Translation quality, examined.
Measure translation quality.
Find the best LLM for translation in your languages using the latest benchmarks. Audit translations against your own quality scorecard.
Backed by Y Combinator
Trusted by international teams managing content at scale.
%20--%3e%3csvg%20version='1.1'%20id='Layer_1'%20xmlns='http://www.w3.org/2000/svg'%20xmlns:xlink='http://www.w3.org/1999/xlink'%20x='0px'%20y='0px'%20viewBox='0%200%20233.1%2072.2'%20style='enable-background:new%200%200%20233.1%2072.2;'%20xml:space='preserve'%3e%3cstyle%20type='text/css'%3e%20.st0{fill:%232a3631;}%20%3c/style%3e%3cg%3e%3cpath%20class='st0'%20d='M49.6,32.8c0.4-0.7,0.7-1.3,1.1-2c5.4-10.2,4.4-14.2,3.4-15.7c-1.3-1.8-4.3-2.7-9.2-2.7%20c-2.9,0-6.2,0.3-9.6,0.9c-0.6-3.1-1.5-5.9-2.4-8.1c-1.2-2.8-2.6-4.5-4.1-5.1C28.3,0.1,27.8,0,27.3,0C24,0,18.5,3.9,12.3,10.6%20C5.5,17.9,1.4,25,0.3,31.1c-1.1,5.9,0.5,13.8,8,18.5c0.7,1.6,1.7,2.9,2.9,4.1C14.7,57.2,20,59,26.8,59c6.4,0,14.5-1.8,21.6-4.8%20c3.4-1.4,6.3-3,8.3-4.6c2.4-1.8,3.7-3.7,3.8-5.4c0-1.6-0.9-3.3-2.8-5.3C55.8,37.1,53,34.9,49.6,32.8z%20M44.9,16.8%20c4,0,5.3,0.7,5.6,0.8c0.2,0.6,0.1,4.1-3.7,11.2c-0.3,0.6-0.6,1.2-1,1.8c-3-1.6-6.1-3-9.1-4.1c0-2.9-0.3-5.9-0.7-8.8%20C39.2,17.1,42.2,16.8,44.9,16.8z%20M36,36.5c0.3-1.7,0.5-3.5,0.6-5.4c2.3,0.9,4.7,2,6.9,3.2c-2.3,3.5-5,6.7-7.6,9.2%20c-1.1,1-2.2,2-3.3,2.8C34.1,43.7,35.3,40.5,36,36.5z%20M32.2,25.1c-3.5-1-6.8-1.5-9.7-1.5c-1.6,0-3.1,0.2-4.6,0.5%20c2.7-1.7,5.7-2.9,7.8-3.7c1.9-0.7,3.8-1.3,5.9-1.8C32,20.8,32.2,22.9,32.2,25.1z%20M7.8,28.5c-0.8,1.6-1.7,3.3-2,5l0,0.2%20c-0.2,0.9-0.6,3-0.1,5.8c-1-2.6-1.6-4.8-1-7.7c0.9-4.8,4.5-11,11.2-18.5c2.3-2.6,4.8-4.9,7-6.6c1.9-1.4,3.6-2.3,4.5-2.4%20c0.8,0.4,2.4,3.8,3.7,9.9c-5,1.2-9.7,3-13.5,5.1C12.9,21.9,9.7,24.9,7.8,28.5z%20M12.2,45.6c-1.4-2.1-2.1-4.7-2-7.8v0%20c0-1.3,0.7-4.9,2.9-7.1c1.8-1.8,5.1-2.8,9.4-2.8c2.8,0,6.1,0.6,9.7,1.6c-0.1,2.2-0.3,4.3-0.6,6.2c-1.6,9-5.3,13.6-10.9,13.6%20C18.4,49.4,14.1,48.4,12.2,45.6z%20M47,50c-6.7,2.9-14,4.6-20.1,4.6c-2.3,0-4.4-0.2-6.2-0.7c1.7,0.2,2.8,0.2,3,0.2%20c3.4,0,7.4-1.5,11.3-4.1c4.5-3.1,8.7-7.6,12.5-13.5c3,1.9,5.5,3.8,7.1,5.5c1.2,1.2,1.6,2,1.6,2.1C56,44.6,53.7,47.1,47,50z'/%3e%3cg%3e%3cpath%20class='st0'%20d='M110.8,2.3c-15.1,0-24.8,0-39.7,0c0,2,0,3.8,0,5.9c6.4,0,10,0,16.7,0c0,16.5,0,31.6,0,47.7c2.2,0,4.2,0,6.4,0%20c0-16.3,0-31.3,0-47.8c6.6,0,10.2,0,16.6,0C110.8,6,110.8,4.2,110.8,2.3z'/%3e%3cpath%20class='st0'%20d='M146.2,17h-6.3c0,4