# machine-translation

Machine translation uses computational systems to automatically translate text or speech from one natural language into another.

This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles.

## Do Rule-Based Tools Help LLMs Translate Endangered Uralic Languages in the Prompt or over MCP?

DevFeed: [Do Rule-Based Tools Help LLMs Translate Endangered Uralic Languages in the Prompt or over MCP?](<https://devfeed.tech/articles/do-rule-based-tools-help-llms-translate-endangered-uralic-languages-in-the-prompt-or-over-mcp-137448.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.19439>)

Author: Iaroslav Chelombitko, Mika H\\"am\\"al\\"ainen

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [machine-translation](<https://devfeed.tech/topics/machine-translation.md>), [prompt](<https://devfeed.tech/topics/prompt.md>), [MSP MCP](<https://devfeed.tech/topics/msp-mcp.md>)

Tags: [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [languages](<https://devfeed.tech/tags/languages.md>), [llms](<https://devfeed.tech/tags/llms.md>), [machine-translation](<https://devfeed.tech/tags/machine-translation.md>), [mcp](<https://devfeed.tech/tags/mcp.md>), [model-context-protocol](<https://devfeed.tech/tags/model-context-protocol.md>), [model-context-protocol-mcp](<https://devfeed.tech/tags/model-context-protocol-mcp.md>), [prompt](<https://devfeed.tech/tags/prompt.md>), [tool-calling](<https://devfeed.tech/tags/tool-calling.md>), [tools](<https://devfeed.tech/tags/tools.md>), [translation](<https://devfeed.tech/tags/translation.md>), [translations](<https://devfeed.tech/tags/translations.md>)

### AI overview

Adding rule-based resources improved translation acceptance in the expert-reviewed evaluation: accepted outputs rose from 13 to 43 of 272 when analyses were placed in the prompt, and to 55 with tools attached. The reported gain came from bilingual dictionary glosses. For Udmurt, whose dictionary lacks an English side, prompt-based access produced no gain; for the other three languages, English glosses covered 74-80% of source words. MCP tool use was harder to make work: 14 of 31 open-weight models never called a tool, including 11 unable to produce a usable call, and half of calls named unsupported languages. The evaluation covered 816 sentences across three conditions, and the authors say they release the dataset, outputs, and call logs.

### Source excerpt

arXiv:2611.19439v1 Announce Type: cross Abstract: Komi-Zyrian, Moksha, Erzya and Udmurt have mature rule-based resources, among them finite-state analysers, constraint grammars and bilingual dictionaries, which language models do not use. We ask whether giving a model access to these resources improves machine translation from the four languages into English, and whether it matters how the access is provided: as analyses computed in advance and written into the prompt, or as tools that the model calls itself through the Model Context Protocol (MCP). We evaluate manually: an expert linguist read every English output of the strongest model under three conditions, 816 sentences in all, accepted or rejected each one, and named the main error in every rejected translation. With the analyses in the prompt, the number of accepted translations rises from 13 to 43 out of 272; with the tools attached it rises to 55. The gain comes from the bilingual dictionary. For Udmurt, whose dictionary has no English side, the prompt-based condition brings no gain at all, while for the other three languages an English gloss reaches the prompt for 74 to 80% of the source words. Tool calling is also much harder to get working than prompt injection. Of 31 open-weight models, 14 never call a tool, 11 of them because they cannot produce a usable call, and half of the calls that are made name a language the toolkit does not support. We release the data set, every model output and all tool-call logs.

## Not What a Child Expressed: Auditing the Sign-to-Text Safety Interface in Child-Facing AI

DevFeed: [Not What a Child Expressed: Auditing the Sign-to-Text Safety Interface in Child-Facing AI](<https://devfeed.tech/articles/not-what-a-child-expressed-auditing-the-sign-to-text-safety-interface-in-child-facing-ai-139243.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.07519>)

Author: Muhammad Rafiullah Memon, Viet Vo, Wanlun Ma, Yang Xiang

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [machine-translation](<https://devfeed.tech/topics/machine-translation.md>), [audit](<https://devfeed.tech/topics/audit.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [auditing](<https://devfeed.tech/tags/auditing.md>), [case-study](<https://devfeed.tech/tags/case-study.md>), [classifiers](<https://devfeed.tech/tags/classifiers.md>), [comparison](<https://devfeed.tech/tags/comparison.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [cs-cy](<https://devfeed.tech/tags/cs-cy.md>), [errors](<https://devfeed.tech/tags/errors.md>), [interface](<https://devfeed.tech/tags/interface.md>), [roles](<https://devfeed.tech/tags/roles.md>), [safety](<https://devfeed.tech/tags/safety.md>), [schema](<https://devfeed.tech/tags/schema.md>), [sign](<https://devfeed.tech/tags/sign.md>), [slt](<https://devfeed.tech/tags/slt.md>), [taxonomy](<https://devfeed.tech/tags/taxonomy.md>), [text](<https://devfeed.tech/tags/text.md>), [translation](<https://devfeed.tech/tags/translation.md>), [trust-and-safety](<https://devfeed.tech/tags/trust-and-safety.md>)

### AI overview

The paper proposes a Deaf-informed pre-deployment audit of how sign-to-text errors may affect child safety systems that act on translated text. It reports finding no publicly documented joint evaluation of sign language translation (SLT) and child-facing safety tools, and says the leading deployed SLT model was neither trained nor formally evaluated on signers under 18. Fluent translations that change negation, participant roles, secrecy, urgency, or help-seeking could alter safety decisions. The proposed audit includes a failure taxonomy, sanitised scenario schema, four comparison conditions, and four outcome measures; Auslan is the planned first case study.

### Source excerpt

arXiv:2610.07519v2 Announce Type: replace-cross Abstract: Automatic sign language translation (SLT) has entered consumer products, turning American Sign Language into English text for dictation, messaging, and queries put to a conversational assistant. Child-facing AI and platform trust-and-safety tooling decide on text, using filters on minor accounts and grooming classifiers that score chat messages. A signing child who uses SLT therefore reaches these safeguards through a translation. We found no publicly documented system in which the two have been jointly evaluated, and the leading deployed SLT model was neither trained nor formally evaluated on signers under 18. Errors that alter negation, participant roles, secrecy, urgency or help-seeking could change a safety decision without disturbing fluency. This paper proposes a Deaf-informed pre-deployment audit of that boundary, with a failure taxonomy, a sanitised scenario schema, four comparison conditions, and four outcome measures. Auslan is the planned first case study.

## When Simpler Is Better: Evaluating Translation Pipelines for Medieval Latin Manuscripts

DevFeed: [When Simpler Is Better: Evaluating Translation Pipelines for Medieval Latin Manuscripts](<https://devfeed.tech/articles/when-simpler-is-better-evaluating-translation-pipelines-for-medieval-latin-manuscripts-134992.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2607.03836>)

Author: Nguyen Kim Hai Bui, Md. Easin Arafat, Tam\\'as G\\'abor Orosz, Mufti Mahmud

Published: 2026-11-16T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [machine-translation](<https://devfeed.tech/topics/machine-translation.md>), [vlm](<https://devfeed.tech/topics/vlm.md>), [Retrieval Augmented Generation (RAG)](<https://devfeed.tech/topics/retrieval-augmented-generation-rag.md>)

Tags: [abstract](<https://devfeed.tech/tags/abstract.md>), [aggregate](<https://devfeed.tech/tags/aggregate.md>), [all](<https://devfeed.tech/tags/all.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [benchmarking](<https://devfeed.tech/tags/benchmarking.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [better](<https://devfeed.tech/tags/better.md>), [capabilities](<https://devfeed.tech/tags/capabilities.md>), [complexity](<https://devfeed.tech/tags/complexity.md>), [component](<https://devfeed.tech/tags/component.md>), [core](<https://devfeed.tech/tags/core.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [domain](<https://devfeed.tech/tags/domain.md>), [error](<https://devfeed.tech/tags/error.md>), [error-propagation](<https://devfeed.tech/tags/error-propagation.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [failure](<https://devfeed.tech/tags/failure.md>), [pipelines](<https://devfeed.tech/tags/pipelines.md>), [translation](<https://devfeed.tech/tags/translation.md>)

### AI overview

On the CATMuS Latin benchmark, domain-specific OCR reduced character error rate by up to 4.3x versus general-purpose vision-language models despite using far fewer parameters. The authors report that a specialized OCR model feeding directly into a VLM produced better aggregate translation quality than tested multi-component pipelines; adding retrieval-augmented generation or post-OCR correction degraded results through prompt saturation and error propagation. They also introduce the Interpres-Parallel-Corpus, with 1,383 aligned manuscript image lines, transcriptions, and expert translations. The findings are based on the reported experiments and address challenges such as shorthand, ligatures, degraded parchment, archaic vocabulary, and noisy input.

### Source excerpt

arXiv:2607.03836v2 Announce Type: replace-cross Abstract: Despite remarkable progress in machine translation, Vision Language Models (VLMs) struggle on historical manuscripts, a domain that stresses core Natural Language Processing (NLP) capabilities: low-resource transliteration, archaic vocabulary, and noisy input signals. We present a systematic framework for evaluating the full image-to-translation pipeline on medieval Latin manuscripts, a setting in which scribal shorthand, ligatures, and parchment degradation expose failure modes that are invisible in clean-text benchmarks. Benchmarking on the CATMuS Latin dataset reveals a specialization gap: domain-specific Optical Character Recognition (OCR) models reduce character error rate by up to 4.3$\times$ compared to general-purpose VLMs, despite operating at orders of magnitude fewer parameters. We introduce the Interpres-Parallel-Corpus (IPC), a novel dataset comprising 1,383 aligned manuscript image lines, transcriptions, and expert translations, to our knowledge the first of its kind for medieval Latin. Our experiments uncover a complexity paradox: the simplest pipeline, a specialized OCR model feeding directly into a VLM, outperforms all multi-component variants. Adding retrieval-augmented generation (RAG) or post-OCR correction introduces prompt saturation and error propagation that degrade aggregate translation quality. These findings offer both a new benchmark and practical guidance for deploying translation systems in low-resource historical settings.

