Prospects for Computer-Assisted Dialect Adaption

David J. Weber, William C. MannACL 1979 · DOI

Abstract

This paper investigates the feasibility of automating text adaptation between closely related dialects of Central Peruvian Quechua. By employing systematic phonological and morphological rules, the researchers developed a pipeline capable of performing the "heavy lifting" of translation. The results indicate that computer-assisted adaptation produces highly comprehensible text, significantly reducing the burden on human editors.

Introduction

Adapting documents across related dialects is traditionally a manual, labor-intensive process requiring rare bidialectal expertise. In highly agglutinative languages like Quechua, where words are formed by complex suffixation, simple string substitution is insufficient. Systematic differences in sound changes, pluralization markers, and morphological constraints necessitate a more sophisticated computational approach. This study explores whether a rule-based system can handle these systematic changes effectively enough to be a viable tool for dialect adaptation.

Methodology: The Analysis-Synthesis Pipeline

The core of the proposed system is a two-stage word-by-word pipeline that converts text from a Source Dialect (SD) to a Target Dialect (TD).

1. Analysis Stage

The word analyzer decomposes SD words into their constituent roots and suffixes. This involves:

  • Recursive Search: Using a dictionary to identify potential components.
  • Morphological Constraints: Applying rules to filter out illegal suffix combinations.
  • Morphophonemic Processing: Accounting for spelling variations caused by neighboring morphemes.
  • Plurality Handling: Identifying and abstracting plural markers for later re-insertion.

2. Synthesis Stage

The word synthesizer reconstructs the word in the target dialect:

  • Root Mapping: Roots are converted using Regular Sound Changes (RSC's) derived from "Proto-Quechua" forms or through direct lookup tables for non-cognates.
  • Re-pluralization: The system places plural markers according to the specific rules of the TD.
  • Allomorph Selection: The correct suffix variants are chosen based on TD phonology before final concatenation.
The word analyzer and synthesizer pipeline showing the transformation of Quechua words from source to target dialect.

Experimental Setup and Results

The system was tested using a 40-page corpus adapted from one source dialect into five target dialects of Central Peruvian Quechua (Panao, Huallaga, Dos-de-Mayo, Llata, and Junín).

The evaluation focused on accuracy and human effort reduction:

  • Accuracy: The program correctly adapted approximately 760 morphemes per 1,000 words.
  • Human Effort: Native speakers required only 190 additional manual changes per 1,000 words to finalize the text.
  • Comprehensibility: Output text was rated as "fully comprehensible" by native speakers, a significant improvement over the "marginally intelligible" source text.

Discussion

The findings support a "95% rule" for Quechua, where focusing on systematic morphological changes handles the vast majority of adaptation needs. By prioritizing systemic rules over rare idiosyncratic exceptions, the system remains computationally manageable while providing a powerful assistant for human editors. The use of Proto-Quechua forms as an intermediate bridge allows the system to scale across multiple dialects efficiently. While a "non-trivial residue" of complex syntactic or stylistic shifts remains, the tool successfully bridges the gap, allowing native speakers to finalize translations with minimal coaching.

Summary updated

BibTeX
@inproceedings{weber1979prospects,
  author = {David J. Weber and William C. Mann},
  title = {{Prospects for Computer-Assisted Dialect Adaption}},
  booktitle = {Annual Meeting of the Association for Computational Linguistics},
  year = {1979},
  doi = {10.3115/982163.982194},
}
Figure