mcaligner¶
mcaligner aligns the words of two parsed versions of the same text, for
example an Arabic translation of Ephesians and its Greek source. Sentences
(here verses) are paired by sent_id, and each word link joins a word of
text A to a word of text B, written Ugarit style: 3-5S (sure) or 12-18P
(probable). The links feed a bilingual lexicon: which lemmas translate which.
A.conllu + B.conllu (+ silver.jsonl)
│
▼
rules ──▶ aligner ──▶ scorer ──▶ rules ... (run)
│
▼
pred.jsonl ──evaluate──▶ P/R/F1, AER
└──lexicon──▶ lemma → aligned lemmas
It is built on MCParser: the
sentences and the links live in MCParser’s corpus store, and the agents reach
them through MCParser’s MCP server. The
same server works from Claude Desktop (mcaligner mcp config).
Main features:
Code before models: rules link identical symbols, transliterated proper names and words whose translation is overwhelming in the existing links, each measured for precision.
Batched, compact agent tasks: several sentence pairs per task, one model request when the evidence in the prompt is enough, answers as short lines that code checks and applies.
A confidence and an author for every link; links below a threshold go back to the aligner.
Model tiers (default
anthropic:claude-opus-5-5for the aligner, Sonnet 5.5 for the scorer) with per-role overrides, prompt caching, and a token and cost report after every run.
Guide
Reference
Acknowledgements¶
The research leading to this software was conducted as part of the BICROSS project (‘The Significance of Bilingual Manuscripts for Detecting Cross-Language Interaction in the New Testament Tradition’), which has received funding from the European Research Council (ERC) under the European Union’s Horizon Europe Research and Innovation Programme (Grant Agreement No. 101043730 – BICROSS – ERC-2021-COG). Funded by the European Union. Views and opinions expressed are however those of the author only and do not necessarily reflect those of the European Union or the European Education and Culture Executive Agency (EACEA). Neither the European Union nor EACEA can be held responsible for them.
Author: Robert Turnbull (Melbourne Data Analytics Platform, University of Melbourne). Sister projects: MCParser and Hexapla.