.1,0,7.8,0,11.5c0,3.7,0.1,7.4-0.1,11.2c-0.4,7.1-4.6,11.1-11.5,11.1%20c-6.9-0.1-11.1-4.1-11.5-11.1c-0.2-3.7-0.1-7.4-0.1-11.2c0-3.7,0-7.4,0-11.5h-6.3c0,1.5,0,2.8,0,4.2c0,5.9-0.1,11.8,0.1,17.6%20c0.1,2.6,0.4,5.3,1.3,7.7c2.4,6.3,8.5,10,16.1,10.1v0c0.1,0,0.2,0,0.3,0c0.1,0,0.2,0,0.3,0v0c7.7-0.2,13.8-3.8,16.1-10.1%20c0.9-2.4,1.2-5.1,1.3-7.7c0.2-5.9,0.1-11.7,0.1-17.6C146.2,19.8,146.2,18.5,146.2,17z'/%3e%3cpath%20class='st0'%20d='M229,24.3c-2.9-3.8-6.8-6.3-11.5-7.1c-6.9-1.3-13.2,0.1-18.5,5.8c-0.1-2.2-0.2-3.7-0.2-5.2c-2.2,0-4.1,0-6,0%20c0,18.3,0,36.3,0,54.4c2.1,0,4,0,6.1,0c0-7.1,0-13.9,0-21.2c10.2,7.6,20.7,7.7,28.2,0.5C234.3,44.5,235.1,32.4,229,24.3z%20M212.5,51.1c-8.1,0-14.1-6.1-14.1-14.4c0-8.1,5.9-14.1,14-14.2c8.2-0.1,14.2,6,14.2,14.2C226.7,45,220.7,51.1,212.5,51.1z'/%3e%3cpath%20class='st0'%20d='M156.9,55.9c2.1,0,3.9,0,5.9,0c0-19.7,0-34.3,0-53.6c-2.1,0-3.9,0-5.9,0C156.9,21.9,156.9,36.5,156.9,55.9z'%20/%3e%3cpath%20class='st0'%20d='M174.9,56c2.1,0,3.9,0,5.9,0c0-13,0-25.6,0-38.2c-2.1,0-4,0-5.9,0C174.9,30.6,174.9,43.2,174.9,56z'/%3e%3cpath%20class='st0'%20d='M177.7,2.3c-2.1,0.1-3.8,1.8-3.8,3.8c0,2,1.8,3.8,3.8,3.8c2.2,0,4-1.7,4-3.9C181.8,3.9,180,2.3,177.7,2.3z'/%3e%3c/g%3e%3c/g%3e%3c/svg%3e)
01 / LOOK CLOSELY
Measure translation quality
Audit human and LLM translations against your quality scorecard. Track quality by language and translation provider.
01
Find changes in meaning.
Check for omitted information, added claims, and wording that changes the source meaning.
- SOURCEThe original passage
- TRANSLATIONThe same passage,translated
- CHANGED MEANINGWording shifts the claim
- OMITTED INFORMATIONThe translation drops a detail
- ADDED CLAIMThe source never says this
- SOURCEThe original passage
- TRANSLATIONThe same passage,translated
- CHANGED MEANINGWording shifts the claim
- OMITTED INFORMATIONThe translation drops a detail
- ADDED CLAIMThe source never says this
02
Install Lingo.dev in the tools you already use.
Connect your TMS, CMS, GitHub, and more. Continuously audit translations against your quality scorecard.
TMSCMSGitHub
- TMS
- CMS
- GitHub
- YOUR QUALITY SCORECARDAudits every connected tool
- YOUR EXISTING TOOLSTMS, CMS, GitHub, and more
- YOUR QUALITY SCORECARDAudits every connected tool
- YOUR EXISTING TOOLSTMS, CMS, GitHub, and more
03
Inspect the passages behind quality flags.
See the source, translation, quality criterion, and reason for each finding.