## MITRA-Swarm: Injecting Explicit Linguistic Structure into Foundation Models for Classical Machine Translation

DevFeed: [MITRA-Swarm: Injecting Explicit Linguistic Structure into Foundation Models for Classical Machine Translation](<https://devfeed.tech/articles/mitra-swarm-injecting-explicit-linguistic-structure-into-foundation-models-for-classical-machine-translation-132036.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.14675>)

Author: Kayshav Bhardwaj, Pragun Seth, Kush Bhardwaj, Vijay Nagasamy, Kurt Keutzer

Published: 2026-11-13T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [machine-translation](<https://devfeed.tech/topics/machine-translation.md>)

Tags: [ablation](<https://devfeed.tech/tags/ablation.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agent-architecture](<https://devfeed.tech/tags/agent-architecture.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [array](<https://devfeed.tech/tags/array.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [cost](<https://devfeed.tech/tags/cost.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [discourse](<https://devfeed.tech/tags/discourse.md>), [domain](<https://devfeed.tech/tags/domain.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [exactly-once](<https://devfeed.tech/tags/exactly-once.md>), [failures](<https://devfeed.tech/tags/failures.md>), [foundation](<https://devfeed.tech/tags/foundation.md>), [foundation-models](<https://devfeed.tech/tags/foundation-models.md>), [machine](<https://devfeed.tech/tags/machine.md>), [machine-translation](<https://devfeed.tech/tags/machine-translation.md>), [models](<https://devfeed.tech/tags/models.md>), [structure](<https://devfeed.tech/tags/structure.md>), [swarm](<https://devfeed.tech/tags/swarm.md>), [translation](<https://devfeed.tech/tags/translation.md>)

### AI overview

MITRA-Swarm reports a modest improvement in Sanskrit translation quality over a monolithic baseline: 85.90 versus 85.27 MQM, with severe per-verse failures dropping from 9.0% to 7.6%, at roughly nine times the inference cost. Its pipeline splits Sandhi, runs generalist and Sanskrit-specialized translators in parallel, then restores omitted discourse subjects using register tags to avoid inserting active-voice subjects in impersonal legal and philosophical texts. Discourse restoration alone contributed +0.44 MQM, and the full system exceeded a call-count-matched self-refinement baseline by 0.51 MQM. The authors use a routing judge for critique and a separate reporting judge; the routing judge scored its own model family's outputs 2-7 MQM higher across all 12 conditions, motivating that separation. The reported gains are small on a metric near 85-86. These findings cover the supplied abstract's benchmark of 1,000 verses across five registers.

### Source excerpt

arXiv:2611.14675v1 Announce Type: cross Abstract: Classical Sanskrit poses three challenges unresolved by single-call monolithic models: morphophonemic fusion Sandhi, dense nominal compounding Sam\=asa, 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.

## Text Grouping Changes Completion-Time Comparisons in Long-Form Simultaneous Speech Translation

DevFeed: [Text Grouping Changes Completion-Time Comparisons in Long-Form Simultaneous Speech Translation](<https://devfeed.tech/articles/text-grouping-changes-completion-time-comparisons-in-long-form-simultaneous-speech-translation-126284.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.12501>)

Author: Hieu Minh Truong, Dung D. Le

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [machine-translation](<https://devfeed.tech/topics/machine-translation.md>), [Speech Processing](<https://devfeed.tech/topics/speech-processing.md>), [asr](<https://devfeed.tech/topics/asr.md>)

Tags: [audio](<https://devfeed.tech/tags/audio.md>), [automatic](<https://devfeed.tech/tags/automatic.md>), [availability](<https://devfeed.tech/tags/availability.md>), [blocks](<https://devfeed.tech/tags/blocks.md>), [change](<https://devfeed.tech/tags/change.md>), [changes](<https://devfeed.tech/tags/changes.md>), [compare](<https://devfeed.tech/tags/compare.md>), [comparison](<https://devfeed.tech/tags/comparison.md>), [comparisons](<https://devfeed.tech/tags/comparisons.md>), [completion](<https://devfeed.tech/tags/completion.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [effects](<https://devfeed.tech/tags/effects.md>), [english](<https://devfeed.tech/tags/english.md>), [fixed](<https://devfeed.tech/tags/fixed.md>), [gap](<https://devfeed.tech/tags/gap.md>), [history](<https://devfeed.tech/tags/history.md>), [speech](<https://devfeed.tech/tags/speech.md>), [speech-translation](<https://devfeed.tech/tags/speech-translation.md>), [text](<https://devfeed.tech/tags/text.md>), [time](<https://devfeed.tech/tags/time.md>), [translation](<https://devfeed.tech/tags/translation.md>)

### AI overview

Grouping translation updates before synthesis changed StreamAtt's measured completion lead, with opposite effects across synthesizers: four-second grouping increased the mean lead from 106.6 to 229.1 seconds with XTTS-v2, but reduced it from 169.5 to 146.5 seconds with Piper. Both produced shorter audio; differences in playback idle accounted for the opposing results. The directions recurred on a separately selected twenty-talk panel. In continuing CosyVoice 3 sessions, grouping left paired audio unchanged and delayed completion on 19 of 20 blocks. The findings indicate that speech-timing comparisons should specify how text is grouped for synthesis as well as which synthesizer is used. Recognition and automatic word timing describe outputs but do not establish translation accuracy or intelligibility.

### Source excerpt

arXiv:2611.12501v1 Announce Type: cross Abstract: Does grouping translation updates before speech synthesis change the measured completion gap between two translators? We compare StreamAtt with its fixed-audio-history baseline on ten English-German talks. For each translation, we keep the intended text and its availability times fixed and vary how updates are grouped for synthesis. Four-second grouping increases StreamAtt's mean completion lead from 106.6 to 229.1 seconds with XTTS-v2, but reduces it from 169.5 to 146.5 seconds with Piper. Both produce shorter audio; different changes in playback idle account for the opposite effects. These directions recur on a separately selected twenty-talk panel. In continuing CosyVoice 3 sessions, grouping instead leaves paired audio unchanged and delays completion on nineteen of twenty blocks. A speech-timing comparison therefore needs to specify how text enters synthesis, not just which synthesizer is used. Recognition and automatic word timing describe the outputs, but do not establish translation accuracy or intelligibility.

## A Large-Scale Dataset for Cross-Lingual Multi-Jurisdiction Patent Drafting from Patent Cooperation Treaty Applications

DevFeed: [A Large-Scale Dataset for Cross-Lingual Multi-Jurisdiction Patent Drafting from Patent Cooperation Treaty Applications](<https://devfeed.tech/articles/a-large-scale-dataset-for-cross-lingual-multi-jurisdiction-patent-drafting-from-patent-cooperation-treaty-applications-126194.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.11891>)

Author: Trung Phuong Le, Hirofumi Nonaka, Koji Marusaki, Katsuhito Sudoh, Seiya Kawano

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [machine-translation](<https://devfeed.tech/topics/machine-translation.md>), [Localization (l10n)](<https://devfeed.tech/topics/localization.md>), [few-shot](<https://devfeed.tech/topics/few-shot.md>)

Tags: [benchmark](<https://devfeed.tech/tags/benchmark.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [languages](<https://devfeed.tech/tags/languages.md>), [office](<https://devfeed.tech/tags/office.md>), [patent](<https://devfeed.tech/tags/patent.md>), [translation](<https://devfeed.tech/tags/translation.md>)

### AI overview

The paper frames cross-lingual patent drafting as generating claim sets for a target language, jurisdiction, and prosecution stage, capturing changes beyond content-preserving translation. It links Japan Patent Office and USPTO records across 884,179 PCT-derived patent families, yielding 4,067,168 directional claim-set instances in six configurations, with four core configurations in the main benchmark. Commercial machine translation and zero- and few-shot prompting capture part of the task; supervised adaptation is the strongest overall baseline.

### Source excerpt

arXiv:2611.11891v1 Announce Type: cross Abstract: Securing patent protection across countries requires drafting claims separately for each national patent office, in the relevant language and according to its legal and stylistic conventions. Patent claims can therefore differ across languages and jurisdictions because of translation, target-office drafting practices, and prosecution-stage changes. We formulate cross-lingual patent drafting as generating plausible candidate claim sets for a specified target language, jurisdiction, and prosecution stage from a source claim set in another language. The task models jurisdiction- and stage-specific transformations beyond content-preserving translation. We construct a large-scale Japanese-English dataset by linking Japan Patent Office (JPO) and United States Patent and Trademark Office (USPTO) records through 884,179 Patent Cooperation Treaty (PCT)-derived patent families. Across six released configurations, these families yield 4,067,168 directional claim-set instances; four core configurations form the main benchmark. Commercial machine translation and zero- and few-shot prompting capture part of the task, while supervised adaptation provides the strongest overall baseline, showing that task-specific patterns can be learned from linked cross-jurisdictional claim sets.

