@inproceedings{borec-etal-2024-unreasonable-ineffectiveness,
title = "The Unreasonable Ineffectiveness of Nucleus Sampling on Mitigating Text Memorization",
author = "Borec, Luka and
Sadler, Philipp and
Schlangen, David",
editor = "Mahamood, Saad and
Minh, Nguyen Le and
Ippolito, Daphne",
booktitle = "Proceedings of the 17th International Natural Language Generation Conference",
month = sep,
year = "2024",
address = "Tokyo, Japan",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2024.inlg-main.30",
pages = "358--370",
abstract = "This work analyses the text memorization behavior of large language models (LLMs) when subjected to nucleus sampling. Stochastic decoding methods like nucleus sampling are typically applied to overcome issues such as monotonous and repetitive text generation, which are often observed with maximization-based decoding techniques. We hypothesize that nucleus sampling might also reduce the occurrence of memorization patterns, because it could lead to the selection of tokens outside the memorized sequence. To test this hypothesis we create a diagnostic dataset with a known distribution of duplicates that gives us some control over the likelihood of memorisation of certain parts of the training data. Our analysis of two GPT-Neo models fine-tuned on this dataset interestingly shows that (i) an increase of the nucleus size reduces memorization only modestly, and (ii) even when models do not engage in {``}hard{''} memorization {--} a verbatim reproduction of training samples {--} they may still display {``}soft{''} memorization whereby they generate outputs that echo the training data but without a complete one-by-one resemblance.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="borec-etal-2024-unreasonable-ineffectiveness">
<titleInfo>
<title>The Unreasonable Ineffectiveness of Nucleus Sampling on Mitigating Text Memorization</title>
</titleInfo>
<name type="personal">
<namePart type="given">Luka</namePart>
<namePart type="family">Borec</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Philipp</namePart>
<namePart type="family">Sadler</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">David</namePart>
<namePart type="family">Schlangen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2024-09</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 17th International Natural Language Generation Conference</title>
</titleInfo>
<name type="personal">
<namePart type="given">Saad</namePart>
<namePart type="family">Mahamood</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nguyen</namePart>
<namePart type="given">Le</namePart>
<namePart type="family">Minh</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Daphne</namePart>
<namePart type="family">Ippolito</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Tokyo, Japan</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>This work analyses the text memorization behavior of large language models (LLMs) when subjected to nucleus sampling. Stochastic decoding methods like nucleus sampling are typically applied to overcome issues such as monotonous and repetitive text generation, which are often observed with maximization-based decoding techniques. We hypothesize that nucleus sampling might also reduce the occurrence of memorization patterns, because it could lead to the selection of tokens outside the memorized sequence. To test this hypothesis we create a diagnostic dataset with a known distribution of duplicates that gives us some control over the likelihood of memorisation of certain parts of the training data. Our analysis of two GPT-Neo models fine-tuned on this dataset interestingly shows that (i) an increase of the nucleus size reduces memorization only modestly, and (ii) even when models do not engage in “hard” memorization – a verbatim reproduction of training samples – they may still display “soft” memorization whereby they generate outputs that echo the training data but without a complete one-by-one resemblance.</abstract>
<identifier type="citekey">borec-etal-2024-unreasonable-ineffectiveness</identifier>
<location>
<url>https://aclanthology.org/2024.inlg-main.30</url>
</location>
<part>
<date>2024-09</date>
<extent unit="page">
<start>358</start>
<end>370</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T The Unreasonable Ineffectiveness of Nucleus Sampling on Mitigating Text Memorization
%A Borec, Luka
%A Sadler, Philipp
%A Schlangen, David
%Y Mahamood, Saad
%Y Minh, Nguyen Le
%Y Ippolito, Daphne
%S Proceedings of the 17th International Natural Language Generation Conference
%D 2024
%8 September
%I Association for Computational Linguistics
%C Tokyo, Japan
%F borec-etal-2024-unreasonable-ineffectiveness
%X This work analyses the text memorization behavior of large language models (LLMs) when subjected to nucleus sampling. Stochastic decoding methods like nucleus sampling are typically applied to overcome issues such as monotonous and repetitive text generation, which are often observed with maximization-based decoding techniques. We hypothesize that nucleus sampling might also reduce the occurrence of memorization patterns, because it could lead to the selection of tokens outside the memorized sequence. To test this hypothesis we create a diagnostic dataset with a known distribution of duplicates that gives us some control over the likelihood of memorisation of certain parts of the training data. Our analysis of two GPT-Neo models fine-tuned on this dataset interestingly shows that (i) an increase of the nucleus size reduces memorization only modestly, and (ii) even when models do not engage in “hard” memorization – a verbatim reproduction of training samples – they may still display “soft” memorization whereby they generate outputs that echo the training data but without a complete one-by-one resemblance.
%U https://aclanthology.org/2024.inlg-main.30
%P 358-370
Markdown (Informal)
[The Unreasonable Ineffectiveness of Nucleus Sampling on Mitigating Text Memorization](https://aclanthology.org/2024.inlg-main.30) (Borec et al., INLG 2024)
ACL