Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

An Empirical Study on Chinese Character Decomposition in Multiword Expression-Aware Neural Machine Translation

Lifeng Han, Gareth J. F. Jones, Alan F. Smeaton

TL;DR

The paper investigates how Chinese character decomposition—ranging from radicals to stroke-level pieces—affects MWE translation in zh→en neural machine translation. It introduces three models (Radical_Sem, Decompose_Rep, BiMWE_Term) spanning BiRNN and Transformer architectures to study semantic radical features, multi-level decomposition, and bilingual MWE augmentation. Through pilot and large-scale experiments with both automatic metrics and human assessments, the study finds that radical semantics can enhance word meaning representation and that deeper decomposition (especially with BiMWE augmentation) can improve MWE translation accuracy, albeit with nuanced performance depending on evaluation method. Expert validation highlights improvements in MWE translations that are not always captured by BLEU, underscoring the practical value of decomposition and BiMWE strategies for ideographic languages in high-resource MT settings.

Abstract

Word meaning, representation, and interpretation play fundamental roles in natural language understanding (NLU), natural language processing (NLP), and natural language generation (NLG) tasks. Many of the inherent difficulties in these tasks stem from Multi-word Expressions (MWEs), which complicate the tasks by introducing ambiguity, idiomatic expressions, infrequent usage, and a wide range of variations. Significant effort and substantial progress have been made in addressing the challenging nature of MWEs in Western languages, particularly English. This progress is attributed in part to the well-established research communities and the abundant availability of computational resources. However, the same level of progress is not true for language families such as Chinese and closely related Asian languages, which continue to lag behind in this regard. While sub-word modelling has been successfully applied to many Western languages to address rare words improving phrase comprehension, and enhancing machine translation (MT) through techniques like byte-pair encoding (BPE), it cannot be applied directly to ideograph language scripts like Chinese. In this work, we conduct a systematic study of the Chinese character decomposition technology in the context of MWE-aware neural machine translation (NMT). Furthermore, we report experiments to examine how Chinese character decomposition technology contributes to the representation of the original meanings of Chinese words and characters, and how it can effectively address the challenges of translating MWEs.

An Empirical Study on Chinese Character Decomposition in Multiword Expression-Aware Neural Machine Translation

TL;DR

The paper investigates how Chinese character decomposition—ranging from radicals to stroke-level pieces—affects MWE translation in zh→en neural machine translation. It introduces three models (Radical_Sem, Decompose_Rep, BiMWE_Term) spanning BiRNN and Transformer architectures to study semantic radical features, multi-level decomposition, and bilingual MWE augmentation. Through pilot and large-scale experiments with both automatic metrics and human assessments, the study finds that radical semantics can enhance word meaning representation and that deeper decomposition (especially with BiMWE augmentation) can improve MWE translation accuracy, albeit with nuanced performance depending on evaluation method. Expert validation highlights improvements in MWE translations that are not always captured by BLEU, underscoring the practical value of decomposition and BiMWE strategies for ideographic languages in high-resource MT settings.

Abstract

Word meaning, representation, and interpretation play fundamental roles in natural language understanding (NLU), natural language processing (NLP), and natural language generation (NLG) tasks. Many of the inherent difficulties in these tasks stem from Multi-word Expressions (MWEs), which complicate the tasks by introducing ambiguity, idiomatic expressions, infrequent usage, and a wide range of variations. Significant effort and substantial progress have been made in addressing the challenging nature of MWEs in Western languages, particularly English. This progress is attributed in part to the well-established research communities and the abundant availability of computational resources. However, the same level of progress is not true for language families such as Chinese and closely related Asian languages, which continue to lag behind in this regard. While sub-word modelling has been successfully applied to many Western languages to address rare words improving phrase comprehension, and enhancing machine translation (MT) through techniques like byte-pair encoding (BPE), it cannot be applied directly to ideograph language scripts like Chinese. In this work, we conduct a systematic study of the Chinese character decomposition technology in the context of MWE-aware neural machine translation (NMT). Furthermore, we report experiments to examine how Chinese character decomposition technology contributes to the representation of the original meanings of Chinese words and characters, and how it can effectively address the challenges of translating MWEs.

Paper Structure

This paper contains 20 sections, 11 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Lexical Granularity for MT
  4. Chinese Character Decomposition in NLP
  5. Investigating MWEs in MT
  6. Model Designs
  7. Pilot Study: $Radical\_Sem$
  8. Model: $Decompose\_Rep$
  9. Model: $BiMWE\_Term$
  10. Experimental Settings and Evaluations
  11. Pilot Study Evaluation: $Radical\_Sem$
  12. Broad Evaluation Metrics
  13. Evaluations on the Single-reference Development Set NIST06
  14. Evaluations on the Multi-reference Testing Set NIST08
  15. Model Evaluations: $Decompose\_Rep$ & $BiMWE\_Term$
  16. ...and 5 more sections

Figures (11)

  • Figure 1: Character radical 木 (mù, wood/tree) from pictograph: the root of semantic representation.
  • Figure 2: Character decomposition examples of qiáo and liáng
  • Figure 3: Chinese sentence example with word segmentation (delimiter '/'), character sequence, radical-level decomposition and pronunciations.
  • Figure 4: A mapping between English and Chinese on compositionality.
  • Figure 5: BiRNN based NMT architecture DBLP:journals/corr/BahdanauCB14 using multi-embeddings of (word, character, radical).
  • ...and 6 more figures