- REASONWhy the audit flagged it
- QUALITY CRITERIONThe rule from your scorecard
- TRANSLATIONThe flagged wording
- SOURCEThe original wording
- ONE QUALITY FLAG
- REASONWhy the audit flagged it
- QUALITY CRITERIONThe rule from your scorecard
- TRANSLATIONThe flagged wording
- SOURCEThe original wording
- ONE QUALITY FLAG
04
Compare audit findings with your team’s reviews.
Examine findings your reviewers disagree with and issues they reported that the audit did not flag.
- AUDIT FINDINGSAgainst your scorecard
- YOUR TEAM'S REVIEWThe same passage
- YOUR REVIEWERS DISAGREE
- Flagged by the audit only
- NOT FLAGGED BY THE AUDITYour reviewers report it
- AUDIT FINDINGSAgainst your scorecard
- YOUR TEAM'S REVIEWThe same passage
- YOUR REVIEWERS DISAGREEFlagged by the audit only
- NOT FLAGGED BY THE AUDITYour reviewers report it
05
Review the most serious findings first.
Separate critical errors from minor issues using your scorecard's severity rules.
- MINOR · REVIEW NEXT
- These findings stay visible
- CRITICAL · REVIEW FIRSTYour rules place these first
- YOUR SEVERITY RULESCritical before minor
- MINOR · REVIEW NEXTThese findings stay visible
- CRITICAL · REVIEW FIRSTYour rules place these first
- YOUR SEVERITY RULESCritical before minor
06
Check your product terminology.
Find glossary terms and product names that differ from your approved wording.
- YOUR GLOSSARYYour team approves each term
- APPROVED WORDINGThis use matches the glossary
- A DIFFERENT TERMThe translation uses another word
- YOUR GLOSSARYYour team approves each term
- APPROVED WORDINGThis use matches the glossary
- A DIFFERENT TERMThe translation uses another word
07
Check translations against your brand voice.
Identify wording that conflicts with your tone and writing rules.
- YOUR TONE AND WRITING RULESYour team writes them once
- CONFLICTING WORDINGThis phrase breaks one rule
- YOUR TONE AND WRITING RULESYour team writes them once
- CONFLICTING WORDINGThis phrase breaks one rule
08
Check consistency across related content.
Find conflicting terminology and names within a document or collection of content.
- ONE COLLECTIONRelated documents
- THE SAME NAMEEach document should use it
- A CONFLICTING NAMEOne document uses another
- ONE COLLECTIONRelated documents
- THE SAME NAMEEach document should use it
- A CONFLICTING NAMEOne document uses another
09
Apply different scorecards to different content.
Assess product interfaces, marketing, and support against the requirements for each content type.
- PRODUCT INTERFACES
- MARKETING
- SUPPORT
- A SCORECARD PER CONTENT TYPEEach holds its own requirements
- PRODUCT INTERFACES
- MARKETING
- SUPPORT
- A SCORECARD PER CONTENT TYPEEach holds its own requirements
10
Recheck translations after edits.
Compare findings for revised text with the previous assessment.
- YOUR TEAM'S EDITThe revised text gets a new assessment
- PREVIOUS ASSESSMENTIt stays on record
- NEW ASSESSMENTCompare both assessments
- YOUR TEAM'S EDITThe revised text gets a new assessment
- PREVIOUS ASSESSMENTIt stays on record
- NEW ASSESSMENTCompare both assessments
EXISTING CONTENT
Lingo.dev reads translations where your teams keep them.
01 / Code repositories
The strings your developers ship.
Examine translated strings in each commit and pull request before a release reaches your users.
- LOCALE FILESSource and translations
- COMMITSEach change, in order
- PULL REQUESTSChecked before merge
- LOCALE FILESSource and translations
- COMMITSEach change, in order
- PULL REQUESTSChecked before merge
02 / Content management systems
The pages your teams maintain.
Keep source content and translations connected to the requirements for that content.
- PRODUCT PAGESLinked field by field
- SUPPORT ARTICLESA second entry type
- SHARED CONTENTReused by both
- PRODUCT PAGESLinked field by field
- SUPPORT ARTICLESA second entry type
- SHARED CONTENTReused by both
03 / Email platforms
The messages your customers read.
Examine translated campaigns, transactional messages and lifecycle sequences against the same quality criteria as the rest of your content.
- CAMPAIGNSVariants per language
- TRANSACTIONAL EMAILReceipts and alerts
- LIFECYCLE SEQUENCESMessages sent in order
- CAMPAIGNSVariants per language
- TRANSACTIONAL EMAILReceipts and alerts
- LIFECYCLE SEQUENCESMessages sent in order
04 / Translation management systems
The workflow your team owns.
