MITRA-Swarm: Injecting Explicit Linguistic Structure into Foundation Models for Classical Machine Translation
Organizations: Purdue University · UC Berkeley · University of Illinois
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 (, same model family as the generators) driving an internal critique loop, and an independent reporting judge (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 MQM against a monolithic baseline of (, , 95% CI ), and isolated discourse restoration alone accounts for MQM (), outperforming a call-count-matched self-refinement baseline by MQM (). Severe per-verse failures (MQM) fall from to , at roughly the per-verse inference cost of the baseline. Additionally, scores its own family's outputs - MQM higher than 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.
Figures & tables
| Register | Description | Pop. | Sample | Voice-preserv. |
|---|---|---|---|---|
| epic | Itihāsa/Purāṇa narrative | 1,720 | 310 | – |
| sutra | Philosophical sūtra, commentary | 1,488 | 268 | ✓ |
| vedic | Vedic saṃhitā/upaniṣad; archaic | 1,000 | 180 | – |
| kavya | Ornate metrical poetry | 844 | 152 | – |
| legal | Dharmaśāstra, nibandha | 500 | 90 | ✓ |
| Total | 5,552 | 1,000 |
| Configuration | Prompt tok. | Compl. tok. | Total/verse | Latency (s) |
|---|---|---|---|---|
| baseline | 91 | 49 | 140 | 2.5 |
| self_refine | 164 | 89 | 253 | 4.9 |
| bo_n | 455 | 258 | 713 | 12.8 |
| full_swarm | 828 | 462 | 1290 | 7.1 |
| Configuration | MQM | SD | P10 | P25 | Min | % 80 | |
|---|---|---|---|---|---|---|---|
| baseline | 1000 | 85.27 | 5.67 | 80.0 | 84.0 | 35 | 9.0 |
| baseline_dup | 1000 | 85.27 | 5.53 | 80.0 | 84.0 | 35 | 8.5 |
| placebo_rejudge | 1000 | 85.26 | 5.71 | 80.0 | 84.0 | 35 | 8.9 |
| bo_n | 1000 | 85.58 | 5.51 | 80.0 | 84.0 | 40 | 8.7 |
| self_refine | 1000 | 85.39 | 5.56 | 80.0 | 84.0 | 35 | 8.3 |
| guide_only | 1000 | 85.23 | 5.65 | 80.0 | 84.0 | 40 | 8.9 |
| Contrast | MQM | 95% CI | W/L/T | |||
|---|---|---|---|---|---|---|
| Full swarm vs monolithic baseline | 999 | +0.63 ∗ | [+0.30, +0.97] | 0.002 | +0.12 | 397/294/308 |
| Re-judging alone (placebo) | 1000 | -0.01 † | [-0.07, +0.01] | 1.000 | -0.02 | 2/2/996 |
| Full swarm vs best-of-N (compute matched) | 999 | +0.32 | [-0.05, +0.67] | 0.460 | +0.06 | 363/333/303 |
| Full swarm vs self-refine (specialisation) | 999 | +0.51 ∗ | [+0.16, +0.85] | 0.029 | +0.09 | 389/301/309 |
| Morphosemantic guide | 999 | -0.01 † | [-0.27, +0.23] | 1.000 | -0.00 | 170/179/650 |
| Ellipsis restoration | 999 | +0.44 ∗ | [+0.12, +0.73] | 0.029 | +0.09 | 323/232/444 |
| Judge diagnostic | Value |
|---|---|
| Reporting judge | openai:gpt-4o-mini |
| Judge failure rate | 0.41% |
| Arithmetic disagrees with rubric | 96.4% (mean = 16.37) |
| Test–retest SD (repeated calls) | 1.45 |
| Self-preference range (min–max, 12 conditions) | +1.94 to +6.79 |
| Tool | Calls | Hit rate | Errors | Latency (ms) |
|---|---|---|---|---|
| TRANSLATION_MEMORY | 5513 | 0.05 | 0 | 513 |
Appendix figures & tables7 assets
Supplementary material from the paper’s appendix.
