On Using Distribution-Based Compositionality Assessment to Evaluate Compositional Generalisation in Machine Translation

Anssi Moisio, Mathias Creutz, Mikko Kurimo

Research output: Chapter in Book/Report/Conference proceedingConference article in proceedingsScientificpeer-review

16 Downloads (Pure)

Abstract

Compositional generalisation (CG), in NLP and in machine learning more generally, has been assessed mostly using artificial datasets. It is important to develop benchmarks to assess CG also in real-world natural language tasks in order to understand the abilities and limitations of systems deployed in the wild. To this end, our GenBench Collaborative Benchmarking Task submission utilises the distribution-based compositionality assessment (DBCA) framework to split the Europarl translation corpus into a training and a test set in such a way that the test set requires compositional generalisation capacity. Specifically, the training and test sets have divergent distributions of dependency relations, testing NMT systems’ capability of translating dependencies that they have not been trained on. This is a fully-automated procedure to create natural language compositionality benchmarks, making it simple and inexpensive to apply it further to other datasets and languages. The code and data for the experiments is available at https://github.com/aalto-speech/dbca.
Original languageEnglish
Title of host publicationProceedings of the 1st GenBench Workshop on (Benchmarking) Generalisation in NLP
PublisherAssociation for Computational Linguistics
Number of pages10
ISBN (Electronic)979-8-89176-042-4
Publication statusPublished - 6 Dec 2023
MoE publication typeA4 Conference publication
EventGenBench: Workshop on generalisation (benchmarking) in NLP - Singapore, Singapore
Duration: 6 Dec 20236 Dec 2023
Conference number: 1

Workshop

WorkshopGenBench: Workshop on generalisation (benchmarking) in NLP
Country/TerritorySingapore
CitySingapore
Period06/12/202306/12/2023

Fingerprint

Dive into the research topics of 'On Using Distribution-Based Compositionality Assessment to Evaluate Compositional Generalisation in Machine Translation'. Together they form a unique fingerprint.

Cite this