## Exploring the world's languages with Language Explorer

DevFeed: [Exploring the world's languages with Language Explorer](<https://devfeed.tech/articles/exploring-the-world-s-languages-with-language-explorer-126470.md>)

Original publisher: [Read original article](<https://research.google/blog/exploring-the-worlds-languages-with-language-explorer/>)

Published: 2026-11-09T20:29:39Z

Content type: article

Language: en

Sources: [The latest research from Google](<https://devfeed.tech/sources/the-latest-research-from-google.md>)

Topics: [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [machine-translation](<https://devfeed.tech/topics/machine-translation.md>)

Tags: [2006](<https://devfeed.tech/tags/2006.md>), [2026](<https://devfeed.tech/tags/2026.md>), [accessible](<https://devfeed.tech/tags/accessible.md>), [africa](<https://devfeed.tech/tags/africa.md>), [ai](<https://devfeed.tech/tags/ai.md>), [android](<https://devfeed.tech/tags/android.md>), [arabic](<https://devfeed.tech/tags/arabic.md>), [berlin](<https://devfeed.tech/tags/berlin.md>), [collaboration](<https://devfeed.tech/tags/collaboration.md>), [collect](<https://devfeed.tech/tags/collect.md>), [communication](<https://devfeed.tech/tags/communication.md>), [communities](<https://devfeed.tech/tags/communities.md>), [coverage](<https://devfeed.tech/tags/coverage.md>), [data](<https://devfeed.tech/tags/data.md>), [explorer](<https://devfeed.tech/tags/explorer.md>), [exploring](<https://devfeed.tech/tags/exploring.md>), [language](<https://devfeed.tech/tags/language.md>), [languages](<https://devfeed.tech/tags/languages.md>), [natural-language-processing](<https://devfeed.tech/tags/natural-language-processing.md>), [open-source-models-datasets](<https://devfeed.tech/tags/open-source-models-datasets.md>), [world](<https://devfeed.tech/tags/world.md>)

### AI overview

Language Explorer is an interactive web platform for discovering and visualizing linguistic diversity worldwide. Developed as part of Google's 1,000 Languages Initiative, it brings together data on thousands of languages using the open-source LinguaMeta dataset, and is intended for researchers, educators, and language enthusiasts. It is also featured as an interactive installation in the Museum of Communication Berlin exhibition *Real Talk! Language. Power. World.*, running in Berlin through February 2028 and at the Museum of Communication Frankfurt through September 2029. The source also points readers to an online companion and the platform itself.

### Source excerpt

Natural Language Processing

## Reasoning Before Translation: Enhancing Legal Machine Translation with Structured Reasoning

DevFeed: [Reasoning Before Translation: Enhancing Legal Machine Translation with Structured Reasoning](<https://devfeed.tech/articles/reasoning-before-translation-enhancing-legal-machine-translation-with-structured-reasoning-123190.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2607.19181>)

Author: Aixiu An, Michael Jungo, Eloi Eynard, Mark Drenhaus, Andreas Fischer, Jean Hennebert, S\\'ebastien Rumley

Published: 2026-11-09T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [machine-translation](<https://devfeed.tech/topics/machine-translation.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Reinforcement learning](<https://devfeed.tech/topics/reinforcement-learning.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>)

Tags: [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [gemma-3](<https://devfeed.tech/tags/gemma-3.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [legal](<https://devfeed.tech/tags/legal.md>), [machine-translation](<https://devfeed.tech/tags/machine-translation.md>), [performance](<https://devfeed.tech/tags/performance.md>), [quality](<https://devfeed.tech/tags/quality.md>), [qwen3-5](<https://devfeed.tech/tags/qwen3-5.md>), [reasoning](<https://devfeed.tech/tags/reasoning.md>), [reasoning-models](<https://devfeed.tech/tags/reasoning-models.md>), [reinforcement-learning](<https://devfeed.tech/tags/reinforcement-learning.md>)

### AI overview

The study compares retraining approaches for legal machine translation on Swiss multilingual statutes, evaluating Qwen3.5 4B and 9B and Gemma 3 12B against frontier reasoning models. It reports that reinforcement learning with verifiable rewards outperforms supervised fine-tuning, substantially improves small base models, and brings them close to--but still below--state-of-the-art reasoning models. Retraining gains diminish as model size increases. The supplied abstract does not provide detailed evaluation methods or quantitative results.

### Source excerpt

arXiv:2607.19181v2 Announce Type: replace-cross Abstract: Neural machine translation (NMT) in the legal domain is a linguistically and conceptually demanding task, primarily due to the complexity of legal language and the high level of precision it requires. The recent emergence of reasoning-capable language models opens new possibilities for tackling such challenges. They add to a set of other previously proposed techniques to enhance the translation quality, which includes supervised fine-tuning and reinforcement learning. In this work, we perform a comparison between these various approaches. More particularly, we evaluate small language models such as Qwen3.5 4B, Qwen3.5 9B, and Gemma 3 12B enhanced with various re-training paradigms and compare their performances against frontier reasoning models. We focus on the Swiss legal system, which -- with its unique multilingual statutes -- offers a particularly challenging testbed for reasoning-augmented models. Our results show that the quality of small ``base'' models can be greatly enhanced, and that reinforcement learning with verifiable rewards can be applied to NMT in the legal domain and surpasses the translation quality of supervised fine-tuning. The performance of enhanced small models is close to the one of state-of-the-art reasoning models yet remains inferior. We also note that re-training paradigms yield diminishing returns as model size increase. The code and models are publicly available at https://github.com/aixiuxiuxiu/Legal-MT-SFT-RL.

## Are These Programmers Ripe Enough to Eat? Evaluating How LLMs Handle Lexical Knowledge Conflicts in Translation

DevFeed: [Are These Programmers Ripe Enough to Eat? Evaluating How LLMs Handle Lexical Knowledge Conflicts in Translation](<https://devfeed.tech/articles/are-these-programmers-ripe-enough-to-eat-evaluating-how-llms-handle-lexical-knowledge-conflicts-in-translation-123455.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.09785>)

Author: Dongxu Wu, Sara Court, Micha Elsner

Published: 2026-11-09T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [machine-translation](<https://devfeed.tech/topics/machine-translation.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>)

Tags: [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [knowledge](<https://devfeed.tech/tags/knowledge.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [llms](<https://devfeed.tech/tags/llms.md>), [translation](<https://devfeed.tech/tags/translation.md>)

### AI overview

The study tests how nine language models handle conflicts between familiar Mandarin word meanings and new meanings supplied in translation prompts. In this controlled task, models relied more on supplied contextual knowledge for less frequent words and when sentence context supported the new meanings. Prompt framing could reduce that reliance, yet seven of nine models favored the supplied meanings even when told to ignore them; within model families, larger models used them more often. Follow-up experiments found that weaker modal wording reduced reliance, repetition effects depended on prompt layout, and reasoning effects varied by model. The abstract reports these results as motivation for better methods to detect and evaluate lexical conflicts in translation.

### Source excerpt

arXiv:2611.09785v1 Announce Type: cross Abstract: External lexical resources can help large language models (LLMs) translate texts in low-resource languages and specialized domains, but these resources may conflict with models' prior lexical knowledge. To isolate this conflict, we construct a controlled Mandarin-to-English translation task in which familiar source words are assigned new meanings within and across semantic categories. We treat these prompt-supplied meanings as contextual knowledge (CK) and systematically evaluate the effects of word frequency, sentence context, and prompt framing on the translated outputs. Across nine models, CK reliance is greater for less frequent source words and in sentence contexts that support the supplied meanings. Prompt framing can reduce CK reliance, but seven of nine models favor CK even when explicitly instructed to ignore the supplied meanings. Within model families, larger models are more likely than smaller models to utilize CK in their outputs. Three follow-up experiments further show that modal weakening lowers CK reliance, the effect of repetition depends on prompt layout, and the effect of reasoning varies across models. Overall, our controlled study reveals how CK reliance varies across conditions and models, motivating new methods to detect and evaluate lexical conflicts and guide models' lexical choices in LLM-based translation.

## From Collaboration to Intelligent Workplace: What's New for Cisco Partners in 2026

DevFeed: [From Collaboration to Intelligent Workplace: What's New for Cisco Partners in 2026](<https://devfeed.tech/articles/from-collaboration-to-intelligent-workplace-what-s-new-for-cisco-partners-in-2026-121720.md>)

Original publisher: [Read original article](<https://blogs.cisco.com/partner/from-collaboration-to-intelligent-workplace-whats-new-for-cisco-partners-in-2026>)

Author: Kristyn Hogan

Published: 2026-11-06T19:30:35Z

Content type: news

Language: en

Sources: [Cisco Blogs](<https://devfeed.tech/sources/cisco-blogs.md>)

Topics: [Cisco](<https://devfeed.tech/topics/cisco.md>), [cisco networking](<https://devfeed.tech/topics/cisco-networking.md>), [ai security](<https://devfeed.tech/topics/ai-security.md>), [machine-translation](<https://devfeed.tech/topics/machine-translation.md>), [Cisco Meraki](<https://devfeed.tech/topics/cisco-meraki.md>), [ThousandEyes](<https://devfeed.tech/topics/thousandeyes.md>), [Cryptography](<https://devfeed.tech/topics/cryptography.md>), [zoom](<https://devfeed.tech/topics/zoom.md>)

Tags: [1-4](<https://devfeed.tech/tags/1-4.md>), [2](<https://devfeed.tech/tags/2.md>), [2026](<https://devfeed.tech/tags/2026.md>), [3](<https://devfeed.tech/tags/3.md>), [about](<https://devfeed.tech/tags/about.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-native](<https://devfeed.tech/tags/ai-native.md>), [cisco](<https://devfeed.tech/tags/cisco.md>), [cisco-partner-summit](<https://devfeed.tech/tags/cisco-partner-summit.md>), [collaboration](<https://devfeed.tech/tags/collaboration.md>), [growth](<https://devfeed.tech/tags/growth.md>), [how-to](<https://devfeed.tech/tags/how-to.md>), [innovation](<https://devfeed.tech/tags/innovation.md>), [native](<https://devfeed.tech/tags/native.md>), [partner](<https://devfeed.tech/tags/partner.md>), [partners](<https://devfeed.tech/tags/partners.md>), [recap](<https://devfeed.tech/tags/recap.md>), [spaces](<https://devfeed.tech/tags/spaces.md>), [summit](<https://devfeed.tech/tags/summit.md>), [watch](<https://devfeed.tech/tags/watch.md>), [workplace](<https://devfeed.tech/tags/workplace.md>)

### AI overview

Cisco's 2026 partner pitch is to sell collaboration devices, networking and workplace spaces as one platform, with the article claiming this can produce 3x larger deals--and 6x when spaces telemetry brings Facilities and Corporate Real Estate into the sale. Announced offerings include RoomOS 27 features, support for Google Meet alongside Webex, Teams and Zoom, and room planning and monitoring tools. The AI-native Webex Suite adds meeting assistance and agents, with MCP support for customer-provided models and governance; Webex Calling customers can add cloud and AI features to UCM without forced migration. Partner incentives include higher Q2 CPI rebates, Contact Center in EA 3.0 and adoption data in PXP. The article recommends sharing the Acceleration Playbook, selecting an account for a combined portfolio pitch, and registering teams for training and events.

