<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zhao-etal-2024-explaining">
    <titleInfo>
        <title>Explaining Language Model Predictions with High-Impact Concepts</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Ruochen</namePart>
        <namePart type="family">Zhao</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Tan</namePart>
        <namePart type="family">Wang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Yongjie</namePart>
        <namePart type="family">Wang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Shafiq</namePart>
        <namePart type="family">Joty</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <originInfo>
        <dateIssued>2024-03</dateIssued>
    </originInfo>
    <typeOfResource>text</typeOfResource>
    <relatedItem type="host">
        <titleInfo>
            <title>Findings of the Association for Computational Linguistics: EACL 2024</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Yvette</namePart>
            <namePart type="family">Graham</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Matthew</namePart>
            <namePart type="family">Purver</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <originInfo>
            <publisher>Association for Computational Linguistics</publisher>
            <place>
                <placeTerm type="text">St. Julian’s, Malta</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">conference publication</genre>
    </relatedItem>
    <abstract>To encourage fairness and transparency, there exists an urgent demand for deriving reliable explanations for large language models (LLMs). One promising solution is concept-based explanations, i.e., human-understandable concepts from internal representations. However, due to the compositional nature of languages, current methods mostly discover correlational explanations instead of causal features. Therefore, we propose a novel framework to provide impact-aware explanations for users to understand the LLM’s behavior, which are robust to feature changes and influential to the model’s predictions. Specifically, we extract predictive high-level features (concepts) from the model’s hidden layer activations. Then, we innovatively optimize for features whose existence causes the output predictions to change substantially. Extensive experiments on real and synthetic tasks demonstrate that our method achieves superior results on predictive impact, explainability, and faithfulness compared to the baselines, especially for LLMs.</abstract>
    <identifier type="citekey">zhao-etal-2024-explaining</identifier>
    <location>
        <url>https://aclanthology.org/2024.findings-eacl.67</url>
    </location>
    <part>
        <date>2024-03</date>
        <extent unit="page">
            <start>995</start>
            <end>1012</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Explaining Language Model Predictions with High-Impact Concepts
%A Zhao, Ruochen
%A Wang, Tan
%A Wang, Yongjie
%A Joty, Shafiq
%Y Graham, Yvette
%Y Purver, Matthew
%S Findings of the Association for Computational Linguistics: EACL 2024
%D 2024
%8 March
%I Association for Computational Linguistics
%C St. Julian’s, Malta
%F zhao-etal-2024-explaining
%X To encourage fairness and transparency, there exists an urgent demand for deriving reliable explanations for large language models (LLMs). One promising solution is concept-based explanations, i.e., human-understandable concepts from internal representations. However, due to the compositional nature of languages, current methods mostly discover correlational explanations instead of causal features. Therefore, we propose a novel framework to provide impact-aware explanations for users to understand the LLM’s behavior, which are robust to feature changes and influential to the model’s predictions. Specifically, we extract predictive high-level features (concepts) from the model’s hidden layer activations. Then, we innovatively optimize for features whose existence causes the output predictions to change substantially. Extensive experiments on real and synthetic tasks demonstrate that our method achieves superior results on predictive impact, explainability, and faithfulness compared to the baselines, especially for LLMs.
%U https://aclanthology.org/2024.findings-eacl.67
%P 995-1012

Markdown (Informal)

[Explaining Language Model Predictions with High-Impact Concepts](https://aclanthology.org/2024.findings-eacl.67) (Zhao et al., Findings 2024)

• Explaining Language Model Predictions with High-Impact Concepts (Zhao et al., Findings 2024)

ACL

• Ruochen Zhao, Tan Wang, Yongjie Wang, and Shafiq Joty. 2024. Explaining Language Model Predictions with High-Impact Concepts. In Findings of the Association for Computational Linguistics: EACL 2024, pages 995–1012, St. Julian’s, Malta. Association for Computational Linguistics.