PhD Spring School: Error Analysis in Natural Language Generation

Natural Language Generation (NLG), the task of generating text from some input, is arguably useful to explain the reasoning of AI models. NLG models are not error free however and developing robust, trustworthy NLG models remains a key goal for explainable AI. In this context, the Nancy Spring School will focus on explainable methods for Natural Language Generation (NLG), methods which aim to detect errors in the output of NLG models and methods which aims to explain the sources of these errors.

Structure

The school will include lectures, hands on work and technical and innovation talks by Industrials.

The lectures, given by members of the NL4XAI consortium (Supervisors and Early Stage Researchers), will provide background knowledge on NLG, the types of errors manifest in NLG outputs, methods that have been used both to identify and to explain these errors and more generally, challenges for explainable NLG.

The hands-on-workshop will give the NL4XAI ESRs the opportunity first, to analyse, discuss and classify the types of errors manifest in NLG outputs; and second, to reflect as to which explainable method discussed by the morning lectures could be exploited to best explain these errors.

Finally, a series of talks by industrials will highlight various approaches and challenges met when processing text in industrial applications.

All events will take place at LORIA (a computer science research lab of Lorraine University). Tea breaks and lunches will take place in or outside the building giving participants plenty of time for informal interactions.

For further information please access the following link: https://members.loria.fr/CGardent/nl4xai-apr2022/nl4xai-ntp5.html