### Source excerpt

Discover how to turn intelligent workplace innovation into profitable growth. From Partner Summit, learn how Cisco's new AI-native collaboration and spaces solutions are helping partners drive 3x-6x larger deals. Watch the recap today.

## CardioLM: A multilingual suite of small language models for the cardiology domain

DevFeed: [CardioLM: A multilingual suite of small language models for the cardiology domain](<https://devfeed.tech/articles/cardiolm-a-multilingual-suite-of-small-language-models-for-the-cardiology-domain-120272.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.07940>)

Author: Manuela-Daniela Danu, Miguel Rodr\\'iguez-Ortega, Alberto Becerra-Tome, Pablo Arancibia, Fernando Gallego-Donoso, Lorenzo Pratesi, Xi Bai, Andrei-Damian Dinescu, Salvador Lima-L\\'opez, Andrea Guala, Sa

Published: 2026-11-06T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [ner](<https://devfeed.tech/topics/ner.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>), [machine-translation](<https://devfeed.tech/topics/machine-translation.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>)

Tags: [annotations](<https://devfeed.tech/tags/annotations.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [biomedical](<https://devfeed.tech/tags/biomedical.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [deployment](<https://devfeed.tech/tags/deployment.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [languages](<https://devfeed.tech/tags/languages.md>), [machine-translation](<https://devfeed.tech/tags/machine-translation.md>), [named-entity-recognition](<https://devfeed.tech/tags/named-entity-recognition.md>), [natural-language-processing](<https://devfeed.tech/tags/natural-language-processing.md>), [pretraining](<https://devfeed.tech/tags/pretraining.md>)

### AI overview

CardioCCC contains 508 Spanish cardiology case reports extended to Czech, Dutch, English, Italian, Romanian, and Swedish, with clinician-reviewed annotations for diseases, symptoms, procedures, and medications. CardioLM pairs language-specific CardioBERTa encoders, continued-pretrained on cardiology and biomedical text, with CardioNER models fine-tuned for entity recognition. Specialty pretraining lowered pseudo-perplexity by up to eightfold but improved strict span F1 by no more than 0.04, mainly where base models lacked biomedical coverage. Jointly trained multilingual XLM-RoBERTa-large models performed best in nearly every language and entity type, beating the strongest monolingual system by up to 0.13 F1. Relaxed boundary matching recovered more F1 for diseases, procedures, and symptoms than medications. Clinicians rated CardioNER higher than a rule-and-dictionary baseline (73% versus 26% mean rating).

### Source excerpt

arXiv:2611.07940v1 Announce Type: cross Abstract: Background: Cardiovascular care generates large volumes of clinical free-text documentation in local languages. However, clinical natural language processing (NLP) resources remain concentrated in English, and no existing benchmark is cardiology-specific, multilingual, and broad in entity coverage. Methods: We present two contributions. The first is CardioCCC, a corpus of 508 cardiology case reports originally published in Spanish and extended to Czech, Dutch, English, Italian, Romanian, and Swedish through machine translation, annotation projection, and clinician review, with annotations covering diseases, symptoms, procedures, and medications. The second is CardioLM, comprising CardioBERTa, a set of language-specific encoders obtained by continued pretraining on cardiology and biomedical corpora, and CardioNER, a set of named entity recognition (NER) models fine-tuned on CardioCCC. Results: Cardiology-specific pretraining reduced pseudo-perplexity by up to a factor of eight, yet changed strict span-level F1 by at most 0.04, with benefits concentrated in languages whose backbones lacked biomedical coverage. The jointly trained multilingual XLM-RoBERTa-large models achieved the best strict F1 in nearly every language and entity type, exceeding the strongest monolingual system by up to 0.13 F1. A large share of residual errors concerned boundary detection: relaxed matching recovered 0.10-0.23 F1 for disease, procedure, and symptom, but only 0.01-0.07 for medication. Clinicians rated CardioNER well above a rule- and dictionary-based baseline, with the mean rating across entity types rising from 26% to 73%. Conclusions: Specialty-specific pretraining is not the decisive factor for downstream entity recognition, whereas multilingual joint training yields the strongest results in nearly all settings and a simpler deployment path for multi-country clinical information extraction.

## Isharah-Selfie: Continuous Sign Language Recognition Dataset for One-handed Signing

DevFeed: [Isharah-Selfie: Continuous Sign Language Recognition Dataset for One-handed Signing](<https://devfeed.tech/articles/isharah-selfie-continuous-sign-language-recognition-dataset-for-one-handed-signing-118286.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.33200>)

Author: Ahmed Abul Hasanaath, Murtadha Aljubran, Sarah Alyami, Muhammad Haris Khan, Hamzah Luqman

Published: 2026-11-05T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [dataset](<https://devfeed.tech/topics/dataset.md>), [machine-translation](<https://devfeed.tech/topics/machine-translation.md>)

Tags: [continuous](<https://devfeed.tech/tags/continuous.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [language](<https://devfeed.tech/tags/language.md>), [recognition](<https://devfeed.tech/tags/recognition.md>), [sign](<https://devfeed.tech/tags/sign.md>)

### AI overview

Isharah-Selfie is an Arabic Sign Language dataset designed to reflect signing on front-facing smartphones in everyday selfie conditions. It contains 8,997 clips from 16 signers covering 1,493 sentences across domains including healthcare, education, and transportation. The videos capture one-handed signing, close viewpoints, partial body visibility, camera motion, varied environments, and different device resolutions; each has gloss annotations for continuous sign recognition and Arabic sentence translations. The dataset defines signer-independent and unseen-sentence evaluation protocols. Benchmarks with RGB and pose data indicate that current recognition and translation methods struggle with these conditions, especially when evaluated on unseen sentences, where compositional generalization is weak.

### Source excerpt

arXiv:2610.33200v2 Announce Type: replace Abstract: Current sign language recognition and translation benchmarks are commonly collected in controlled settings using fixed or studio-style cameras, particularly limiting their ability to reflect how signers communicate in everyday mobile scenarios. As such, this gap becomes increasingly evident when considering the Arabic Sign Language (ArSL), which is a primary sign language used across Arab countries. In this work, we introduce Isharah-Selfie, a large-scale ArSL dataset collected entirely using front-facing smartphone cameras under unconstrained selfie conditions. The dataset comprises 8,997 video clips performed by 16 signers across 1,493 unique sentences, covering several domains such as healthcare, education, and transportation. Unlike conventional datasets dominated by fixed-camera, two-handed signing, Isharah-Selfie captures realistic mobile communication characterized by one-handed signing, close-range viewpoints, partial body visibility, camera motion, unconstrained environments, and heterogeneous device resolutions. Each video is annotated with both gloss sequences for continuous sign language recognition (CSLR) and Arabic sentence translations for sign language translation (SLT). Moreover, we define signer-independent and unseen-sentences evaluation protocols to assess generalization across unseen signers and novel sentence compositions. Extensive benchmarks using RGB-based and pose-based data show that current CSLR and SLT methods remain challenged by selfie-style signing conditions, particularly under unseen-sentence evaluation, where compositional generalization remains weak. By releasing Isharah-Selfie, we aim to support research toward robust, user-centered sign language technologies that operate reliably in realistic smartphone-based communication settings. The Isharah-Selfie dataset is available on https://isharah-selfie.github.io/.

## Universal Conceptual Structure in Neural Translation: Probing NLLB-200's Multilingual Geometry

DevFeed: [Universal Conceptual Structure in Neural Translation: Probing NLLB-200's Multilingual Geometry](<https://devfeed.tech/articles/universal-conceptual-structure-in-neural-translation-probing-nllb-200-s-multilingual-geometry-116670.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2603.02258>)

Author: Kyle Elliott Mathewson

Published: 2026-11-04T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [machine-translation](<https://devfeed.tech/topics/machine-translation.md>), [concept-learning](<https://devfeed.tech/topics/concept-learning.md>), [scaling laws](<https://devfeed.tech/topics/scaling-laws.md>)

Tags: [color](<https://devfeed.tech/tags/color.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [hierarchy](<https://devfeed.tech/tags/hierarchy.md>), [language](<https://devfeed.tech/tags/language.md>), [meaning](<https://devfeed.tech/tags/meaning.md>), [model](<https://devfeed.tech/tags/model.md>), [neural](<https://devfeed.tech/tags/neural.md>), [semantic](<https://devfeed.tech/tags/semantic.md>), [space](<https://devfeed.tech/tags/space.md>), [structure](<https://devfeed.tech/tags/structure.md>), [translation](<https://devfeed.tech/tags/translation.md>)

### AI overview

In NLLB-200's encoder, multilingual representations show both language-specific shifts and shared conceptual structure. Across 135 languages and 101 Swadesh concepts, language shifts explain more variance for bare words than contextualized words (45.7% versus 13.7%); mean-centering improves concept separation in both settings, most strongly for bare words, with separation peaking at Layer 10. Semantic category explains substantially more Swadesh convergence variance than orthography or phonology, though the Swadesh advantage over other concrete nouns is mediated by subword compactness. Within language subfamilies, embedding distances track phonetic distances; colexified concepts are closer under reported controls, and color-term embeddings recover the Berlin and Kay hierarchy. The authors report releasing InterpretCognates, an interactive toolkit and reproducible suite.

