPC group and supervise master students where appropriate.
• Contribute to open science and collaborative dissemination of findings.

Skills

Python

PyTorch

GPU clusters

English proficiency

Education

PhD in a relevant discipline

Job description

Postdoctoral Position in Speech Language Modelling

Post Doc

part-time

We are looking for a postdoc to join the Language and Predictive Computation research group, led by Dr. Micha Heilbron, at the Max Planck Institute for Psycholinguistics.

Language models acquire remarkable linguistic abilities, but are using a pre-defined vocabulary and text pre-segmented in a way that matches this vocabulary. Human listeners face a continuous acoustic stream, and must work out what the relevant units are before they can predict them. Models that learn from speech alone face the same problem, which makes them a unique direct testbed for theories of how linguistic structure can be learned. Current speech models tackle this problem by predicting very low-level acoustic segments, rather than predicting any meaningful linguistic unit. Models that learn from speech alone face the same problem, which makes them a unique direct testbed for theories of how linguistic structure can be learned. Current speech models tackle this problem by predicting very low-level acoustic segments, rather than predicting any meaningful linguistic unit. The research team will consist of the postdoctoral researcher together with Dr Micha Heilbron. The successful candidate will be expected to take a leading role in shaping this line of work within the group, and may also supervise master students working on related projects.

Requirements

What we expect from you:

• A PhD - or equivalent - in a relevant discipline (e.g., computational linguistics, speech technology, AI, cognitive science, computer science, psycholinguistics). Candidates who have submitted their thesis, or expect to defend before the starting date, are welcome to apply.
• Demonstrable experience developing neural models of speech, including training models from scratch rather than only fine-tuning or evaluating existing ones.
• Strong programming skills in Python; proficiency with PyTorch or equivalent ML frameworks; experience working on GPU clusters.
• Experience with methods for analysing model-internal representations (e.g., probing classifiers, representational similarity analysis, discrimination or structural probes), or closely related interpretability work.
• A track record of peer-reviewed publications in a relevant area.
• Excellent written and spoken English.
• The ability to work independently within an interdisciplinary team.

Desirable (traits that would give you advantage)

• Formal training in linguistics or a closely related field - in particular phonetics and phonology - and genuine interest in what models learning from speech can and cannot tell us about linguistic theory.
• Experience designing psycholinguistically controlled stimulus sets or experimental paradigms to test hypotheses about model behaviour.
• Experience relating model representations to human behavioural or neural data (EEG, MEG, fMRI or intracranial recordings), or a clear interest in developing this.
• Experience with large-scale or distributed model training on HPC infrastructure.
• An explicit interest in, and commitment to, Open Science - including the release of code, trained models and intermediate checkpoints.
• Interest in supervising students and in contributing to the wider research community (e.g. workshops, tutorials, outreach).

What we offer you

• A 0.8 FTE position fully funded for 1 year. The intended starting date is October/November 2026.
• Salary according to the German TVöD (Tarifvertrag für den öffentlichen Dienst) in the salary group/level E13: depending on the experience of the applicant, between EUR 5.709,87 and EUR 7.025,87 gross per month, based on full time employment, excluding an 8% holiday bonus.
• 30 holidays per year, based on full-time employment; in addition, we honour both Dutch and German public holidays.
• Enrolment in a personal pension scheme to which both employer and employee pay a monthly contribution.

The deadline for applications is 28 August.

For questions and informal enquiries, please contact Dr. Micha Heilbron ([e-mail] ).

About our institute

The Max Planck Institute for Psycholinguistics is a world-leading research institute devoted to interdisciplinary studies of the science of language and communication, including departments on genetics, psychology, development, neurobiology and multimodalityof these fundamental human abilities.
We investigate how children and adults acquire their language(s), how speaking and listening happen in real time, how the brain processes language, how the human genome contributes to building a language-ready brain, how multiple modalities (as in speech, gesture and sign) shape language and its use in diverse languages and how language is related to cognition and culture, and shaped by evolution.
We are part of the Max Planck Society , an independent non-governmental association of German-funded research institutes dedicated to fundamental research in the natural sciences, life sciences, social sciences, and the humanities.

The Max Planck Society is an equal opportunities employer . We recognise the positive value of diversity and inclusion, promote equity and challenge discrimination. We aim to provide a working environment with room for differences, where everyone feels a sense of belonging. Therefore, we welcome applications from all suitably qualified candidates.

About the Language and Predictive Computation (LPC) group

How do we understand language? What does the brain actually do when we read or hear a sentence? And how is it that Large Language Models - trained on nothing but predicting the next word - have not only mastered language, but also turned out to be the most accurate models of human brain responses to language?

In the LPC Group, we use the tools of modern AI to model language in the human mind and brain. We also run the logic in reverse: drawing on what psychology and neuroscience teach us about human language processing, we build language models constrained by the human cognitive architecture.

Ultimately, we aim to understand how the human brain learns and represents language, and to build more cognitively faithful models of human language processing.