Computational Learning & Computational Linguistics

The Computational Learning and Computational Linguistics group develops techniques to improve our understanding of how linguistic structural and meaning properties are represented in current machine learning models, across many languages.

All speakers can understand a sentence never heard before or derive the meaning of a word or a sentence from its parts. And yet, these basic linguistic skills are still hard to reach by computational models. The current reported success of machine learning architectures is based on computationally expensive algorithms and prohibitively large amounts of data. We develop tasks and data that help us understand their current generalisation abilities and their more complex compositional skills.

We identify higher-level human linguistic abilities as (i) the ability to infer patterns of regularities in unstructured data, (ii) learn from few examples and (iii) develop abstract representations that are valid across, possibly very different, languages. We develop novel structured data that define patterns spanning several sentences with attention to variety of phenomena and languages; we define prediction tasks inspired by human IQ intelligence tests to learn these complex linguistic patterns and paradigms; we develop interpretative models that can identify the generative components of these data. We focus on verbs and their argument structure, the building blocks of any language, expressing core events and actions.

The larger contribution of our research lies in tackling a mixture of language tasks and abstract rule learning and reasoning that takes us closer to investigations of human linguistic intelligence.