@inproceedings{feldhus-etal-2023-saliency,
title = "Saliency Map Verbalization: Comparing Feature Importance Representations from Model-free and Instruction-based Methods",
author = {Feldhus, Nils and
Hennig, Leonhard and
Nasert, Maximilian Dustin and
Ebert, Christopher and
Schwarzenberg, Robert and
M{\"o}ller, Sebastian},
editor = "Dalvi Mishra, Bhavana and
Durrett, Greg and
Jansen, Peter and
Neves Ribeiro, Danilo and
Wei, Jason",
booktitle = "Proceedings of the 1st Workshop on Natural Language Reasoning and Structured Explanations (NLRSE)",
month = jun,
year = "2023",
address = "Toronto, Canada",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2023.nlrse-1.4",
doi = "10.18653/v1/2023.nlrse-1.4",
pages = "30--46",
abstract = "Saliency maps can explain a neural model{'}s predictions by identifying important input features. They are difficult to interpret for laypeople, especially for instances with many features. In order to make them more accessible, we formalize the underexplored task of translating saliency maps into natural language and compare methods that address two key challenges of this approach {--} what and how to verbalize. In both automatic and human evaluation setups, using token-level attributions from text classification tasks, we compare two novel methods (search-based and instruction-based verbalizations) against conventional feature importance representations (heatmap visualizations and extractive rationales), measuring simulatability, faithfulness, helpfulness and ease of understanding. Instructing GPT-3.5 to generate saliency map verbalizations yields plausible explanations which include associations, abstractive summarization and commonsense reasoning, achieving by far the highest human ratings, but they are not faithfully capturing numeric information and are inconsistent in their interpretation of the task. In comparison, our search-based, model-free verbalization approach efficiently completes templated verbalizations, is faithful by design, but falls short in helpfulness and simulatability. Our results suggest that saliency map verbalization makes feature attribution explanations more comprehensible and less cognitively challenging to humans than conventional representations.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="feldhus-etal-2023-saliency">
<titleInfo>
<title>Saliency Map Verbalization: Comparing Feature Importance Representations from Model-free and Instruction-based Methods</title>
</titleInfo>
<name type="personal">
<namePart type="given">Nils</namePart>
<namePart type="family">Feldhus</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Leonhard</namePart>
<namePart type="family">Hennig</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Maximilian</namePart>
<namePart type="given">Dustin</namePart>
<namePart type="family">Nasert</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Christopher</namePart>
<namePart type="family">Ebert</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Robert</namePart>
<namePart type="family">Schwarzenberg</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sebastian</namePart>
<namePart type="family">Möller</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2023-06</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 1st Workshop on Natural Language Reasoning and Structured Explanations (NLRSE)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Bhavana</namePart>
<namePart type="family">Dalvi Mishra</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Greg</namePart>
<namePart type="family">Durrett</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Peter</namePart>
<namePart type="family">Jansen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Danilo</namePart>
<namePart type="family">Neves Ribeiro</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jason</namePart>
<namePart type="family">Wei</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Toronto, Canada</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>Saliency maps can explain a neural model’s predictions by identifying important input features. They are difficult to interpret for laypeople, especially for instances with many features. In order to make them more accessible, we formalize the underexplored task of translating saliency maps into natural language and compare methods that address two key challenges of this approach – what and how to verbalize. In both automatic and human evaluation setups, using token-level attributions from text classification tasks, we compare two novel methods (search-based and instruction-based verbalizations) against conventional feature importance representations (heatmap visualizations and extractive rationales), measuring simulatability, faithfulness, helpfulness and ease of understanding. Instructing GPT-3.5 to generate saliency map verbalizations yields plausible explanations which include associations, abstractive summarization and commonsense reasoning, achieving by far the highest human ratings, but they are not faithfully capturing numeric information and are inconsistent in their interpretation of the task. In comparison, our search-based, model-free verbalization approach efficiently completes templated verbalizations, is faithful by design, but falls short in helpfulness and simulatability. Our results suggest that saliency map verbalization makes feature attribution explanations more comprehensible and less cognitively challenging to humans than conventional representations.</abstract>
<identifier type="citekey">feldhus-etal-2023-saliency</identifier>
<identifier type="doi">10.18653/v1/2023.nlrse-1.4</identifier>
<location>
<url>https://aclanthology.org/2023.nlrse-1.4</url>
</location>
<part>
<date>2023-06</date>
<extent unit="page">
<start>30</start>
<end>46</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Saliency Map Verbalization: Comparing Feature Importance Representations from Model-free and Instruction-based Methods
%A Feldhus, Nils
%A Hennig, Leonhard
%A Nasert, Maximilian Dustin
%A Ebert, Christopher
%A Schwarzenberg, Robert
%A Möller, Sebastian
%Y Dalvi Mishra, Bhavana
%Y Durrett, Greg
%Y Jansen, Peter
%Y Neves Ribeiro, Danilo
%Y Wei, Jason
%S Proceedings of the 1st Workshop on Natural Language Reasoning and Structured Explanations (NLRSE)
%D 2023
%8 June
%I Association for Computational Linguistics
%C Toronto, Canada
%F feldhus-etal-2023-saliency
%X Saliency maps can explain a neural model’s predictions by identifying important input features. They are difficult to interpret for laypeople, especially for instances with many features. In order to make them more accessible, we formalize the underexplored task of translating saliency maps into natural language and compare methods that address two key challenges of this approach – what and how to verbalize. In both automatic and human evaluation setups, using token-level attributions from text classification tasks, we compare two novel methods (search-based and instruction-based verbalizations) against conventional feature importance representations (heatmap visualizations and extractive rationales), measuring simulatability, faithfulness, helpfulness and ease of understanding. Instructing GPT-3.5 to generate saliency map verbalizations yields plausible explanations which include associations, abstractive summarization and commonsense reasoning, achieving by far the highest human ratings, but they are not faithfully capturing numeric information and are inconsistent in their interpretation of the task. In comparison, our search-based, model-free verbalization approach efficiently completes templated verbalizations, is faithful by design, but falls short in helpfulness and simulatability. Our results suggest that saliency map verbalization makes feature attribution explanations more comprehensible and less cognitively challenging to humans than conventional representations.
%R 10.18653/v1/2023.nlrse-1.4
%U https://aclanthology.org/2023.nlrse-1.4
%U https://doi.org/10.18653/v1/2023.nlrse-1.4
%P 30-46
Markdown (Informal)
[Saliency Map Verbalization: Comparing Feature Importance Representations from Model-free and Instruction-based Methods](https://aclanthology.org/2023.nlrse-1.4) (Feldhus et al., NLRSE 2023)
ACL