TY - GEN
T1 - Exploring Neural Language Models via Analysis of Local and Global Self-Attention Spaces
AU - Škrlj, Blaž
AU - Sheehan, Shane
AU - Eržen, Nika
AU - Robnik-Šikonja, Marko
AU - Luz, Saturnino
AU - Pollak, Senja
N1 - Funding Information:
We acknowledge European Union’s Horizon 2020 research and innovation programme under grant agreement No 825153, project EMBEDDIA (Cross-Lingual Embeddings). The first author was also funded by Slovenian Research Agency as a young researcher.
Publisher Copyright:
© Association for Computational Linguistics
PY - 2021/4/16
Y1 - 2021/4/16
N2 - Large pretrained language models using the transformer neural network architecture are becoming a dominant methodology for many natural language processing tasks, such as question answering, text classification, word sense disambiguation, text completion and machine translation. Commonly comprising hundreds of millions of parameters, these models offer state-of-the-art performance, but at the expense of interpretability. The attention mechanism is the main component of transformer networks. We present AttViz, a method for exploration of self-attention in transformer networks, which can help in explanation and debugging of the trained models by showing associations between text tokens in an input sequence. We show that existing deep learning pipelines can be explored with AttViz, which offers novel visualizations of the attention heads and their aggregations. We implemented the proposed methods in an online toolkit and an offline library. Using examples from news analysis, we demonstrate how AttViz can be used to inspect and potentially better understand what a model has learned.
AB - Large pretrained language models using the transformer neural network architecture are becoming a dominant methodology for many natural language processing tasks, such as question answering, text classification, word sense disambiguation, text completion and machine translation. Commonly comprising hundreds of millions of parameters, these models offer state-of-the-art performance, but at the expense of interpretability. The attention mechanism is the main component of transformer networks. We present AttViz, a method for exploration of self-attention in transformer networks, which can help in explanation and debugging of the trained models by showing associations between text tokens in an input sequence. We show that existing deep learning pipelines can be explored with AttViz, which offers novel visualizations of the attention heads and their aggregations. We implemented the proposed methods in an online toolkit and an offline library. Using examples from news analysis, we demonstrate how AttViz can be used to inspect and potentially better understand what a model has learned.
UR - http://www.scopus.com/inward/record.url?scp=85123279871&partnerID=8YFLogxK
M3 - Conference contribution
AN - SCOPUS:85123279871
T3 - EACL Hackashop on News Media Content Analysis and Automated Report Generation, Hackashop 2021 at 16th conference of the European Chapter of the Association for Computational Linguistics, EACL 2021 - Proceedings
SP - 76
EP - 83
BT - EACL Hackashop on News Media Content Analysis and Automated Report Generation, Hackashop 2021 at 16th conference of the European Chapter of the Association for Computational Linguistics, EACL 2021 - Proceedings
A2 - Toivonen, Hannu
A2 - Boggia, Michele
PB - Association for Computational Linguistics (ACL)
T2 - 2021 EACL Hackashop on News Media Content Analysis and Automated Report Generation, Hackashop 2021
Y2 - 19 April 2021
ER -