@inproceedings{sancheti-etal-2025-less,
title = "Less Mature is More Adaptable for Sentence-level Language Modeling",
author = "Sancheti, Abhilasha and
Dale, David and
Kozhevnikov, Artyom and
Elbayad, Maha",
editor = "Che, Wanxiang and
Nabende, Joyce and
Shutova, Ekaterina and
Pilehvar, Mohammad Taher",
booktitle = "Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
month = jul,
year = "2025",
address = "Vienna, Austria",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.acl-long.573/",
doi = "10.18653/v1/2025.acl-long.573",
pages = "11680--11695",
ISBN = "979-8-89176-251-0",
abstract = "This work investigates sentence-level models ($\textit{i.e.}$, models that operate at the sentence-level) to study how sentence representations from various encoders influence downstream task performance, and which syntactic, semantic, and discourse-level properties are essential for strong performance. Our experiments encompass encoders with diverse training regimes and pretraining domains, as well as various pooling strategies applied to multi-sentence input tasks (including sentence ordering, sentiment classification, and natural language inference) requiring coarse-to-fine-grained reasoning. We find that ``less mature'' representations ($\textit{e.g.}$, mean-pooled representations from BERT{'}s first or last layer, or representations from encoders with limited fine-tuning) exhibit greater generalizability and adaptability to downstream tasks compared to representations from extensively fine-tuned models ($\textit{e.g.}$, SBERT or SimCSE). These findings are consistent across different pretraining seed initializations for BERT. Our probing analysis reveals that syntactic and discourse-level properties are stronger indicators of downstream performance than MTEB scores or decodability. Furthermore, the data and time efficiency of sentence-level models, often outperforming token-level models, underscores their potential for future research."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="sancheti-etal-2025-less">
<titleInfo>
<title>Less Mature is More Adaptable for Sentence-level Language Modeling</title>
</titleInfo>
<name type="personal">
<namePart type="given">Abhilasha</namePart>
<namePart type="family">Sancheti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">David</namePart>
<namePart type="family">Dale</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Artyom</namePart>
<namePart type="family">Kozhevnikov</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Maha</namePart>
<namePart type="family">Elbayad</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2025-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Wanxiang</namePart>
<namePart type="family">Che</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Joyce</namePart>
<namePart type="family">Nabende</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ekaterina</namePart>
<namePart type="family">Shutova</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mohammad</namePart>
<namePart type="given">Taher</namePart>
<namePart type="family">Pilehvar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Vienna, Austria</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-251-0</identifier>
</relatedItem>
<abstract>This work investigates sentence-level models (i.e., models that operate at the sentence-level) to study how sentence representations from various encoders influence downstream task performance, and which syntactic, semantic, and discourse-level properties are essential for strong performance. Our experiments encompass encoders with diverse training regimes and pretraining domains, as well as various pooling strategies applied to multi-sentence input tasks (including sentence ordering, sentiment classification, and natural language inference) requiring coarse-to-fine-grained reasoning. We find that “less mature” representations (e.g., mean-pooled representations from BERT’s first or last layer, or representations from encoders with limited fine-tuning) exhibit greater generalizability and adaptability to downstream tasks compared to representations from extensively fine-tuned models (e.g., SBERT or SimCSE). These findings are consistent across different pretraining seed initializations for BERT. Our probing analysis reveals that syntactic and discourse-level properties are stronger indicators of downstream performance than MTEB scores or decodability. Furthermore, the data and time efficiency of sentence-level models, often outperforming token-level models, underscores their potential for future research.</abstract>
<identifier type="citekey">sancheti-etal-2025-less</identifier>
<identifier type="doi">10.18653/v1/2025.acl-long.573</identifier>
<location>
<url>https://aclanthology.org/2025.acl-long.573/</url>
</location>
<part>
<date>2025-07</date>
<extent unit="page">
<start>11680</start>
<end>11695</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Less Mature is More Adaptable for Sentence-level Language Modeling
%A Sancheti, Abhilasha
%A Dale, David
%A Kozhevnikov, Artyom
%A Elbayad, Maha
%Y Che, Wanxiang
%Y Nabende, Joyce
%Y Shutova, Ekaterina
%Y Pilehvar, Mohammad Taher
%S Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)
%D 2025
%8 July
%I Association for Computational Linguistics
%C Vienna, Austria
%@ 979-8-89176-251-0
%F sancheti-etal-2025-less
%X This work investigates sentence-level models (i.e., models that operate at the sentence-level) to study how sentence representations from various encoders influence downstream task performance, and which syntactic, semantic, and discourse-level properties are essential for strong performance. Our experiments encompass encoders with diverse training regimes and pretraining domains, as well as various pooling strategies applied to multi-sentence input tasks (including sentence ordering, sentiment classification, and natural language inference) requiring coarse-to-fine-grained reasoning. We find that “less mature” representations (e.g., mean-pooled representations from BERT’s first or last layer, or representations from encoders with limited fine-tuning) exhibit greater generalizability and adaptability to downstream tasks compared to representations from extensively fine-tuned models (e.g., SBERT or SimCSE). These findings are consistent across different pretraining seed initializations for BERT. Our probing analysis reveals that syntactic and discourse-level properties are stronger indicators of downstream performance than MTEB scores or decodability. Furthermore, the data and time efficiency of sentence-level models, often outperforming token-level models, underscores their potential for future research.
%R 10.18653/v1/2025.acl-long.573
%U https://aclanthology.org/2025.acl-long.573/
%U https://doi.org/10.18653/v1/2025.acl-long.573
%P 11680-11695
Markdown (Informal)
[Less Mature is More Adaptable for Sentence-level Language Modeling](https://aclanthology.org/2025.acl-long.573/) (Sancheti et al., ACL 2025)
ACL
- Abhilasha Sancheti, David Dale, Artyom Kozhevnikov, and Maha Elbayad. 2025. Less Mature is More Adaptable for Sentence-level Language Modeling. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 11680–11695, Vienna, Austria. Association for Computational Linguistics.