@article{liu-etal-2023-compositional,
title = "Compositional Zero-Shot Domain Transfer with Text-to-Text Models",
author = "Liu, Fangyu and
Liu, Qianchu and
Bannur, Shruthi and
P{\'e}rez-Garc{\'\i}a, Fernando and
Usuyama, Naoto and
Zhang, Sheng and
Naumann, Tristan and
Nori, Aditya and
Poon, Hoifung and
Alvarez-Valle, Javier and
Oktay, Ozan and
Hyland, Stephanie L.",
journal = "Transactions of the Association for Computational Linguistics",
volume = "11",
year = "2023",
address = "Cambridge, MA",
publisher = "MIT Press",
url = "https://aclanthology.org/2023.tacl-1.62",
doi = "10.1162/tacl_a_00585",
pages = "1097--1113",
abstract = "Label scarcity is a bottleneck for improving task performance in specialized domains. We propose a novel compositional transfer learning framework (DoT51) for zero-shot domain transfer. Without access to in-domain labels, DoT5 jointly learns domain knowledge (from masked language modelling of unlabelled in-domain free text) and task knowledge (from task training on more readily available general-domain data) in a multi-task manner. To improve the transferability of task training, we design a strategy named NLGU: We simultaneously train natural language generation (NLG) for in-domain label-to-data generation, which enables data augmentation for self-finetuning and natural language understanding (NLU) for label prediction. We evaluate DoT5 on the biomedical domain and the resource-lean subdomain of radiology, focusing on natural language inference, text summarization, and embedding learning. DoT5 demonstrates the effectiveness of compositional transfer learning through multi-task learning. In particular, DoT5 outperforms the current state-of-the-art in zero-shot transfer by over 7 absolute points in accuracy on RadNLI. We validate DoT5 with ablations and a case study demonstrating its ability to solve challenging NLI examples requiring in-domain expertise.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="liu-etal-2023-compositional">
<titleInfo>
<title>Compositional Zero-Shot Domain Transfer with Text-to-Text Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Fangyu</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Qianchu</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shruthi</namePart>
<namePart type="family">Bannur</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Fernando</namePart>
<namePart type="family">Pérez-García</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Naoto</namePart>
<namePart type="family">Usuyama</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sheng</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tristan</namePart>
<namePart type="family">Naumann</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Aditya</namePart>
<namePart type="family">Nori</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hoifung</namePart>
<namePart type="family">Poon</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Javier</namePart>
<namePart type="family">Alvarez-Valle</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ozan</namePart>
<namePart type="family">Oktay</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Stephanie</namePart>
<namePart type="given">L</namePart>
<namePart type="family">Hyland</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>Label scarcity is a bottleneck for improving task performance in specialized domains. We propose a novel compositional transfer learning framework (DoT51) for zero-shot domain transfer. Without access to in-domain labels, DoT5 jointly learns domain knowledge (from masked language modelling of unlabelled in-domain free text) and task knowledge (from task training on more readily available general-domain data) in a multi-task manner. To improve the transferability of task training, we design a strategy named NLGU: We simultaneously train natural language generation (NLG) for in-domain label-to-data generation, which enables data augmentation for self-finetuning and natural language understanding (NLU) for label prediction. We evaluate DoT5 on the biomedical domain and the resource-lean subdomain of radiology, focusing on natural language inference, text summarization, and embedding learning. DoT5 demonstrates the effectiveness of compositional transfer learning through multi-task learning. In particular, DoT5 outperforms the current state-of-the-art in zero-shot transfer by over 7 absolute points in accuracy on RadNLI. We validate DoT5 with ablations and a case study demonstrating its ability to solve challenging NLI examples requiring in-domain expertise.</abstract>
<identifier type="citekey">liu-etal-2023-compositional</identifier>
<identifier type="doi">10.1162/tacl_a_00585</identifier>
<location>
<url>https://aclanthology.org/2023.tacl-1.62</url>
</location>
<part>
<date>2023</date>
<detail type="volume"><number>11</number></detail>
<extent unit="page">
<start>1097</start>
<end>1113</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Journal Article
%T Compositional Zero-Shot Domain Transfer with Text-to-Text Models
%A Liu, Fangyu
%A Liu, Qianchu
%A Bannur, Shruthi
%A Pérez-García, Fernando
%A Usuyama, Naoto
%A Zhang, Sheng
%A Naumann, Tristan
%A Nori, Aditya
%A Poon, Hoifung
%A Alvarez-Valle, Javier
%A Oktay, Ozan
%A Hyland, Stephanie L.
%J Transactions of the Association for Computational Linguistics
%D 2023
%V 11
%I MIT Press
%C Cambridge, MA
%F liu-etal-2023-compositional
%X Label scarcity is a bottleneck for improving task performance in specialized domains. We propose a novel compositional transfer learning framework (DoT51) for zero-shot domain transfer. Without access to in-domain labels, DoT5 jointly learns domain knowledge (from masked language modelling of unlabelled in-domain free text) and task knowledge (from task training on more readily available general-domain data) in a multi-task manner. To improve the transferability of task training, we design a strategy named NLGU: We simultaneously train natural language generation (NLG) for in-domain label-to-data generation, which enables data augmentation for self-finetuning and natural language understanding (NLU) for label prediction. We evaluate DoT5 on the biomedical domain and the resource-lean subdomain of radiology, focusing on natural language inference, text summarization, and embedding learning. DoT5 demonstrates the effectiveness of compositional transfer learning through multi-task learning. In particular, DoT5 outperforms the current state-of-the-art in zero-shot transfer by over 7 absolute points in accuracy on RadNLI. We validate DoT5 with ablations and a case study demonstrating its ability to solve challenging NLI examples requiring in-domain expertise.
%R 10.1162/tacl_a_00585
%U https://aclanthology.org/2023.tacl-1.62
%U https://doi.org/10.1162/tacl_a_00585
%P 1097-1113
Markdown (Informal)
[Compositional Zero-Shot Domain Transfer with Text-to-Text Models](https://aclanthology.org/2023.tacl-1.62) (Liu et al., TACL 2023)
ACL
- Fangyu Liu, Qianchu Liu, Shruthi Bannur, Fernando Pérez-García, Naoto Usuyama, Sheng Zhang, Tristan Naumann, Aditya Nori, Hoifung Poon, Javier Alvarez-Valle, Ozan Oktay, and Stephanie L. Hyland. 2023. Compositional Zero-Shot Domain Transfer with Text-to-Text Models. Transactions of the Association for Computational Linguistics, 11:1097–1113.