arXiv cs.AI· Kayshav Bhardwaj, Pragun Seth, Kush Bhardwaj, Vijay Nagasamy, Kurt Keutzer·· 1 小时前AI 评分35
MITRA-Swarm：向基础模型注入显式语言结构以实现古典梵文机器翻译
MITRA-Swarm: Injecting Explicit Linguistic Structure into Foundation Models for Classical Machine Translation
AI 导读
MITRA-Swarm 是一种多智能体架构，为古典梵文翻译注入显式语言结构，包含 Sandhi 拆分 agent、gemma-4-26B-A4B-it 与 gemma-2-mitra-it 译者和语篇省略恢复器。
正文 · 原文
Abstract:Classical Sanskrit poses three challenges unresolved by single-call monolithic models: morphophonemic fusion Sandhi, dense nominal compounding Samāsa, and discourse-level zero-anaphora forced by poetic meter. We present MITRA-Swarm, a multi-agent architecture injecting explicit linguistic structure into an otherwise monolithic pipeline: a Sandhi-splitting agent, a parallel array of generalist gemma-4-26B-A4B-it and domain-specialized gemma-2-mitra-it translators, and a discourse-aware Ellipsis Restorer that consults corpus-derived register tags to avoid inventing active-voice subjects in impersonal legal and philosophical registers. Two separated judges evaluate quality: a routing judge ($J_{\text{route}}$, same model family as the generators) driving an internal critique loop, and an independent reporting judge ($J_{\text{report}}=$gpt-4o-mini, no shared vendor lineage) scoring every output exactly once. On a stratified benchmark of 1,000 verses across five registers (23,990 scored verdicts, 12-condition pre-registered ablation), the full swarm reaches $85.90$ MQM against a monolithic baseline of $85.27$ ($\Delta=+0.63$, $p_{\text{adj}}=0.0018$, 95% CI $[+0.30,+0.97]$), and isolated discourse restoration alone accounts for $+0.44$ MQM ($p_{\text{adj}}=0.0288$), outperforming a call-count-matched self-refinement baseline by $+0.51$ MQM ($p_{\text{adj}}=0.0288$). Severe per-verse failures (MQM$<80$) fall from $9.0\%$ to $7.6\%$, at roughly $9\times$ the per-verse inference cost of the baseline. Additionally, $J_{\text{route}}$ scores its own family's outputs $2$-$7$ MQM higher than $J_{\text{report}}$ on every one of the 12 conditions, empirically motivating the two-judge split; and a preflight screen of a 192,151-entry retrieval index against all 12 evaluation works verifies 0 reference leaks across 269 runtime checks. These are small gains on a metric saturating near 85-86.
| Comments: | 17 pages, accepted to NeurIPS 2026 LP4FM Workshop |
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2611.14675 [cs.CL] |
| (or arXiv:2611.14675v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2611.14675 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Kayshav Bhardwaj [view email]
[v1]
Tue, 10 Nov 2026 19:31:07 UTC (56 KB)
来源：arXiv cs.AI · arxiv.org