Keep translation production, review, and approval in the systems your team already uses.
- TRANSLATIONSFrom linguists
- REVIEW RECORDSFrom reviewers
- YOUR REQUIREMENTSOne check fails
- TRANSLATIONSFrom linguists
- REVIEW RECORDSFrom reviewers
- YOUR REQUIREMENTSOne check fails
05 / Databases
The records your product stores.
Examine translated records next to their source text, in the tables where your product reads them.
- TRANSLATED FIELDSOne column per language
- PRODUCT CATALOGSChecked where they live
- SOURCE RECORDSOriginal text per row
- TRANSLATED FIELDSOne column per language
- PRODUCT CATALOGSChecked where they live
- SOURCE RECORDSOriginal text per row
06 / API
Send translations from your own tools.
Submit source text and translations from your application for analysis.
- SOURCE TEXTOriginal passages
- TRANSLATIONSTarget-language passages
- ANALYSISInspect both together
- FINDINGSPassages to review
- SOURCE TEXTOriginal passages
- TRANSLATIONSTarget-language passages
- ANALYSISInspect both together
- FINDINGSPassages to review
Lingo.dev in numbers
300M+
Words processed
- Languages
- 100+
- Enterprise uptime SLA
- 99.99%
- Model authors
- 80
02 / CHOOSE DELIBERATELY
Decide what to improve next
Compare quality, reviewer edits, and costs. Decide which providers to use and which problems to address first.
01
Choose the next improvement to test.
Use recurring findings to decide whether to investigate terminology, missing context, or translation instructions.
- TERMINOLOGYGlossary and product names
- MISSING CONTEXTWhat translators did not know
- TRANSLATION INSTRUCTIONSGuidance sent with the work
- YOUR NEXT TESTYour team picks one cause to test
- RECURRING FINDINGSFindings that repeat, grouped by possible cause
- TERMINOLOGYGlossary and product names
- MISSING CONTEXTWhat translators did not know
- TRANSLATION INSTRUCTIONSGuidance sent with the work
- YOUR NEXT TESTYour team picks one cause to test
- RECURRING FINDINGSFindings that repeat, grouped by possible cause
02
Decide where LLM translation fits.
Compare human and LLM results against your requirements for each language and content type.
- LLM
- Human
- LLM
- LLM
- Human
- Human
- LLM
- LLM
- LLM
- DE
- JA
- PT
- INTERFACE
- MARKETING
- SUPPORT
- YOUR REQUIREMENTSPer language and content type
- LLM TRANSLATION FITSLLM results meet the bar
- HUMAN TRANSLATION
- Your team keeps it here
| INTERFACE | MARKETING | SUPPORT | |
|---|---|---|---|
| DE | LLM | Human | LLM |
| JA | LLM | Human | Human |
| PT | LLM | LLM | LLM |
- YOUR REQUIREMENTSPer language and content type
- LLM TRANSLATION FITSLLM results meet the bar
- HUMAN TRANSLATIONYour team keeps it here
03
Compare translation and review costs.
Weigh quality against recorded translation, review, and correction costs.
- QUALITYYour audit results
- RECORDED COSTS
- Translation, review, correction
- YOUR TEAM WEIGHS BOTH
- QUALITYYour audit results
- RECORDED COSTSTranslation, review, correction
- YOUR TEAM WEIGHS BOTH
04
Evaluate a translation provider before switching.
Compare candidate and current providers on matched content under your scorecard.
- MATCHED CONTENT
- YOUR SCORECARDThe same criteria on both sides
- CURRENT PROVIDER
- CANDIDATE PROVIDERYour team compares both first
- MATCHED CONTENT
- YOUR SCORECARDThe same criteria on both sides
- CURRENT PROVIDER
- CANDIDATE PROVIDERYour team compares both first
05
Make the case for a change.
Show your team how a trial compares with your current process on quality, reviewer edits, and cost.
- QUALITY
- REVIEWER
- EDITS
- COST
- THE TRIALUses the same content
- YOUR CURRENT PROCESS
- QUALITY
- REVIEWEREDITS
- COST
- THE TRIALUses the same content
- YOUR CURRENT PROCESS
06
See what reviewers change.
Track how often translations are edited and how much text changes, by language and provider.