### Source excerpt

arXiv:2603.02258v2 Announce Type: replace-cross Abstract: Do multilingual translation models organize meaning around a shared conceptual space, or do they primarily encode language identity and surface form? We probe the encoder geometry of Meta's 600M-parameter NLLB-200 model across 135 languages and 101 Swadesh concepts in contextualized carrier-frame and decontextualized bare-word regimes against null and confound controls. First, two-way variance partitioning reveals an additive language-shift plus conceptual-manifold geometry: language shifts account for 45.7% of variance on bare words vs. 13.7% in context, and mean-centering improves the between-to-within-concept distance ratio (vs. a 0.99 permutation null) by 2.14x on bare words (1.15 to 2.46; 2.27x across disjoint scripts) and 1.19x in context (2.25 to 2.69), with conceptual separation peaking at Layer 10 (mutual k-NN overlap jumping 4.35x). Second, semantic category explains 25.4% of Swadesh convergence variance ($F(7,93)=4.53$, $p=2.21\times 10^{-4}$; partial $F=7.74$ controlling for fertility and length)---over 20x orthography or phonology ($R^2\le 0.012$)---while the advantage of Swadesh over non-loanword concrete nouns ($d=0.54$--$0.69$) is mediated by Zipfian subword compactness ($p=7.43\times 10^{-7}$). Third, language distances track ASJP NLD and LDND phonetic distances within subfamilies under block-permutation and cross-script controls ($\rho_s=0.65$--$0.67$, block $p=10^{-4}$; global bare-word Mantel $\rho_s=0.24$, $p=0.001$), and CLICS$^3$ colexified concepts lie closer across string-distinct, within-category ($\rho_s=0.32$), and bare-word ($d=1.79$) controls ($d=1.43$, $p=1.55\times 10^{-33}$). Finally, basic color terms recover the Berlin and Kay hierarchy ($\rho_s=0.91$--$0.93$, $p\le 2.18\times 10^{-4}$), and semantic offsets exceed distance-matched controls ($p=3.55\times 10^{-5}$, $d=1.06$). We release InterpretCognates, an interactive toolkit and reproducible suite.

## FlowTranslator: Training-Free In-Image Machine Translation with Flow-Based Editing

DevFeed: [FlowTranslator: Training-Free In-Image Machine Translation with Flow-Based Editing](<https://devfeed.tech/articles/flowtranslator-training-free-in-image-machine-translation-with-flow-based-editing-117461.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.04303>)

Author: Zhixian Li, Xingsong Ye, Yongkun Du, Yunyan Xiao, Yanglin Huang, Pinxuan Chen, Jiyu Zhuang, Cong Xu, Zhineng Chen

Published: 2026-11-04T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [machine-translation](<https://devfeed.tech/topics/machine-translation.md>), [image editing](<https://devfeed.tech/topics/image-editing.md>), [ai-image-generator](<https://devfeed.tech/topics/ai-image-generator.md>), [Image](<https://devfeed.tech/topics/image.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>)

Tags: [bleu](<https://devfeed.tech/tags/bleu.md>), [comet](<https://devfeed.tech/tags/comet.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [flow](<https://devfeed.tech/tags/flow.md>), [free](<https://devfeed.tech/tags/free.md>), [image](<https://devfeed.tech/tags/image.md>), [llm](<https://devfeed.tech/tags/llm.md>), [machine-translation](<https://devfeed.tech/tags/machine-translation.md>), [ocr](<https://devfeed.tech/tags/ocr.md>), [prim](<https://devfeed.tech/tags/prim.md>), [text-to-image](<https://devfeed.tech/tags/text-to-image.md>), [training](<https://devfeed.tech/tags/training.md>), [translation](<https://devfeed.tech/tags/translation.md>)

### AI overview

FlowTranslator performs in-image machine translation without task-specific training: an OCR model and an LLM create a translation prompt for a text-to-image diffusion model, with Region Coupling Release altering noise within the text region early in generation and Coupling Expansion amplifying the edit mid-trajectory. Under EasyOCR read-back, it reports average PRIM BLEU of 20.6 and COMET of 57.2, substantially outperforming the training-based UniTranslator baseline. On whole-image ScenePair photographs translated into Chinese, it reports 40.1% exact-match accuracy and surpasses existing editing systems, some of which handle only text rendering rather than the complete translation task. These are abstract-reported results; no evaluation details beyond those stated are provided.

### Source excerpt

arXiv:2611.04303v1 Announce Type: cross Abstract: In-image machine translation (IIMT) replaces text in an image with its translation while preserving the surrounding visual context. Existing task-specific models achieve convincing performance but require massive paired training images before and after translation. As advanced text-to-image (T2I) models have reached a practically usable level of multilingual text generation, we argue that this capability can be transferred to IIMT and present FlowTranslator. It is a training-free IIMT framework that first employs an OCR model and an LLM to generate the translation prompt, which is injected into the T2I-based diffusion model. To optimize the accurate and natural generation of translated text, we introduce two trajectory controls: Region Coupling Release (RCR) partially decorrelates target-branch noise inside the text region early in the trajectory, while Coupling Expansion (CE) amplifies the emerging edit displacement mid-trajectory. Under the EasyOCR read-back protocol, FlowTranslator achieves average PRIM BLEU and COMET scores of 20.6 and 57.2, respectively, substantially outperforming the strong training-based baseline UniTranslator. On the more challenging and realistic whole-image ScenePair photographs translated into Chinese, it reaches 40.1\% exact-match accuracy and surpasses existing editing systems, some of which only address the text-rendering stage rather than the complete IIMT task. These results establish flow-based editing as a strong, fully training-free baseline for IIMT.

## Maswao: An automatically collected, prepared, and benchmarked dataset for the South African Sign language

DevFeed: [Maswao: An automatically collected, prepared, and benchmarked dataset for the South African Sign language](<https://devfeed.tech/articles/maswao-an-automatically-collected-prepared-and-benchmarked-dataset-for-the-south-african-sign-language-115398.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.02903>)

Author: Thapelo Sindane, Unarine Netshifhefhe, Vukosi Marivate, Abiodun Modupe, Herkulaas Combrink, Ketro Sithole

Published: 2026-11-03T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Language Engineering](<https://devfeed.tech/topics/language-engineering.md>), [machine-translation](<https://devfeed.tech/topics/machine-translation.md>), [asr](<https://devfeed.tech/topics/asr.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [managing large-scale text data in AI](<https://devfeed.tech/topics/managing-large-scale-text-data-in-ai.md>)

Tags: [asr](<https://devfeed.tech/tags/asr.md>), [audio](<https://devfeed.tech/tags/audio.md>), [automatically](<https://devfeed.tech/tags/automatically.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [bleu](<https://devfeed.tech/tags/bleu.md>), [collect](<https://devfeed.tech/tags/collect.md>), [collection](<https://devfeed.tech/tags/collection.md>), [crop](<https://devfeed.tech/tags/crop.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [framework](<https://devfeed.tech/tags/framework.md>), [gradient](<https://devfeed.tech/tags/gradient.md>), [gradient-descent](<https://devfeed.tech/tags/gradient-descent.md>), [labour](<https://devfeed.tech/tags/labour.md>), [language](<https://devfeed.tech/tags/language.md>), [languages](<https://devfeed.tech/tags/languages.md>), [model](<https://devfeed.tech/tags/model.md>), [models](<https://devfeed.tech/tags/models.md>), [motion](<https://devfeed.tech/tags/motion.md>), [sign](<https://devfeed.tech/tags/sign.md>)

### AI overview

Maswao is a weakly sentence-aligned South African Sign Language dataset built with minimal human intervention. The framework collects public SASL videos, including news broadcasts, converts accompanying audio to text with available ASR tools, crops the signing area, and uses gradient descent to detect non-motion valleys as sentence boundaries. On sign-to-text translation, Transformer models outperformed the recurrent baselines tested. The best Transformer scored BLEU-1 through BLEU-4 of 13.78, 5.21, 2.52, and 1.54, with ChrF 18.09 and ROUGE-L 10.51. The reported scores suggest the task remains open for further investigation.

### Source excerpt

arXiv:2611.02903v1 Announce Type: cross Abstract: Recent advancements in sign language processing can be attributed to developments in both large-scale dataset collection and algorithmic complexity in areas such as pose estimation techniques, signal processing through Transformer models, and model optimization. However, only a handful of sign languages, such as the American, British, and German, from over 200 globally recognized sign languages, reap the benefits of such advancements, while the rest lag behind. In this study, we aim to address this technological disparity by proposing a framework for the automatic collection and preparation of sign language datasets, with very minimal human-labour intervention, using the South African Sign Language (SASL) as a testbed. Specifically, we automatically collect SASL video content using public resources such as news broadcasts. Additionally, we convert their accompanying audio streams into text using available Automatic Speech Recognition (ASR) tools. Furthermore, we crop the sign language area in the videos and use gradient descent to locate valleys of non-motion to determine end-of-sentence points within long videos to create our weakly sentence-aligned parallel corpora -- the Maswao (Signs) dataset. Finally, we benchmark the weakly aligned dataset on the task of Sign-to-text Translation using Recurrent Neural Networks and Transformer models to compare their individual intricacies. Our results show that Transformer models proposed in the literature outperform recurrent baselines, with the best-performing Transformer achieving BLEU-1 to BLEU-4 scores of 13.78, 5.21, 2.52, and 1.54, respectively, as well as ChrF = 18.09 and ROUGE-L = 10.51. These results indicate that the task remains promising and warrants further investigations.

