<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="oh-schuler-2023-surprisal">
    <titleInfo>
        <title>Why Does Surprisal From Larger Transformer-Based Language Models Provide a Poorer Fit to Human Reading Times?</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Byung-Doh</namePart>
        <namePart type="family">Oh</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">William</namePart>
        <namePart type="family">Schuler</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <originInfo>
        <dateIssued>2023</dateIssued>
    </originInfo>
    <typeOfResource>text</typeOfResource>
    <genre authority="bibutilsgt">journal article</genre>
    <relatedItem type="host">
        <titleInfo>
            <title>Transactions of the Association for Computational Linguistics</title>
        </titleInfo>
        <originInfo>
            <issuance>continuing</issuance>
            <publisher>MIT Press</publisher>
            <place>
                <placeTerm type="text">Cambridge, MA</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">periodical</genre>
        <genre authority="bibutilsgt">academic journal</genre>
    </relatedItem>
    <abstract>This work presents a linguistic analysis into why larger Transformer-based pre-trained language models with more parameters and lower perplexity nonetheless yield surprisal estimates that are less predictive of human reading times. First, regression analyses show a strictly monotonic, positive log-linear relationship between perplexity and fit to reading times for the more recently released five GPT-Neo variants and eight OPT variants on two separate datasets, replicating earlier results limited to just GPT-2 (Oh et al., 2022). Subsequently, analysis of residual errors reveals a systematic deviation of the larger variants, such as underpredicting reading times of named entities and making compensatory overpredictions for reading times of function words such as modals and conjunctions. These results suggest that the propensity of larger Transformer-based models to ‘memorize’ sequences during training makes their surprisal estimates diverge from humanlike expectations, which warrants caution in using pre-trained language models to study human language processing.</abstract>
    <identifier type="citekey">oh-schuler-2023-surprisal</identifier>
    <identifier type="doi">10.1162/tacl_a_00548</identifier>
    <location>
        <url>https://aclanthology.org/2023.tacl-1.20</url>
    </location>
    <part>
        <date>2023</date>
        <detail type="volume"><number>11</number></detail>
        <extent unit="page">
            <start>336</start>
            <end>350</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Journal Article
%T Why Does Surprisal From Larger Transformer-Based Language Models Provide a Poorer Fit to Human Reading Times?
%A Oh, Byung-Doh
%A Schuler, William
%J Transactions of the Association for Computational Linguistics
%D 2023
%V 11
%I MIT Press
%C Cambridge, MA
%F oh-schuler-2023-surprisal
%X This work presents a linguistic analysis into why larger Transformer-based pre-trained language models with more parameters and lower perplexity nonetheless yield surprisal estimates that are less predictive of human reading times. First, regression analyses show a strictly monotonic, positive log-linear relationship between perplexity and fit to reading times for the more recently released five GPT-Neo variants and eight OPT variants on two separate datasets, replicating earlier results limited to just GPT-2 (Oh et al., 2022). Subsequently, analysis of residual errors reveals a systematic deviation of the larger variants, such as underpredicting reading times of named entities and making compensatory overpredictions for reading times of function words such as modals and conjunctions. These results suggest that the propensity of larger Transformer-based models to ‘memorize’ sequences during training makes their surprisal estimates diverge from humanlike expectations, which warrants caution in using pre-trained language models to study human language processing.
%R 10.1162/tacl_a_00548
%U https://aclanthology.org/2023.tacl-1.20
%U https://doi.org/10.1162/tacl_a_00548
%P 336-350

Markdown (Informal)

[Why Does Surprisal From Larger Transformer-Based Language Models Provide a Poorer Fit to Human Reading Times?](https://aclanthology.org/2023.tacl-1.20) (Oh & Schuler, TACL 2023)

• Why Does Surprisal From Larger Transformer-Based Language Models Provide a Poorer Fit to Human Reading Times? (Oh & Schuler, TACL 2023)

ACL

• Byung-Doh Oh and William Schuler. 2023. Why Does Surprisal From Larger Transformer-Based Language Models Provide a Poorer Fit to Human Reading Times?. Transactions of the Association for Computational Linguistics, 11:336–350.