@inproceedings{liang-chen-2026-escaping,
title = "Escaping the Probability Trap: Mitigating Semantic Drift in {C}antonese-{M}andarin Translation",
author = "Liang, Yuzhi and
Chen, Fangqi",
editor = "Hettiarachchi, Hansi and
Ranasinghe, Tharindu and
Plum, Alistair and
Rayson, Paul and
Mitkov, Ruslan and
Gaber, Mohamed and
Premasiri, Damith and
Tan, Fiona Anting and
Uyangodage, Lasitha",
booktitle = "Proceedings of the Second Workshop on Language Models for Low-Resource Languages ({L}o{R}es{LM} 2026)",
month = mar,
year = "2026",
address = "Rabat, Morocco",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.loreslm-1.41/",
pages = "471--483",
ISBN = "979-8-89176-377-7",
abstract = "Fine-tuning multilingual models for low-resource dialect translation frequently encounters a ``plausibility over faithfulness'' dilemma, resulting in severe semantic drift on dialect-specific tokens. We term this phenomenon the ``Probability Trap,'' where models prioritize statistical fluency over semantic fidelity. To address this, we propose MVS-Rank (Multi-View Scoring Reranking), a generate-then-rerank framework that decouples evaluation from generation. Our method assesses translation candidates through three complementary perspectives: (1) Source-Side Faithfulness via a Reverse Translation Model to anchor semantic fidelity; (2) Local Fluency using Masked Language Models to ensure syntactic precision; and (3) Global Fluency leveraging Large Language Models to capture discourse coherence. Extensive experiments on Cantonese-Mandarin benchmarks demonstrate that MVS-Rank achieves state-of-the-art performance, significantly outperforming strong fine-tuning baselines by effectively rectifying hallucinations while maintaining high fluency."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="liang-chen-2026-escaping">
<titleInfo>
<title>Escaping the Probability Trap: Mitigating Semantic Drift in Cantonese-Mandarin Translation</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yuzhi</namePart>
<namePart type="family">Liang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Fangqi</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-03</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the Second Workshop on Language Models for Low-Resource Languages (LoResLM 2026)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hansi</namePart>
<namePart type="family">Hettiarachchi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tharindu</namePart>
<namePart type="family">Ranasinghe</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Alistair</namePart>
<namePart type="family">Plum</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Paul</namePart>
<namePart type="family">Rayson</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ruslan</namePart>
<namePart type="family">Mitkov</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mohamed</namePart>
<namePart type="family">Gaber</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Damith</namePart>
<namePart type="family">Premasiri</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Fiona</namePart>
<namePart type="given">Anting</namePart>
<namePart type="family">Tan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Lasitha</namePart>
<namePart type="family">Uyangodage</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Rabat, Morocco</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-377-7</identifier>
</relatedItem>
<abstract>Fine-tuning multilingual models for low-resource dialect translation frequently encounters a “plausibility over faithfulness” dilemma, resulting in severe semantic drift on dialect-specific tokens. We term this phenomenon the “Probability Trap,” where models prioritize statistical fluency over semantic fidelity. To address this, we propose MVS-Rank (Multi-View Scoring Reranking), a generate-then-rerank framework that decouples evaluation from generation. Our method assesses translation candidates through three complementary perspectives: (1) Source-Side Faithfulness via a Reverse Translation Model to anchor semantic fidelity; (2) Local Fluency using Masked Language Models to ensure syntactic precision; and (3) Global Fluency leveraging Large Language Models to capture discourse coherence. Extensive experiments on Cantonese-Mandarin benchmarks demonstrate that MVS-Rank achieves state-of-the-art performance, significantly outperforming strong fine-tuning baselines by effectively rectifying hallucinations while maintaining high fluency.</abstract>
<identifier type="citekey">liang-chen-2026-escaping</identifier>
<location>
<url>https://aclanthology.org/2026.loreslm-1.41/</url>
</location>
<part>
<date>2026-03</date>
<extent unit="page">
<start>471</start>
<end>483</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Escaping the Probability Trap: Mitigating Semantic Drift in Cantonese-Mandarin Translation
%A Liang, Yuzhi
%A Chen, Fangqi
%Y Hettiarachchi, Hansi
%Y Ranasinghe, Tharindu
%Y Plum, Alistair
%Y Rayson, Paul
%Y Mitkov, Ruslan
%Y Gaber, Mohamed
%Y Premasiri, Damith
%Y Tan, Fiona Anting
%Y Uyangodage, Lasitha
%S Proceedings of the Second Workshop on Language Models for Low-Resource Languages (LoResLM 2026)
%D 2026
%8 March
%I Association for Computational Linguistics
%C Rabat, Morocco
%@ 979-8-89176-377-7
%F liang-chen-2026-escaping
%X Fine-tuning multilingual models for low-resource dialect translation frequently encounters a “plausibility over faithfulness” dilemma, resulting in severe semantic drift on dialect-specific tokens. We term this phenomenon the “Probability Trap,” where models prioritize statistical fluency over semantic fidelity. To address this, we propose MVS-Rank (Multi-View Scoring Reranking), a generate-then-rerank framework that decouples evaluation from generation. Our method assesses translation candidates through three complementary perspectives: (1) Source-Side Faithfulness via a Reverse Translation Model to anchor semantic fidelity; (2) Local Fluency using Masked Language Models to ensure syntactic precision; and (3) Global Fluency leveraging Large Language Models to capture discourse coherence. Extensive experiments on Cantonese-Mandarin benchmarks demonstrate that MVS-Rank achieves state-of-the-art performance, significantly outperforming strong fine-tuning baselines by effectively rectifying hallucinations while maintaining high fluency.
%U https://aclanthology.org/2026.loreslm-1.41/
%P 471-483
Markdown (Informal)
[Escaping the Probability Trap: Mitigating Semantic Drift in Cantonese-Mandarin Translation](https://aclanthology.org/2026.loreslm-1.41/) (Liang & Chen, LoResLM 2026)
ACL