ALTA - Spoken Language Processing Technology Research

ALTA Spoken Language Processing Technology (SLPT) Project

[ Overview | News | Personnel ]
[ Presentations | Publications | PhD Theses]

Speak & Improve Challenge 2025

8 teams took part in the Speak & Improve Challenge 2025 on ASR, free (open) monlogic speaking assessment, spoken grammatical error correction (SGEC) and SGEC feedback. Come to the ISCA SLaTE 2025 Workshop to find out about how they did and what approaches they took to improve spoken language assessment and feedback.

The Challenge was run with Cambridge English from Cambridge University Press & Assessment (CUPA). Participants used a pre-release of the Speak & Improve dataset from CUPA. This dataset is derived from the Cambridge English Speak & Improve L2 (second language) English speaking practice tool and contains annotated recordings of a wide variety of L2 English learner speech on open (spontaneous) speaking tasks.

Latest News

We have a new paper on arXiv that explores spoken grammatical error correction (GEC) for use in assessment and to provide feedback on L2 English speaking to learners and teachers. This work focuses on End-to-End Spoken Grammatical Error Correction
We presented our work at the UK and Ireland Speech Meeting 16-17 June, York, UK. Posters on improving spoken grammatical error correction, assessing L2 oral proficiency with speech LLMs and with zero-shot Natural Language-based Assessment, the Speak & Improve Corpus and Challenge 2025 and evaluating synthetic articulation (useful for giving pronunciation feedback). Chloe Yiu gave a talk on bias and fairness in automatic spoken language assessment.
Kate Knill presented the work of Mengjie Qian. Chloe Yee Man Yiu and herself on bias and fairness in automatic free (open) monologic speaking assessment at the EDI/AI EALTA SIG Meeting, Salzburg, Austria.
Paper on exploiting the English Vocabulary Profile for L2 word-level vocabulary assessment with LLMs has been accepted at BEA 2025 Workshop.
Papers on spoken grammatical error correction, assessment of L2 oral proficiency using Speech LLMs and training articulatory inversion models for inter-speaker consistency have been accepted for presentation at Interspeech 2025
Paper on uncertainty estimation for comparative assessment accepted at UAI 2025
Paper on ASR error correction has been published in IEEE Transactions on Audio, Speech and Language Processing
Our paper on Cross-Lingual Transfer Learning for Speech Translation was presented at NAACL 2025
Our work was featured in the main program at COLING 2025 with our paper on fine-tuning LLMs for comparative assessment and Dr Stefano Banno was a co-presenter of the tutorial on LLMs in Education: Novel Perspectives, Challenges, and Opportunities
Adian Liusie defended his PhD on mitigating and analysing bias in LLMs

Openings

We do not have any open positions currently.

Overview

The Cambridge Institute for Automated Language Teaching and Assessment (ALTA) is a virtual institute that brings together researchers from Cambridge University Press & Assessment and the departments of Computer Science and Technology and Engineering. Using techniques from Machine Learning and Natural Language Processing, ALTA's primary objective is to develop technology to enhance the experience of language learning and to develop cutting-edge approaches to automated assessment that will benefit learners and teachers worldwide. It forms part of Cambridge University's Language Sciences Interdisciplinary Research Center.

Sponsorship from Cambridge Assessment English supports a team of PhD students and post-doctoral researchers who investigate a range of topics, including:

language assessment;
teaching feedback;
adaptive learning;
content creation and calibration.

The institute comprises researchers at the Department of Computer Science and Technology (CST) and the Department of Engineering (CUED), organised into a consortium of projects. The CUED project is:

Spoken Language Processing Technology (SLPT)

The ALTA research in the CUED Speech Group is focused on spoken language assessment and learning. The group works on all aspects of this area from core speech recognition and machine learning, to spoken language assessment, to feedback and the nature of the test.

Try it out: practise speaking English with Speak&Improve

Please email Kate Knill kmk1001 @ cam.ac.uk if you have any questions.

top