Appendix
| Source | indriyārtheṣu sarveṣu na prasajyeta kāmataḥ / atiprasaktiṃ caiteṣāṃ manasā saṃnivartayet // |
|---|---|
| Gold Reference | “…let him carefully obviate an excessive attachment to them, by (reflecting on their worthlessness in) his heart.” |
| no_ellipsis (guard inactive) | “…furthermore, one should turn the mind away from excessive attachment to these.” |
| full_swarm ( legal passive) | “…furthermore, the mind should be turned away from excessive attachment to these.” |
| Configuration | Guide | Ellip. | Crit. | Tools | TM | QA | Notes |
|---|---|---|---|---|---|---|---|
| baseline | – | – | – | – | – | – | – |
| baseline_dup | – | – | – | – | – | – | seed |
| placebo_rejudge | – | – | – | – | – | – | re-judge only |
| bo_n | – | – | – | – | – | – | |
| self_refine | – | – | ✓ | – | – | ✓ | unspecialised |
| guide_only | ✓ | – | – | – | – | – | – |
| Register | baseline | baseline_dup | bo_n | full_swarm | guide_ellipsis | guide_only | no_critique | no_ellipsis | no_guide | no_tools | placebo_rejudge | self_refine |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| epic | 86.0 | 85.8 | 85.8 | 86.0 | 85.8 | 85.9 | 85.8 | 86.0 | 86.4 | 86.0 | 86.0 | 85.9 |
| kavya | 85.4 | 85.5 | 85.3 | 85.8 | 85.7 | 85.4 | 85.7 | 85.4 | 85.7 | 85.8 | 85.4 | 85.4 |
| legal | 85.1 | 84.7 | 85.0 | 85.4 | 85.1 | 85.0 | 85.1 | 84.8 | 85.2 | 85.4 | 85.1 | 84.6 |
| sutra | 85.7 | 85.8 | 85.9 | 86.0 | 86.0 | 85.5 | 86.0 | 85.8 | 86.2 | 86.0 | 85.7 | 86.1 |
| vedic | 83.6 | 83.9 | 83.7 | 84.5 | 84.6 | 83.4 | 84.7 | 83.5 | 84.4 | 84.6 | 83.6 | 83.7 |
| Configuration | Prompt tok. | Compl. tok. | Total/verse | Latency (s) |
|---|---|---|---|---|
| baseline | 91 | 49 | 140 | 2.5 |
| baseline_dup | 91 | 48 | 140 | 2.4 |
| bo_n | 455 | 258 | 713 | 12.8 |
| full_swarm | 828 | 462 | 1290 | 7.1 |
| guide_ellipsis | 738 | 430 | 1169 | 5.1 |
| guide_only | 482 | 389 | 871 | 2.5 |
| Configuration | addition | mistranslation | omission | untranslated | awkward | grammar |
|---|---|---|---|---|---|---|
| baseline | 0.07 | 1.62 | 1.35 | 0.02 | 0.02 | 0.13 |
| baseline_dup | 0.07 | 1.62 | 1.36 | 0.02 | 0.02 | 0.13 |
| bo_n | 0.07 | 1.63 | 1.37 | 0.01 | 0.02 | 0.13 |
| full_swarm | 0.06 | 1.68 | 1.31 | 0.01 | 0.02 | 0.13 |
| guide_ellipsis | 0.06 | 1.69 | 1.31 | 0.01 | 0.02 | 0.13 |
| guide_only | 0.06 | 1.66 | 1.32 | 0.02 | 0.02 | 0.15 |
| Judge diagnostic | Value |
|---|---|
| Reporting judge | openai:gpt-4o-mini |
| Judge failure rate | 0.41% |
| Judge arithmetic disagrees with rubric | 96.4% (mean = 16.37) |
| Test–retest SD (same input, repeated calls) | 1.45 |
| Self-preference, baseline ( ) | +2.84 |
| Self-preference, baseline_dup ( ) | +2.72 |
Machine Translation
Machine translation develops computational methods for converting text, speech, and visual content between languages while preserving meaning and purpose.
See all 283 papers