## NVIDIA Nemotron 3.5 Transcribe: A Complete Guide to the Speech AI Model

DevFeed: [NVIDIA Nemotron 3.5 Transcribe: A Complete Guide to the Speech AI Model](<https://devfeed.tech/articles/nvidia-nemotron-3-5-transcribe-a-complete-guide-to-the-speech-ai-model-107725.md>)

Original publisher: [Read original article](<https://hackernoon.com/nvidia-nemotron-35-transcribe-a-complete-guide-to-the-speech-ai-model?source=rss>)

Author: aimodels44

Published: 2026-10-30T05:19:14Z

Content type: article

Language: en

Sources: [HackerNoon](<https://devfeed.tech/sources/hackernoon.md>)

Topics: [Nemotron](<https://devfeed.tech/topics/nemotron.md>), [asr](<https://devfeed.tech/topics/asr.md>), [model architecture](<https://devfeed.tech/topics/model-architecture.md>), [LLM evaluation / benchmarking](<https://devfeed.tech/topics/llm-evaluation-benchmarking.md>), [machine-translation](<https://devfeed.tech/topics/machine-translation.md>), [Streaming](<https://devfeed.tech/topics/streaming.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [ai-model](<https://devfeed.tech/tags/ai-model.md>), [artificial-intelligence](<https://devfeed.tech/tags/artificial-intelligence.md>), [asr](<https://devfeed.tech/tags/asr.md>), [audio](<https://devfeed.tech/tags/audio.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [comparison](<https://devfeed.tech/tags/comparison.md>), [comparisons](<https://devfeed.tech/tags/comparisons.md>), [content-creation](<https://devfeed.tech/tags/content-creation.md>), [context](<https://devfeed.tech/tags/context.md>), [customer-success](<https://devfeed.tech/tags/customer-success.md>), [data-science](<https://devfeed.tech/tags/data-science.md>), [deep-learning](<https://devfeed.tech/tags/deep-learning.md>), [deployment](<https://devfeed.tech/tags/deployment.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [hardware](<https://devfeed.tech/tags/hardware.md>), [hybrid](<https://devfeed.tech/tags/hybrid.md>), [nemotron](<https://devfeed.tech/tags/nemotron.md>), [nemotron-3-5](<https://devfeed.tech/tags/nemotron-3-5.md>), [nemotron-3-5-transcribe](<https://devfeed.tech/tags/nemotron-3-5-transcribe.md>), [nemotron-transcribe-30b](<https://devfeed.tech/tags/nemotron-transcribe-30b.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [nvidia-nemotron-3-5](<https://devfeed.tech/tags/nvidia-nemotron-3-5.md>), [speech](<https://devfeed.tech/tags/speech.md>)

### AI overview

NVIDIA Nemotron 3.5 Transcribe 30B A4B is an offline, open-weight model for audio transcription, translation, language identification, speaker attribution, timestamps, and audio questions in 40 languages, with recordings up to one hour. Instructions and optional context can guide verbatim or cleaned transcripts, formatting, and translation targets. The README specifies one H100 80 GB deployment configuration; it does not describe real-time streaming, and internal chunking does not make it a streaming service. Reported mean WERs are 4.71 for English ASR, 10.98 for long-form ASR, and 4.62 for multilingual ASR on listed benchmarks, where Qwen3-ASR-1.7B and Parakeet-TDT-0.6B-v3 are identified as next-best open-weight models for the respective evaluations. These results do not establish performance on other workloads. A related Nemotron Speech Streaming English 0.6B is described as streaming-capable. The excerpt says vLLM word timestamps require its V1 model runner. This listing is the BF16 reference; FP8 and NVFP4 variants are also available. The excerpt does not include the referenced language support matrix.

### Source excerpt

NVIDIA-Nemotron-3.5-Transcribe-30B-A4B [https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B/resolve/main/architecture.png] Image source: AIM...

## Cross Lingual Abstractive Summarization for Marathi

DevFeed: [Cross Lingual Abstractive Summarization for Marathi](<https://devfeed.tech/articles/cross-lingual-abstractive-summarization-for-marathi-108049.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.38267>)

Author: Asang Dani, Shailesh R Sathe

Published: 2026-10-30T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [machine-translation](<https://devfeed.tech/topics/machine-translation.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>), [Transformers](<https://devfeed.tech/topics/transformers.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>)

Tags: [architectures](<https://devfeed.tech/tags/architectures.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [machine-translation](<https://devfeed.tech/tags/machine-translation.md>), [metrics](<https://devfeed.tech/tags/metrics.md>), [natural-language-processing](<https://devfeed.tech/tags/natural-language-processing.md>), [performance](<https://devfeed.tech/tags/performance.md>), [text-summarization](<https://devfeed.tech/tags/text-summarization.md>)

### AI overview

The study fine-tunes IndicTrans2 for cross-lingual summarization using training data built from the monolingual Marathi MahaSum corpus and the smaller cross-lingual PMIndiaSum dataset. Evaluation uses ROUGE, chrF, and chrF++, with primary focus on Marathi-to-English due to computational constraints; the authors also discuss applicability to English-to-Marathi. They report substantially better summarization performance than IndicBART- and mBART-50-based baselines from previous work. The supplied abstract does not give metric scores or results for the reverse direction.

### Source excerpt

arXiv:2610.38267v1 Announce Type: cross Abstract: Text summarization is an important task for advancing natural language processing (NLP) research and applications in Indic languages. Despite progress in extractive and abstractive summarization, both monolingual summarization (MS) and cross-lingual summarization (CLS) remain relatively underexplored for these languages. Recent advances in Transformer architectures, pretrained language models (PLMs), and dataset availability offer new opportunities to address this gap. In this work, we construct cross-lingual training data from the monolingual Marathi MahaSum corpus and combine it with the smaller cross-lingual PMIndiaSum dataset to fine-tune IndicTrans2, an Indic neural machine translation model. We evaluate the resulting models using ROUGE, chrF and chrF++ metrics, focusing primarily on Marathi-to-English summarization due to computational constraints. We also discuss the applicability of our approach to the English-to-Marathi direction. Our experiments demonstrate that leveraging a pretrained translation model yields substantially better summarization performance than the IndicBART and mBART-50 based baselines reported in previous work on cross-lingual summarization.

## Component Translation and Judge-Labelled Refusal: A Multilingual Safety Evaluation

DevFeed: [Component Translation and Judge-Labelled Refusal: A Multilingual Safety Evaluation](<https://devfeed.tech/articles/component-translation-and-judge-labelled-refusal-a-multilingual-safety-evaluation-109033.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.38181>)

Author: Yongxi Zhou, Junwei Yao, Wenbo Ye, Jiahui Ge

Published: 2026-10-30T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [LLM evaluation / benchmarking](<https://devfeed.tech/topics/llm-evaluation-benchmarking.md>), [machine-translation](<https://devfeed.tech/topics/machine-translation.md>), [Dataset development](<https://devfeed.tech/topics/dataset-development.md>)

Tags: [component](<https://devfeed.tech/tags/component.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [safety](<https://devfeed.tech/tags/safety.md>), [translation](<https://devfeed.tech/tags/translation.md>)

### AI overview

The study tests whether refusal rates change when the same language is assigned to different parts of a harmful request. It independently assigns English, Chinese, and Spanish to the request and operation components, producing nine configurations across 190 behaviours; paired comparisons hold behaviour, English framing, and translated-span count fixed, though component identity and non-English text quantity change together. Refusal is the primary outcome, scored by language-model judges, with unsafe compliance reported separately. For Gemini 2.5 Flash and DeepSeek, four reported refusal swaps are roughly 5.3 percentage points for three comparisons and -0.5 for the fourth; none passes the pre-specified six-test Holm correction. In an exploratory extension, Mistral Small 4's Spanish swap is 13.7 points (95% interval 8.4-18.9) and passes a twelve-test Holm sensitivity correction under the primary judge; a second judge estimates 12.1 points. GPT-5.4-mini estimates are within 2 points of zero. The authors recommend recording component languages and translated text quantity, matching behaviours at equal translated-span count, and interpreting refusal separately from unsafe compliance.

### Source excerpt

arXiv:2610.38181v1 Announce Type: cross Abstract: Whole-prompt translation evaluates a request in different languages but does not compare which component carries a language. We introduce a paired reversal: translate the complete request component or the complete operation component into the same target language, holding behaviour, English frame and translated-span count fixed. Component identity and non-English text quantity change together. Assigning English, Chinese and Spanish independently to the two components yields nine configurations on 190 of 300 sampled harmful behaviours retained after consensus annotation and automatic gating. Refusal scored by language-model judges is the primary outcome; unsafe compliance is reported separately. On Gemini 2.5 Flash and DeepSeek, three refusal swaps are about 5.3 percentage points and the fourth is -0.5; none survives the pre-specified six-test Holm correction. In an exploratory extension, Mistral Small 4 has a 13.7-point Spanish swap (95% interval [8.4, 18.9]), which survives the twelve-test Holm sensitivity correction under the primary judge, and 12.1 points under a second judge. GPT-5.4-mini estimates are within 2 points of zero. Component assignment supplies an additional evaluation dimension. Evaluations should record component languages and translated text quantity, comparing matched behaviours at equal translated-span count while interpreting refusal and unsafe compliance separately.

## One Confidence Threshold, Unequal Coverage: Language Gaps in Typed Decision Interfaces and How Translating Inputs Reduces Them

DevFeed: [One Confidence Threshold, Unequal Coverage: Language Gaps in Typed Decision Interfaces and How Translating Inputs Reduces Them](<https://devfeed.tech/articles/one-confidence-threshold-unequal-coverage-language-gaps-in-typed-decision-interfaces-and-how-translating-inputs-reduces-them-107946.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.37688>)

Author: Junwei Yao, Yongxi Zhou, Zike Yang, Zhijun Wang, Jianan Liu, Yanxian Chen, Wenbo Ye

Published: 2026-10-30T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [machine-translation](<https://devfeed.tech/topics/machine-translation.md>), [Data analysis](<https://devfeed.tech/topics/data-analysis.md>)

Tags: [apis](<https://devfeed.tech/tags/apis.md>), [classifiers](<https://devfeed.tech/tags/classifiers.md>), [confidence](<https://devfeed.tech/tags/confidence.md>), [coverage](<https://devfeed.tech/tags/coverage.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [gap](<https://devfeed.tech/tags/gap.md>), [interfaces](<https://devfeed.tech/tags/interfaces.md>), [language](<https://devfeed.tech/tags/language.md>), [study](<https://devfeed.tech/tags/study.md>), [translation](<https://devfeed.tech/tags/translation.md>)

### AI overview

Across three vendors' typed decision interfaces, one 0.9 confidence threshold produced lower automatic coverage in every tested language than in English. In a confirmatory study of 1,000 MASSIVE items across 51 languages, English cleared the threshold 8.6-11.7 percentage points more often on average, even for items answered correctly in both languages; automatically handled requests were rarely wrong. Reweighting languages to equal accuracy did not remove the gap, and ECE missed it on one interface. Exploratory analyses found English still had higher coverage with language-specific thresholds targeting 5% error. Translating inputs into English nearly closed the coverage gap at 0.9, but not at that error target; native African-language requests showed a gap in the same direction against human English translations. The authors recommend reporting coverage by language at the operating threshold.

