liu.seSearch for publications in DiVA
Endre søk
RefereraExporteraLink to record
Permanent link

Direct link
Referera
Referensformat
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • oxford
  • Annet format
Fler format
Språk
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Annet språk
Fler språk
Utmatningsformat
  • html
  • text
  • asciidoc
  • rtf
Human Ratings Do Not Reflect Downstream Utility: A Study of Free-Text Explanations for Model Predictions
Linköpings universitet, Institutionen för datavetenskap, Artificiell intelligens och integrerade datorsystem. Linköpings universitet, Tekniska fakulteten. (NLP)ORCID-id: 0009-0006-1001-0546
Linköpings universitet, Institutionen för datavetenskap. Linköpings universitet, Tekniska fakulteten.
Linköpings universitet, Institutionen för datavetenskap, Artificiell intelligens och integrerade datorsystem. Linköpings universitet, Tekniska fakulteten. (NLP)
Linköpings universitet, Institutionen för datavetenskap, Artificiell intelligens och integrerade datorsystem. Linköpings universitet, Tekniska fakulteten. (NLP)ORCID-id: 0000-0002-2492-9872
2022 (engelsk)Inngår i: Proceedings of the Fifth BlackboxNLP Workshop on Analyzing and Interpreting Neural Networks for NLP, 2022, Vol. 5, s. 164-177, artikkel-id 2022.blackboxnlp-1.14Konferansepaper, Publicerat paper (Fagfellevurdert)
Abstract [en]

Models able to generate free-text rationales that explain their output have been proposed as an important step towards interpretable NLP for “reasoning” tasks such as natural language inference and commonsense question answering. However, the relative merits of different architectures and types of rationales are not well understood and hard to measure. In this paper, we contribute two insights to this line of research: First, we find that models trained on gold explanations learn to rely on these but, in the case of the more challenging question answering data set we use, fail when given generated explanations at test time. However, additional fine-tuning on generated explanations teaches the model to distinguish between reliable and unreliable information in explanations. Second, we compare explanations by a generation-only model to those generated by a self-rationalizing model and find that, while the former score higher in terms of validity, factual correctness, and similarity to gold explanations, they are not more useful for downstream classification. We observe that the self-rationalizing model is prone to hallucination, which is punished by most metrics but may add useful context for the classification step.

sted, utgiver, år, opplag, sider
2022. Vol. 5, s. 164-177, artikkel-id 2022.blackboxnlp-1.14
Emneord [en]
Large Language Models, Neural Networks, Transformers, Interpretability, Explainability
HSV kategori
Identifikatorer
URN: urn:nbn:se:liu:diva-195615OAI: oai:DiVA.org:liu-195615DiVA, id: diva2:1773126
Konferanse
BlackboxNLP Workshop on Analyzing and Interpreting Neural Networks for NLP, December 8, 2022
Tilgjengelig fra: 2023-06-22 Laget: 2023-06-22 Sist oppdatert: 2024-04-02bibliografisk kontrollert
Inngår i avhandling
1. Understanding Large Language Models: Towards Rigorous and Targeted Interpretability Using Probing Classifiers and Self-Rationalisation
Åpne denne publikasjonen i ny fane eller vindu >>Understanding Large Language Models: Towards Rigorous and Targeted Interpretability Using Probing Classifiers and Self-Rationalisation
2024 (engelsk)Doktoravhandling, med artikler (Annet vitenskapelig)
Abstract [en]

Large language models (LLMs) have become the base of many natural language processing (NLP) systems due to their performance and easy adaptability to various tasks. However, much about their inner workings is still unknown. LLMs have many millions or billions of parameters, and large parts of their training happen in a self-supervised fashion: They simply learn to predict the next word, or missing words, in a sequence. This is effective for picking up a wide range of linguistic, factual and relational information, but it implies that it is not trivial what exactly is learned, and how it is represented within the LLM. 

In this thesis, I present our work on methods contributing to better understanding LLMs. The work can be grouped into two approaches. The first lies within the field of interpretability, which is concerned with understanding the internal workings of the LLMs. Specifically, we analyse and refine a tool called probing classifiers that inspects the intermediate representations of LLMs, focusing on what roles the various layers of the neural model play. This helps us to get a global understanding of how information is structured in the model. I present our work on assessing and improving the probing methodologies. We developed a framework to clarify the limitations of past methods, showing that all common controls are insufficient. Based on this, we proposed more restrictive probing setups by creating artificial distribution shifts. We developed new metrics for the evaluation of probing classifiers that move the focus from the overall information that the layer contains to differences in information content across the LLM. 

The second approach is concerned with explainability, specifically with self-rationalising models that generate free-text explanations along with their predictions. This is an instance of local understandability: We obtain justifications for individual predictions. In this setup, however, the generation of the explanations is just as opaque as the generation of the predictions. Therefore, our work in this field focuses on better understanding the properties of the generated explanations. We evaluate the downstream performance of a classifier with explanations generated by different model pipelines and compare it to human ratings of the explanations. Our results indicate that the properties that increase the downstream performance differ from those that humans appreciate when evaluating an explanation. Finally, we annotate explanations generated by an LLM for properties that human explanations typically have and discuss the effects those properties have on different user groups. 

While a detailed understanding of the inner workings of LLMs is still unfeasible, I argue that the techniques and analyses presented in this work can help to better understand LLMs, the linguistic knowledge they encode and their decision-making process. Together with knowledge about the models’ architecture, training data and training objective, such techniques can help us develop a robust high-level understanding of LLMs that can guide decisions on their deployment and potential improvements. 

sted, utgiver, år, opplag, sider
Linköping: Linköping University Electronic Press, 2024. s. 81
Serie
Linköping Studies in Science and Technology. Dissertations, ISSN 0345-7524 ; 2364
HSV kategori
Identifikatorer
urn:nbn:se:liu:diva-201985 (URN)10.3384/9789180754712 (DOI)9789180754705 (ISBN)9789180754712 (ISBN)
Disputas
2024-04-18, Ada Lovelace, B-building, Campus Valla, Linköping, 14:00 (engelsk)
Opponent
Veileder
Tilgjengelig fra: 2024-04-02 Laget: 2024-04-02 Sist oppdatert: 2024-04-02bibliografisk kontrollert

Open Access i DiVA

Fulltekst mangler i DiVA

Andre lenker

https://aclanthology.org/2022.blackboxnlp-1.14

Person

Kunz, JennyHolmström, OskarKuhlmann, Marco

Søk i DiVA

Av forfatter/redaktør
Kunz, JennyJirénius, MartinHolmström, OskarKuhlmann, Marco
Av organisasjonen

Søk utenfor DiVA

GoogleGoogle Scholar

urn-nbn

Altmetric

urn-nbn
Totalt: 100 treff
RefereraExporteraLink to record
Permanent link

Direct link
Referera
Referensformat
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • oxford
  • Annet format
Fler format
Språk
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Annet språk
Fler språk
Utmatningsformat
  • html
  • text
  • asciidoc
  • rtf