DEJAPT
- DE
- JA
- PT
- HOW MUCH TEXT CHANGESHigher stacks mean more edits
- YOUR REVIEWERS' EDITSThe report counts edits by language and provider
- HOW MUCH TEXT CHANGESHigher stacks mean more edits
- YOUR REVIEWERS' EDITSThe report counts edits by language and provider
07
Check whether targeted problems recur.
Follow the errors your team addressed through changes to glossaries, instructions, or review processes.
- YOUR TEAM'S CHANGEA glossary, instruction, or review update
- ONE TARGETED ERROR
- Findings before the change
- THE SAME ERROR AFTER ITYour team checks whether it recurs
- YOUR TEAM'S CHANGEA glossary, instruction, or review update
- ONE TARGETED ERRORFindings before the change
- THE SAME ERROR AFTER ITYour team checks whether it recurs
08
Give providers specific feedback.
Share recurring findings and affected passages when asking a provider to address a quality problem.
- RECURRING FINDINGSThe problem your team wants fixed
- AFFECTED PASSAGES
- Every example, with its source
- YOUR PROVIDERReceives specific evidence
- RECURRING FINDINGSThe problem your team wants fixed
- AFFECTED PASSAGESEvery example, with its source
- YOUR PROVIDERReceives specific evidence
09
Compare language spending with content use.
Use translation spending and readership by language to explain the budget you recommend.
DEJAPTKO
- DE
- JA
- PT
- KO
- TRANSLATION SPENDINGWhat each language costs
- READERSHIPWho reads each language
- TRANSLATION SPENDINGWhat each language costs
- READERSHIPWho reads each language
10
See when your scorecard changed.
Check the criteria and weights behind each report before interpreting changes in scores.
- YOUR SCORECARD CHANGED HERERead scores on each side separately
- CRITERIA AND WEIGHTSThe version each report used
- REPORTS IN TIME ORDER
- YOUR SCORECARD CHANGED HERERead scores on each side separately
- CRITERIA AND WEIGHTSThe version each report used
- REPORTS IN TIME ORDER
From customer stories
Customers describe working with Lingo.dev.
“The knowledge stays with us – what quality looks like, what needs translating, for what audience.”
Annika Palm
Localization Engineer, SoSafe
“When we needed adjustments, they delivered fixes faster than I’ve seen from any vendor.”
Keith Williams
Head of Engineering, Cal.com
“It was clear they had thought through all the complexities of localization, especially for MDX files which were a particular pain point for us.”
Iuliia Shnai
Founder, Papermark
“We’re in a shared Slack channel with the engineers who actually build the platform.”
Nick Bazley
Staff Product Manager, Laurel
“Their co-founder didn’t just understand the problem – she shipped a solution in a couple of days.”
Sebastiaan van Leeuwen
Product Manager, Truely
“And because of how their team is set up, we talk directly to the engineers building the infrastructure. Edge cases get resolved the same day.”
Max Höffner
Director of Product Engineering, SoSafe
“Lingo.dev gave us a localization engine we run from an API, an MCP, and a CLI, right inside our own stack.”
Clément Champau
CEO and Co-Founder, Scribe
“I would recommend revamping the process so the humans employed can actually focus on what matters.”
Sheree Foltin
Content Engineering Teamlead, SoSafe
“We used to constantly worry about keeping translations in sync. Now we just build features, and localization happens automatically.”
Quentin Decré
Co-founder, Jarvi
03 / EXPLORE THE RESEARCH
Compare LLMs for your languages
Access Lingo.dev's translation benchmarks. Choose the models worth testing on your content.
01
Compare model rankings by language pair.
Find the strongest benchmark results for the source and target languages you translate.
EN → JAEN → DEFR → ENEN → PT
- 1 · MODEL CStrongest result for this pair
- 2 · MODEL A
- 3 · MODEL D
- EN → JA
- EN → DE
- FR → EN
- EN → PT
- YOUR LANGUAGE PAIRThe study ranks models for this pair
- STUDY CRITERIASet by the published study
- PUBLISHED STUDYOne ranking per language pair.Each ranking applies only to its pair.
- 1 · MODEL CStrongest result for this pair
- 2 · MODEL A
- 3 · MODEL D
- YOUR LANGUAGE PAIRThe study ranks models for this pair
- STUDY CRITERIASet by the published study
- PUBLISHED STUDYOne ranking per language pair.Each ranking applies only to its pair.
02
Choose candidates for your content type.