### Source excerpt

arXiv:2610.37688v1 Announce Type: cross Abstract: Deployed classifiers often act on a prediction only when its confidence clears a threshold and pass the rest to a person. We ask whether one threshold serves all languages equally on typed decision interfaces, APIs that return a probability for each option of a fixed multiple-choice question. In a confirmatory study with a written analysis plan, three vendors' interfaces classified 1,000 MASSIVE items, each in 51 languages, with a threshold of 0.9. Even when a request is answered correctly in both English and another language, the English version clears the threshold more often, by 8.6-11.7 percentage points on average. Observed coverage, the share of requests handled automatically, is lower than English's in every other language, while the requests the gate handles are rarely wrong. The gap remains when every language is reweighted to the same accuracy, and a standard calibration metric (ECE) does not detect it on one interface. In exploratory analyses, English's coverage stays higher when each language gets its own threshold fitted to the same 5% error target. Translating inputs into English nearly closes the gap at 0.9 but leaves English ahead at that error target. Requests written natively in African languages show a gap in the same direction when English is their human translation. We recommend reporting per-language coverage at the operating threshold.

## Nvidia releases all-in-one adaptable Nemotron 3.5 audio transcription and translation model

DevFeed: [Nvidia releases all-in-one adaptable Nemotron 3.5 audio transcription and translation model](<https://devfeed.tech/articles/nvidia-releases-all-in-one-adaptable-nemotron-3-5-audio-transcription-and-translation-model-104541.md>)

Original publisher: [Read original article](<https://siliconangle.com/2026/10/28/nvidia-releases-all-in-one-adaptable-nemotron-3-5-audio-transcription-and-translation-model/>)

Author: Kyt Dotson

Published: 2026-10-28T15:00:59Z

Content type: news

Language: en

Sources: [SiliconANGLE](<https://devfeed.tech/sources/siliconangle.md>)

Topics: [Nemotron](<https://devfeed.tech/topics/nemotron.md>), [machine-translation](<https://devfeed.tech/topics/machine-translation.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Nvidia](<https://devfeed.tech/topics/nvidia.md>), [Local AI](<https://devfeed.tech/topics/local-ai.md>), [Digital Public Good](<https://devfeed.tech/topics/digital-public-goods.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [ai-model](<https://devfeed.tech/tags/ai-model.md>), [artificial-intelligence](<https://devfeed.tech/tags/artificial-intelligence.md>), [audio](<https://devfeed.tech/tags/audio.md>), [blackwell](<https://devfeed.tech/tags/blackwell.md>), [data-pipeline](<https://devfeed.tech/tags/data-pipeline.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [granary-v2](<https://devfeed.tech/tags/granary-v2.md>), [nasdaq-nvda](<https://devfeed.tech/tags/nasdaq-nvda.md>), [nemotron](<https://devfeed.tech/tags/nemotron.md>), [nemotron-3-5-transcribe](<https://devfeed.tech/tags/nemotron-3-5-transcribe.md>), [news](<https://devfeed.tech/tags/news.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [speakers](<https://devfeed.tech/tags/speakers.md>), [the-latest](<https://devfeed.tech/tags/the-latest.md>), [top-story-4](<https://devfeed.tech/tags/top-story-4.md>), [top-story-5](<https://devfeed.tech/tags/top-story-5.md>), [transcription](<https://devfeed.tech/tags/transcription.md>), [translation](<https://devfeed.tech/tags/translation.md>)

### AI overview

Nvidia's Nemotron 3.5 Transcribe combines speech detection, transcription, speaker labeling and question answering in one model pass. The open-weight 30-billion-parameter mixture-of-experts model (4 billion active parameters) supports 40 languages, up to eight speakers and questions about as much as 100 hours of audio. Nvidia says it runs on one GPU, from edge devices to data-center hardware; a Blackwell Server Edition GPU using NVFP4 and one-quarter model compression processed 100 hours in under three minutes. Reported word error rate is 30.8% out of the box, 20.8% with careful prompting and 11.8% after fine-tuning on an Indian medical speech benchmark. Prompts can request translations, speaker labels, meeting decisions or different transcript styles, and supplied jargon can improve recognition. Nvidia also released Granary v2 and an open NeMo Curator pipeline for adaptation: the dataset covers 843,579 hours across 48 languages and offers three transcript styles. Nvidia clarifies these figures describe the dataset release, not the model's full training mixture or sampling weights. A checkpoint and demo are available through Hugging Face.

### Source excerpt

Nvidia Corp. today released a leaderboard-winning open artificial intelligence model for multilingual audio transcription and translation, along with an open data pipeline for fine-tuning Nemotron 3.5 Transcribe. Traditional transcription AI models follow a workflow that detects speech, transcribes it, identifies and labels different speakers, then answers questions about the transcription. Transcribe does all of this [...] The post Nvidia releases all-in-one adaptable Nemotron 3.5 audio transcription and translation model appeared first on SiliconANGLE.

## Developing a Speech Language Model for Basque

DevFeed: [Developing a Speech Language Model for Basque](<https://devfeed.tech/articles/developing-a-speech-language-model-for-basque-103205.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.36911>)

Author: Andoni Sudupe, Inma Hernaez, Ander Arriandiaga, Christoforos Souganidis, Ibon Saratxaga, Eva Navas

Published: 2026-10-28T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>), [LLM evaluation / benchmarking](<https://devfeed.tech/topics/llm-evaluation-benchmarking.md>), [machine-translation](<https://devfeed.tech/topics/machine-translation.md>), [Internationalization (i18n)](<https://devfeed.tech/topics/i18n.md>)

Tags: [architecture](<https://devfeed.tech/tags/architecture.md>), [audio](<https://devfeed.tech/tags/audio.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [data](<https://devfeed.tech/tags/data.md>), [developing](<https://devfeed.tech/tags/developing.md>), [eess-as](<https://devfeed.tech/tags/eess-as.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [interaction](<https://devfeed.tech/tags/interaction.md>), [language](<https://devfeed.tech/tags/language.md>), [language-model](<https://devfeed.tech/tags/language-model.md>), [low-latency](<https://devfeed.tech/tags/low-latency.md>), [model](<https://devfeed.tech/tags/model.md>), [modular](<https://devfeed.tech/tags/modular.md>), [modules](<https://devfeed.tech/tags/modules.md>), [open](<https://devfeed.tech/tags/open.md>), [quality](<https://devfeed.tech/tags/quality.md>), [speech](<https://devfeed.tech/tags/speech.md>), [synthetic](<https://devfeed.tech/tags/synthetic.md>), [synthetic-data](<https://devfeed.tech/tags/synthetic-data.md>), [system](<https://devfeed.tech/tags/system.md>), [tts](<https://devfeed.tech/tags/tts.md>)

### AI overview

The paper presents an open, low-latency speech language model for Basque built on the LLaMA-Omni architecture, using Latxa as its backbone and fine-tuning several modules to transfer Basque language capabilities to spoken interaction. To address limited training data, the authors translated the English VoiceAssistant-400K corpus into Basque and synthesized its audio with a Basque text-to-speech system. Reported evaluation scores are 3.86/5 for content quality, 9.94% word error rate for spoken intelligibility, and 3.07 UTMOS for audio quality. These abstract-level results indicate feasibility; the supplied excerpt does not provide evaluation details or comparisons.

### Source excerpt

arXiv:2610.36911v1 Announce Type: cross Abstract: Although the recent development of LLMs has paved the way for multi-modal interaction, minority languages face difficulties keeping pace. This work presents the first open Speech Language Model for Basque, designed for low-latency, end-to-end spoken interaction. Based on the LLaMA-Omni architecture, the system integrates a Latxa model as its backbone, transferring its Basque capabilities to the spoken modality by fine-tuning several modules. To address data scarcity, the English VoiceAssistant-400K corpus was translated into Basque, and the audio was synthesized using a Basque TTS. Evaluation results confirm the system's viability, achieving 3.86/5 in content quality, a 9.94% WER for spoken intelligibility, and a 3.07 UTMOS for audio quality. This work demonstrates that advanced spoken interaction systems can be developed using modular architectures and synthetic data even in low-resource languages.

## "Not" is Not Lost in Translation: A Causal Analysis of Negation Representations in Multilingual LLMs

DevFeed: ["Not" is Not Lost in Translation: A Causal Analysis of Negation Representations in Multilingual LLMs](<https://devfeed.tech/articles/not-is-not-lost-in-translation-a-causal-analysis-of-negation-representations-in-multilingual-llms-99986.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.32281>)

Author: Giang Son Nguyen, Wray Buntine, Dung D. Le, Wenya Wang

Published: 2026-10-26T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Large Language Model](<https://devfeed.tech/topics/llm.md>), [machine-translation](<https://devfeed.tech/topics/machine-translation.md>), [Representation learning](<https://devfeed.tech/topics/representation-learning.md>), [causality](<https://devfeed.tech/topics/causality.md>), [Internationalization (i18n)](<https://devfeed.tech/topics/i18n.md>)

Tags: [activation](<https://devfeed.tech/tags/activation.md>), [autoencoder](<https://devfeed.tech/tags/autoencoder.md>), [code](<https://devfeed.tech/tags/code.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [data](<https://devfeed.tech/tags/data.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [english](<https://devfeed.tech/tags/english.md>), [feature](<https://devfeed.tech/tags/feature.md>), [features](<https://devfeed.tech/tags/features.md>), [language](<https://devfeed.tech/tags/language.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [languages](<https://devfeed.tech/tags/languages.md>), [large-language-models](<https://devfeed.tech/tags/large-language-models.md>), [llms](<https://devfeed.tech/tags/llms.md>), [model](<https://devfeed.tech/tags/model.md>), [models](<https://devfeed.tech/tags/models.md>), [multilingual-llms](<https://devfeed.tech/tags/multilingual-llms.md>), [open](<https://devfeed.tech/tags/open.md>), [study](<https://devfeed.tech/tags/study.md>), [translation](<https://devfeed.tech/tags/translation.md>)

### AI overview

Across six open multilingual LLMs, negation appears to have a largely shared, causally controllable representation across English, French, Vietnamese, and Chinese. On synthetic contrastive and FLORES-200-derived translation data, activation directions extracted in one language could induce or remove negation in the others, and directions from different languages were substantially aligned. Each model's single SAE feature was reported to control polarity selectively for negated inputs. Dense and sparse interventions also generalized to four additional languages and FLORES-derived sentences.

