@inproceedings{turan-etal-2022-adapting,
title = "Adapting Language Models When Training on Privacy-Transformed Data",
author = "Turan, Tugtekin and
Klakow, Dietrich and
Vincent, Emmanuel and
Jouvet, Denis",
editor = "Calzolari, Nicoletta and
B{\'e}chet, Fr{\'e}d{\'e}ric and
Blache, Philippe and
Choukri, Khalid and
Cieri, Christopher and
Declerck, Thierry and
Goggi, Sara and
Isahara, Hitoshi and
Maegaard, Bente and
Mariani, Joseph and
Mazo, H{\'e}l{\`e}ne and
Odijk, Jan and
Piperidis, Stelios",
booktitle = "Proceedings of the Thirteenth Language Resources and Evaluation Conference",
month = jun,
year = "2022",
address = "Marseille, France",
publisher = "European Language Resources Association",
url = "https://aclanthology.org/2022.lrec-1.465/",
pages = "4367--4373",
abstract = "In recent years, voice-controlled personal assistants have revolutionized the interaction with smart devices and mobile applications. The collected data are then used by system providers to train language models (LMs). Each spoken message reveals personal information, hence removing private information from the input sentences is necessary. Our data sanitization process relies on recognizing and replacing named entities by other words from the same class. However, this may harm LM training because privacy-transformed data is unlikely to match the test distribution. This paper aims to fill the gap by focusing on the adaptation of LMs initially trained on privacy-transformed sentences using a small amount of original untransformed data. To do so, we combine class-based LMs, which provide an effective approach to overcome data sparsity in the context of n-gram LMs, and neural LMs, which handle longer contexts and can yield better predictions. Our experiments show that training an LM on privacy-transformed data result in a relative 11{\%} word error rate (WER) increase compared to training on the original untransformed data, and adapting that model on a limited amount of original untransformed data leads to a relative 8{\%} WER improvement over the model trained solely on privacy-transformed data."
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="turan-etal-2022-adapting">
<titleInfo>
<title>Adapting Language Models When Training on Privacy-Transformed Data</title>
</titleInfo>
<name type="personal">
<namePart type="given">Tugtekin</namePart>
<namePart type="family">Turan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Dietrich</namePart>
<namePart type="family">Klakow</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Emmanuel</namePart>
<namePart type="family">Vincent</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Denis</namePart>
<namePart type="family">Jouvet</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2022-06</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the Thirteenth Language Resources and Evaluation Conference</title>
</titleInfo>
<name type="personal">
<namePart type="given">Nicoletta</namePart>
<namePart type="family">Calzolari</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Frédéric</namePart>
<namePart type="family">Béchet</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Philippe</namePart>
<namePart type="family">Blache</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Khalid</namePart>
<namePart type="family">Choukri</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Christopher</namePart>
<namePart type="family">Cieri</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Thierry</namePart>
<namePart type="family">Declerck</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sara</namePart>
<namePart type="family">Goggi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hitoshi</namePart>
<namePart type="family">Isahara</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Bente</namePart>
<namePart type="family">Maegaard</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Joseph</namePart>
<namePart type="family">Mariani</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hélène</namePart>
<namePart type="family">Mazo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jan</namePart>
<namePart type="family">Odijk</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Stelios</namePart>
<namePart type="family">Piperidis</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>European Language Resources Association</publisher>
<place>
<placeTerm type="text">Marseille, France</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>In recent years, voice-controlled personal assistants have revolutionized the interaction with smart devices and mobile applications. The collected data are then used by system providers to train language models (LMs). Each spoken message reveals personal information, hence removing private information from the input sentences is necessary. Our data sanitization process relies on recognizing and replacing named entities by other words from the same class. However, this may harm LM training because privacy-transformed data is unlikely to match the test distribution. This paper aims to fill the gap by focusing on the adaptation of LMs initially trained on privacy-transformed sentences using a small amount of original untransformed data. To do so, we combine class-based LMs, which provide an effective approach to overcome data sparsity in the context of n-gram LMs, and neural LMs, which handle longer contexts and can yield better predictions. Our experiments show that training an LM on privacy-transformed data result in a relative 11% word error rate (WER) increase compared to training on the original untransformed data, and adapting that model on a limited amount of original untransformed data leads to a relative 8% WER improvement over the model trained solely on privacy-transformed data.</abstract>
<identifier type="citekey">turan-etal-2022-adapting</identifier>
<location>
<url>https://aclanthology.org/2022.lrec-1.465/</url>
</location>
<part>
<date>2022-06</date>
<extent unit="page">
<start>4367</start>
<end>4373</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Adapting Language Models When Training on Privacy-Transformed Data
%A Turan, Tugtekin
%A Klakow, Dietrich
%A Vincent, Emmanuel
%A Jouvet, Denis
%Y Calzolari, Nicoletta
%Y Béchet, Frédéric
%Y Blache, Philippe
%Y Choukri, Khalid
%Y Cieri, Christopher
%Y Declerck, Thierry
%Y Goggi, Sara
%Y Isahara, Hitoshi
%Y Maegaard, Bente
%Y Mariani, Joseph
%Y Mazo, Hélène
%Y Odijk, Jan
%Y Piperidis, Stelios
%S Proceedings of the Thirteenth Language Resources and Evaluation Conference
%D 2022
%8 June
%I European Language Resources Association
%C Marseille, France
%F turan-etal-2022-adapting
%X In recent years, voice-controlled personal assistants have revolutionized the interaction with smart devices and mobile applications. The collected data are then used by system providers to train language models (LMs). Each spoken message reveals personal information, hence removing private information from the input sentences is necessary. Our data sanitization process relies on recognizing and replacing named entities by other words from the same class. However, this may harm LM training because privacy-transformed data is unlikely to match the test distribution. This paper aims to fill the gap by focusing on the adaptation of LMs initially trained on privacy-transformed sentences using a small amount of original untransformed data. To do so, we combine class-based LMs, which provide an effective approach to overcome data sparsity in the context of n-gram LMs, and neural LMs, which handle longer contexts and can yield better predictions. Our experiments show that training an LM on privacy-transformed data result in a relative 11% word error rate (WER) increase compared to training on the original untransformed data, and adapting that model on a limited amount of original untransformed data leads to a relative 8% WER improvement over the model trained solely on privacy-transformed data.
%U https://aclanthology.org/2022.lrec-1.465/
%P 4367-4373
Markdown (Informal)
[Adapting Language Models When Training on Privacy-Transformed Data](https://aclanthology.org/2022.lrec-1.465/) (Turan et al., LREC 2022)
ACL