Work

Imitation learning for language generation from unaligned data

Year: 2016

Type: conference-paper

Abstract: Natural language generation (NLG) is the task of generating natural language from a meaning representation. \nRule-based approaches require domain-specific and manually constructed linguistic \nresources, while most corpus based approaches rely on aligned training data and/or phrase templates. \nThe latter are needed to restrict the search space for the structured prediction task defined \nby the unaligned datasets. In this work we propose the use of imitation learning for structured \nprediction which learns an incremental model that handles the large search space while avoiding \nexplicitly enumerating it. We adapted the Locally Optimal Learning to Search (Chang et \nal., 2015) framework which allows us to train against non-decomposable loss functions such \nas the BLEU or ROUGE scores while not assuming gold standard alignments. We evaluate our \napproach on three datasets using both automatic measures and human judgements and achieve \nresults comparable to the state-of-the-art approaches developed for each of them. Furthermore, \nwe performed an analysis of the datasets which examines common issues with NLG evaluation.

Source: White Rose Research Online (University of Leeds, The University of Sheffield, University of York)

Authors: Γεράσιμος Λάμπουρας, Andreas Vlachos

Institutions: Athens University of Economics and Business, University of Sheffield

Language: English

Topic: Natural Language Processing Techniques

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

Open Access status: green

Funder: Engineering and Physical Sciences Research Council

Award: DILiGENt: Domain-Independent Language Generation

OpenAlex ID: W2578330760