### Source excerpt

arXiv:2610.32281v1 Announce Type: cross Abstract: Negation occurs in every language, yet its surface realization differs drastically. Despite this, multilingual large language models (LLMs) translate negation robustly, raising the question of whether they share an underlying representation of negation across languages. We study this question in six open models across English, French, Vietnamese, and Chinese, using causal interventions on a synthetic contrastive dataset and a negation translation dataset derived from FLORES-200. Across models, we find strong evidence that negation representations are causally controllable and largely shared across languages. In particular, we show that a contrastive activation direction extracted from one language can induce or remove negation in the others, and directions extracted from different languages are substantially aligned. We further find that negation is represented in sparse features: a single sparse autoencoder (SAE) feature per model suffices to control polarity across languages, while responding selectively to negated inputs. Finally, both dense and sparse interventions generalize to four additional languages and FLORES-derived sentences. We release data and code to support future work: https://anonymous.4open.science/r/negation-translation/.

## Context-Aware Subtitle Translation on CPU: Deployment-Driven Training, Evaluation, and Latency Design

DevFeed: [Context-Aware Subtitle Translation on CPU: Deployment-Driven Training, Evaluation, and Latency Design](<https://devfeed.tech/articles/context-aware-subtitle-translation-on-cpu-deployment-driven-training-evaluation-and-latency-design-97952.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2607.09957>)

Author: Tsz-To Wong, Shyang-En Weng

Published: 2026-10-23T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [machine-translation](<https://devfeed.tech/topics/machine-translation.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>), [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>), [tokenization](<https://devfeed.tech/topics/tokenization.md>), [Post-training optimization](<https://devfeed.tech/topics/post-training-optimization.md>)

Tags: [browser](<https://devfeed.tech/tags/browser.md>), [context](<https://devfeed.tech/tags/context.md>), [cpu](<https://devfeed.tech/tags/cpu.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [decoding](<https://devfeed.tech/tags/decoding.md>), [deployment](<https://devfeed.tech/tags/deployment.md>), [distillation](<https://devfeed.tech/tags/distillation.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [free](<https://devfeed.tech/tags/free.md>), [google-translate](<https://devfeed.tech/tags/google-translate.md>), [latency](<https://devfeed.tech/tags/latency.md>), [llm](<https://devfeed.tech/tags/llm.md>), [open](<https://devfeed.tech/tags/open.md>), [optimization](<https://devfeed.tech/tags/optimization.md>), [pretraining](<https://devfeed.tech/tags/pretraining.md>), [taiwan](<https://devfeed.tech/tags/taiwan.md>), [tokens](<https://devfeed.tech/tags/tokens.md>), [training](<https://devfeed.tech/tags/training.md>), [translation](<https://devfeed.tech/tags/translation.md>)

### AI overview

The paper reports an English-to-Traditional-Chinese (Taiwan) subtitle translator designed to run on a viewer's CPU within about 500 ms per subtitle. Its Sotto models use task-specific vocabulary and several training steps, including continued pretraining, supervised fine-tuning, distillation, and preference optimization. Since subtitle references can be unreliable, the authors compare translations pairwise with Google Translate, using two LLM judges who assess each pair in both orders. On 1,820 held-out subtitles, Sotto-1.7B has a 0.576 win rate against Google Translate, beats four open translation models, and performs slightly below a 27B model. Idle-time prefill reduces mean CPU end-to-end latency by 31%, while the task-specific vocabulary represents the same Chinese text with 28% fewer tokens. The deployed Sotto-0.6B remains within 500 ms for every subtitle in the reported latency test.

### Source excerpt

arXiv:2607.09957v3 Announce Type: replace-cross Abstract: We study English-to-Traditional-Chinese (Taiwan) subtitle translation for a free browser extension, where the model must run on the viewer's CPU in about 500 ms per subtitle. We train two models built on the LMT-60 base models, Sotto-1.7B and the deployed Sotto-0.6B, with a task-specific vocabulary, continued pretraining, supervised fine-tuning, weight interpolation, logit distillation and on-policy preference optimization. Because reference translations of subtitles are unreliable, we evaluate by pairwise comparison against one fixed baseline, Google Translate, with each pair judged in both orders by a panel of two LLM judges. On 1,820 held-out subtitles, Sotto-1.7B has a win rate of 0.576 against the baseline and wins head-to-head against four open translation models, and is slightly below a 27B model. We split latency into prefill and decoding: idle-time prefill, which prefills the known part of the prompt while waiting for the next subtitle, cuts mean end-to-end latency by 31% on a CPU, and the task-specific vocabulary encodes the same Chinese text in 28% fewer tokens. The deployed model stays within 500 ms for every subtitle in our latency test. The code is available at https://github.com/aiden1020/localsubs .

## IBM AI Releases Granite Speech 5.0 2B for Multilingual Transcription and Translation

DevFeed: [IBM AI Releases Granite Speech 5.0 2B for Multilingual Transcription and Translation](<https://devfeed.tech/articles/ibm-ai-releases-granite-speech-5-0-2b-for-multilingual-transcription-and-translation-95134.md>)

Original publisher: [Read original article](<https://www.marktechpost.com/2026/10/22/ibm-granite-speech-5-0-2b-release/>)

Author: Asif Razzaq

Published: 2026-10-22T07:00:56Z

Content type: release

Language: en

Sources: [MarkTechPost](<https://devfeed.tech/sources/marktechpost.md>)

Topics: [asr](<https://devfeed.tech/topics/asr.md>), [ibm](<https://devfeed.tech/topics/ibm.md>), [machine-translation](<https://devfeed.tech/topics/machine-translation.md>), [Speech Processing](<https://devfeed.tech/topics/speech-processing.md>), [llama.cpp](<https://devfeed.tech/topics/llama-cpp.md>), [lora](<https://devfeed.tech/topics/lora.md>), [Transformers](<https://devfeed.tech/topics/transformers.md>)

Tags: [agentic-ai](<https://devfeed.tech/tags/agentic-ai.md>), [ai-shorts](<https://devfeed.tech/tags/ai-shorts.md>), [apache-2-0](<https://devfeed.tech/tags/apache-2-0.md>), [applications](<https://devfeed.tech/tags/applications.md>), [artificial-intelligence](<https://devfeed.tech/tags/artificial-intelligence.md>), [audio-language-model](<https://devfeed.tech/tags/audio-language-model.md>), [editors-pick](<https://devfeed.tech/tags/editors-pick.md>), [enterprise-ai](<https://devfeed.tech/tags/enterprise-ai.md>), [granite](<https://devfeed.tech/tags/granite.md>), [ibm](<https://devfeed.tech/tags/ibm.md>), [language-model](<https://devfeed.tech/tags/language-model.md>), [large-language-model](<https://devfeed.tech/tags/large-language-model.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [model](<https://devfeed.tech/tags/model.md>), [new-releases](<https://devfeed.tech/tags/new-releases.md>), [open](<https://devfeed.tech/tags/open.md>), [open-asr-leaderboard](<https://devfeed.tech/tags/open-asr-leaderboard.md>), [releases](<https://devfeed.tech/tags/releases.md>), [speech](<https://devfeed.tech/tags/speech.md>), [speech-recognition](<https://devfeed.tech/tags/speech-recognition.md>), [staff](<https://devfeed.tech/tags/staff.md>), [tech-news](<https://devfeed.tech/tags/tech-news.md>), [technology](<https://devfeed.tech/tags/technology.md>), [transcription](<https://devfeed.tech/tags/transcription.md>), [translation](<https://devfeed.tech/tags/translation.md>), [voice-ai](<https://devfeed.tech/tags/voice-ai.md>)

### AI overview

IBM released Granite Speech 5.0 2B, an Apache 2.0 speech model for transcription and translation in six languages. The excerpt reports 5.68% average English WER and a #2 position on the Open ASR Leaderboard, behind Qwen3-ASR-1.7B; results vary from 0.99% on LibriSpeech Clean to 13.79% on a private conversational set, and Japanese is its weakest listed language. Language prompting and keyword lists of up to 2,000 terms are supported; on Earnings-22, keyword F1 rises from 67.2 to 90.1 with correct lists. It can run through Transformers or vLLM on GPU, mlx-audio on Apple Silicon, and llama.cpp on CPU or Mac; IBM recommends Q8_0 GGUF because lower-bit quantization loses accuracy on audio over about a minute. The accompanying Turbo CTC update retains its 470M size, adds punctuation and capitalization across seven languages, and reports 19,870 RTFx. The excerpt notes benchmark differences by dataset and that English-to-Mandarin translation was not evaluated.

### Source excerpt

IBM released Granite Speech 5.0 2B, an Apache 2.0 speech recognition and translation model. It ranks #2 on the Open ASR Leaderboard with 5.68% average English WER, alongside updated Turbo CTC and 4.1 2B Plus models. The post IBM AI Releases Granite Speech 5.0 2B for Multilingual Transcription and Translation appeared first on MarkTechPost.

[Next page](<https://devfeed.tech/topics/machine-translation.md?cursor=WyIyMDI2LTEwLTIyVDA3OjAwOjU2KzAwOjAwIiwgImNmMTA5NzU4LTZjYzAtNGZhOC04NDYzLWM2YTkyYzczMWJhYyJd>)