# Cross-lingual sentence retrieval on Tatoeba Parallel 14 language pairs Source: [Cross-lingual sentence retrieval on Tatoeba Parallel 14 language pairs]() Collection index: [llms.txt]() Data updated: 2026-02-26T05:29:00.781Z Evaluation of cross-lingual sentence retrieval performance across 14 language pairs using the Tatoeba dataset, reporting per-language and average scores. A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. | Scope | Recorded value | | --- | --- | | Task | Cross-lingual sentence retrieval | | Dataset (including any stated split) | Tatoeba Parallel 14 language pairs | | Metric | Accuracy | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 28 of 28 | | Coverage | All results currently returned for this benchmark | Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol. ## Metrics - [Accuracy](): Higher is better; unit: PERCENT. - [Retrieval Score (ar)](): Higher is better; unit: PERCENT. - [Retrieval Score (bg)](): Higher is better; unit: PERCENT. - [Retrieval Score (zh)](): Higher is better; unit: PERCENT. - [Retrieval Score (de)](): Higher is better; unit: PERCENT. - [Retrieval Score (el)](): Higher is better; unit: PERCENT. - [Retrieval Score (fr)](): Higher is better; unit: PERCENT. - [Retrieval Score (hi)](): Higher is better; unit: PERCENT. - [Retrieval Score (ru)](): Higher is better; unit: PERCENT. - [Retrieval Score (es)](): Higher is better; unit: PERCENT. - [Retrieval Score (sw)](): Higher is better; unit: PERCENT. - [Retrieval Score (th)](): Higher is better; unit: PERCENT. - [Retrieval Score (tr)](): Higher is better; unit: PERCENT. - [Retrieval Score (ur)](): Higher is better; unit: PERCENT. - [Retrieval Score (vi)](): Higher is better; unit: PERCENT. - [Average Retrieval Score](): Higher is better; unit: PERCENT. - [Average Accuracy@1](): Higher is better; unit: PERCENT. ## Recorded results | Method | Accuracy | Result date | Report paper | | --- | --- | --- | --- | | InfoXLM | 80.6 | Not recorded | 68fa6fb31e209131cea46f22 | | InfoXLM | 77.8 | Not recorded | 68fa6fb31e209131cea46f22 | | XLM-P+ | 77.2 | Not recorded | 68fa6fb31e209131cea46f22 | | XLM-E | 74.4 | Not recorded | 68fa6fb31e209131cea46f22 | | XLM-P+ | 73.2 | Not recorded | 68fa6fb31e209131cea46f22 | | XLM-E | 72.3 | Not recorded | 68fa6fb31e209131cea46f22 | | InfoXLM (w/o XLCO) | 65.3 | Not recorded | 68fa6fb31e209131cea46f22 | | InfoXLM (w/o XLCO) | 64.6 | Not recorded | 68fa6fb31e209131cea46f22 | | XLM-P | 64 | Not recorded | 68fa6fb31e209131cea46f22 | | XLM-P | 61.7 | Not recorded | 68fa6fb31e209131cea46f22 | | XLM-R | 59.5 | Not recorded | 68fa6fb31e209131cea46f22 | | XLM-R | 57.6 | Not recorded | 68fa6fb31e209131cea46f22 | | XLM-E (w/o TRTD) | 55.8 | Not recorded | 68fa6fb31e209131cea46f22 | | XLM-E (w/o TRTD) | 55.1 | Not recorded | 68fa6fb31e209131cea46f22 | ### InfoXLM Accuracy: 80.6. Result ID: 6990fa1c3a6c31484904bf74. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### InfoXLM Accuracy: 77.8. Result ID: 6990fa1c3a6c31484904bf7b. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### XLM-P+ Accuracy: 77.2. Result ID: 6990fa1c3a6c31484904bf7e. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"post-training":"MLM + InfoNCE"}. This record does not assert independent reproduction. ### XLM-E Accuracy: 74.4. Result ID: 6990fa1c3a6c31484904bf72. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### XLM-P+ Accuracy: 73.2. Result ID: 6990fa1c3a6c31484904bf77. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"post-training":"MLM + InfoNCE"}. This record does not assert independent reproduction. ### XLM-E Accuracy: 72.3. Result ID: 6990fa1c3a6c31484904bf79. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### InfoXLM (w/o XLCO) Accuracy: 65.3. Result ID: 6990fa1c3a6c31484904bf7c. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"variant":"without XLCO"}. This record does not assert independent reproduction. ### InfoXLM (w/o XLCO) Accuracy: 64.6. Result ID: 6990fa1c3a6c31484904bf75. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"variant":"without XLCO"}. This record does not assert independent reproduction. ### XLM-P Accuracy: 64. Result ID: 6990fa1c3a6c31484904bf76. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### XLM-P Accuracy: 61.7. Result ID: 6990fa1c3a6c31484904bf7d. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### XLM-R Accuracy: 59.5. Result ID: 6990fa1c3a6c31484904bf71. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### XLM-R Accuracy: 57.6. Result ID: 6990fa1c3a6c31484904bf78. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction. ### XLM-E (w/o TRTD) Accuracy: 55.8. Result ID: 6990fa1c3a6c31484904bf73. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"variant":"without TRTD"}. This record does not assert independent reproduction. ### XLM-E (w/o TRTD) Accuracy: 55.1. Result ID: 6990fa1c3a6c31484904bf7a. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"variant":"without TRTD"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.