Compare model performance on product interfaces, marketing, support, and other studied content.
1 · A2 · C3 · B1 · C2 · B3 · A1 · B2 · A3 · C
- 1 · A
- 2 · C
- 3 · B
- PRODUCT INTERFACES
- 1 · C
- 2 · B
- 3 · A
- MARKETING
- 1 · B
- 2 · A
- 3 · C
- SUPPORT
- ONE RANKING PER TYPEThe model order changes
- PRODUCT INTERFACES
- MARKETING
- SUPPORT
- ONE RANKING PER TYPEThe model order changes
03
Compare quality with model pricing.
See how benchmark results differ between lower-cost and higher-cost models.
ABCD
- A
- B
- C
- D
- LOWER-COST MODELS
- HIGHER-COST MODELSEach step costs more
- BENCHMARK RESULTEach column, one model
- LOWER-COST MODELS
- HIGHER-COST MODELSEach step costs more
- BENCHMARK RESULTEach column, one model
04
Check what a model update changes.
Compare translation quality before and after a model-version change.
- AFTER THE UPDATEThe new model version
- BEFORE THE UPDATEThe previous version
- THE SAME TEST SETThe study compares passages
- ONE PASSAGE GOT WORSEAnother improved
- AFTER THE UPDATEThe new model version
- BEFORE THE UPDATEThe previous version
- THE SAME TEST SETThe study compares passages
- ONE PASSAGE GOT WORSEAnother improved
05
Learn where each model makes mistakes.
Examine recurring meaning, terminology, and tone errors found in benchmark output.
- MEANING ERRORSThe sense changes
- TERMINOLOGY ERRORSA product term is wrong
- TONE ERRORSThe tone misses the brand
- BENCHMARK
- OUTPUTSFrom one model
- THE MODEL YOU CHOSE
- The benchmark shows
- recurring errors
- MEANING ERRORSThe sense changes
- TERMINOLOGY ERRORSA product term is wrong
- TONE ERRORSThe tone misses the brand
- BENCHMARKOUTPUTSFrom one model
- THE MODEL YOU CHOSEThe benchmark showsrecurring errors
06
Read the full benchmark reports.
Explore each study's findings, quality criteria, and examples behind the rankings.
123
- 1
- 2
- 3
- FINDINGS
- What the study found
- QUALITY CRITERIA
- How the study judges quality
- EXAMPLES
- Passages behind each result
- RANKING
- The order the study reports
- FINDINGSWhat the study found
- QUALITY CRITERIAHow the study judges quality
- EXAMPLESPassages behind each result
- RANKINGThe order the study reports
07
Compare human and LLM benchmark results.
Examine translations of matched material assessed against shared criteria.
- HUMAN TRANSLATION
- LLM TRANSLATION
- SHARED CRITERIAOne rule for both
- MATCHED MATERIALThe same source for both
- HUMAN TRANSLATION
- LLM TRANSLATION
- SHARED CRITERIAOne rule for both
- MATCHED MATERIALThe same source for both
08
See how product context affects translation.
Read studies comparing model output with and without relevant context.
- PRODUCT CONTEXTTerms, screens, and audience
- WITHOUT CONTEXT
- The model guesses a term
- WITH CONTEXTThe study compares both outputs
- EMPTY SLOTNo context given
- PRODUCT CONTEXTTerms, screens, and audience
- WITHOUT CONTEXTThe model guesses a term
- WITH CONTEXTThe study compares both outputs
- EMPTY SLOTNo context given
09
Compare reviewer edits across models.
Read studies of how often reviewers change translations and how much wording changes.
- MODEL A
- MODEL B
- MODEL C
- EDIT COUNTOne block per reviewer edit
- REVIEWERS' EDITSHow much wording changes
- MODEL A
- MODEL B
- MODEL C
- EDIT COUNTOne block per reviewer edit
- REVIEWERS' EDITSHow much wording changes
10
Check how consistent a model is.
Examine variation across repeated translations and different passages.
- A RUN THAT DIFFERSTwo of five runs change one clause
- A DIFFERENT PASSAGEFive runs, the same wording
- FIVE RUNS,
- ONE MODEL
- A RUN THAT DIFFERSTwo of five runs change one clause
- A DIFFERENT PASSAGEFive runs, the same wording
- FIVE RUNS,ONE MODEL
Ongoing research
We study how models translate.
Make your localization data-driven.
Start with translations your team has already reviewed.