MultilingualWebLT MultilingualWeb-LT: Quality Data Categories Administered by: Funded by:
Target audience MultilingualWebLT • Localization Service Providers doing Quality Assessment tasks • Content creators doing quality veri fi cation • Authors marking up authoring errors • Posteditors providing information on e ffi ciency/MT quality Administered by: Funded by:
Motivations MultilingualWebLT • Streamline linguistic review process – Reduce multiple reviews of similar content – Use provenance of segments to reinforce assessment • Build real-time work fl ow rules based on past translator performance. • Build content pro fi le from translation agent, MT con fi dence, domain, purpose, etc. • Use Linked Data to capture cross-silo process pro fi le of content. Administered by: Funded by:
Description MultilingualWebLT • Two (complex) data categories: – Errors themselves – Overall error pro fi les • Provides fl exible way to use di ff erent error models in a consistent, standardized way. • Implement standard marking regardless of metric used. Administered by: Funded by:
qualityerror purpose MultilingualWebLT • Mark quality issues against a speci fi ed framework • Provide an audit trail for quality resolution • Flexible framework for various tasks Administered by: Funded by:
qualityerror description MultilingualWebLT • type? • ruleset? • severity? • note? • agent? The <span � � its-qa-type="syntactic error” � � its-qa-ruleset="SAE J2450” � � its-qa-severity="major” � � its-qa-note="bad grammar” � � its-qa-agent="ABCReview">verbs agrees</span> with the subject. � Administered by: Funded by:
qualitypro fi le purpose MultilingualWebLT • De fi nes the overall pro fi le(s) used. • Allows for fl exibility in picking an appropriate pro fi le/metric • Does not require local tagging: can be summary only (e.g., the traditional LISA QA Model software or SAE J2450 were used with no in-line tagging) • Defaults to LISA QA Model Administered by: Funded by:
qualitypro fi le description MultilingualWebLT • name? • uri? • pass? • score? • agent? <meta � � name="qualityProfile” � We ¡know ¡this ¡would ¡not ¡ � content="name:LISA QA Model; � work ¡exactly ¡as ¡is, ¡but ¡it ¡ � uri:http://www.example.com; � shows ¡the ¡concept. ¡ � pass:no;score:85%; � � agent:ABCReview” � /> � Administered by: Funded by:
Issues and Notes MultilingualWebLT • Verboseness • Would these be used in general? (More sophisticated than most current processes) • Implementation commitments and timeframe • Represents perspective of upcoming QT Launchpad project (which aims to standardize/unify error metrics). Administered by: Funded by:
Recommend
More recommend