Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Typologically-Informed Candidate Reranking for LLM-based Translation into Low-Resource Languages

Nipuna Abeykoon, Ashen Weerathunga, Pubudu Wijesinghe, Parameswari Krishnamurthy

TL;DR

The paper tackles systematic typological bias in LLM-driven translation for low-resource languages by proposing the Universal Metalinguistic Framework (UMF), a model-agnostic, inference-time metalinguistic layer that uses a 16-dimension typological profile to guide candidate reranking. It combines a Semantic Constraint Layer for lexical disambiguation with a Typological Scoring component that enforces target-language constraints, controlled by a tunable mixing parameter. Across nine language pairs and 341 sentences, UMF shows a strong link between typological distance and interception rate, delivering high Gain-Risk performance in structurally distant languages while exposing limitations in morphologically dense or conservatively treated languages, as validated by human judgments and a detailed error taxonomy. The work demonstrates that typology-informed reranking can improve structural correctness without parallel data or retraining, offering practical value for deploying MT to under-resourced languages and guiding future refinements in linguistic representations and evaluation beyond surface similarity metrics.

Abstract

Large language models trained predominantly on high-resource languages exhibit systematic biases toward dominant typological patterns, leading to structural non-conformance when translating into typologically divergent low-resource languages. We present a framework that leverages linguistic typology to improve translation quality without parallel training data or model retraining. The framework consists of two components: the Universal Metalinguistic Framework (UMF), which represents languages as structured profiles across 16 typological dimensions with divergence-weighted scoring, and the Computational Engine, which operates through linguistic disambiguation during generation and typological compliance scoring during selection. Evaluation across nine language pairs demonstrates intervention rates strongly correlating with typological distance from English. In experiments on 341 English sentences each having different morphological and syntactic phenomena, the framework shows an intervention precision of 48.16% for conservatively treated languages, 28.15% for morphologically dense languages, and 86.26% for structurally profiled languages. The framework requires no parallel training data and operates with any LLM capable of producing multiple candidate outputs, enabling practical deployment for under-resourced languages.

Typologically-Informed Candidate Reranking for LLM-based Translation into Low-Resource Languages

TL;DR

The paper tackles systematic typological bias in LLM-driven translation for low-resource languages by proposing the Universal Metalinguistic Framework (UMF), a model-agnostic, inference-time metalinguistic layer that uses a 16-dimension typological profile to guide candidate reranking. It combines a Semantic Constraint Layer for lexical disambiguation with a Typological Scoring component that enforces target-language constraints, controlled by a tunable mixing parameter. Across nine language pairs and 341 sentences, UMF shows a strong link between typological distance and interception rate, delivering high Gain-Risk performance in structurally distant languages while exposing limitations in morphologically dense or conservatively treated languages, as validated by human judgments and a detailed error taxonomy. The work demonstrates that typology-informed reranking can improve structural correctness without parallel data or retraining, offering practical value for deploying MT to under-resourced languages and guiding future refinements in linguistic representations and evaluation beyond surface similarity metrics.

Abstract

Large language models trained predominantly on high-resource languages exhibit systematic biases toward dominant typological patterns, leading to structural non-conformance when translating into typologically divergent low-resource languages. We present a framework that leverages linguistic typology to improve translation quality without parallel training data or model retraining. The framework consists of two components: the Universal Metalinguistic Framework (UMF), which represents languages as structured profiles across 16 typological dimensions with divergence-weighted scoring, and the Computational Engine, which operates through linguistic disambiguation during generation and typological compliance scoring during selection. Evaluation across nine language pairs demonstrates intervention rates strongly correlating with typological distance from English. In experiments on 341 English sentences each having different morphological and syntactic phenomena, the framework shows an intervention precision of 48.16% for conservatively treated languages, 28.15% for morphologically dense languages, and 86.26% for structurally profiled languages. The framework requires no parallel training data and operates with any LLM capable of producing multiple candidate outputs, enabling practical deployment for under-resourced languages.
Paper Structure (46 sections, 13 equations, 7 figures, 10 tables)

This paper contains 46 sections, 13 equations, 7 figures, 10 tables.

Table of Contents

  1. Introduction
  2. Background and Related Work
  3. Typology in Metalinguistic and MT Frameworks: Prior Approaches
  4. Typological Bias in LLMs
  5. Low-Resource Machine Translation
  6. Candidate Reranking and Self-Evaluation
  7. Quality Estimation
  8. Controllable Generation
  9. The UMF Framework and Computational Engine
  10. Overview
  11. Language Profile Structure
  12. Typological Divergence Calculation
  13. Directive Vector Construction
  14. Semantic Constraint Layer
  15. Typological Scoring
  16. ...and 31 more sections

Figures (7)

  • Figure 1: UMF Framework Architecture. The source text is processed through an LLM to generate N candidates. The Universal Metalinguistic Framework (right, orange-dashed) computes a 16-dimensional directive vector from source and target language profiles. The Computational Engine (center, green-dashed) applies dual-layer evaluation: semantic constraints for lexical disambiguation and typological scoring for structural compliance.
  • Figure 2: UMF Translation Pipeline. The seven-stage process transforms source text into typologically-compliant translations through linguistic analysis, divergence quantification, multi-candidate generation, and dual-layer evaluation.
  • Figure 3: Divergence vector visualization for English $\rightarrow$ Sinhala. Word order (SVO$\rightarrow$SOV), case marking, and information structure show maximum divergence (0.8--1.0), while classifiers and evidentiality show zero divergence (both languages lack these features).
  • Figure 4: Typological distance spectrum of evaluated languages from English. Languages are grouped into three clusters based on their combined word order and morphological divergence. High-divergence languages (red) require the most structural transformation during translation.
  • Figure 5: Gain-Risk Ratio distribution across nine target languages. Languages above the 1.0 threshold (vertical line) show net positive impact from UMF reranking. Hindi and Chinese achieve the highest efficiency, while Thai, Tamil, Swahili, and Sinhala show ratios well below 1.0, indicating that errors outweigh improvements in these languages.
  • ...and 2 more figures