Standards

Shared frameworks for multilingual technology.

I have worked on localization standards and shared technical frameworks throughout my career, from early interoperability standards for translation technology to current work on translation quality. My involvement has included framework development, standards committees, industry representation in formal standardization, and governance of shared technical approaches.

01

Standards & governance

I currently work on maintaining, extending, and formalizing shared approaches to translation quality across research and industry.

Multidimensional Quality Metrics

MQM · MQM Council · ASTM

Multidimensional Quality Metrics (MQM) is a framework for describing and evaluating translation quality. MQM began as a response to limitations in the quality models then available to the translation and localization industry, which were proprietary and non-exchangeable. Today it is widely used in industry - including in many localization tools such as translation management systems (TMSes) and research – by groups such as Google Research and the WMT shared quality tasks – and has become the gold standard for human quality evaluation.

I initiated the creation of MQM and have worked on it for 15 years. I remained involved as it moved from its genesis in the EU-funded QTLaunchpad project while I was at the German Research Center for Artificial Intelligence (DFKI) to its widespread adoption in industry and integration into ISO 5060 (which uses a subset of MQM for its error typology) and imminent formalization as a standard in ASTM.

Today, I serve as president of the MQM Council, the organization responsible for the stewardship and continued development of MQM. The Council brings together organizations and individuals from around the globe with an interest in maintaining MQM as an open, shared framework rather than one controlled by an individual company or technology provider. Increasingly, it functions as a provider of authoritative information on all aspects of translation quality.

This work involved more than specifying an error typology. MQM has to be flexible to accommodate different translation requirements and specifications, changing technologies, and uses ranging from human evaluation to the assessment of machine translation and AI systems. Governance therefore requires maintaining enough consistency for results to be meaningful while allowing the framework to adapt to different applications and to changes in the technologies being evaluated.

02

Foundational standards & interoperability

My earliest standards work focused on a basic problem of localization technology: how to move content, linguistic resources, processing rules, and metrics reliably between systems built by different organizations.

LISA OSCAR

Localization interoperability

Much of my early standards work took place as chair of LISA OSCAR, the standards initiative of the Localization Industry Standards Association (LISA). OSCAR developed a family of standards intended to make localization resources and processes interoperable across tools. Today they serve – along with XLIFF – as the backbone of data exchange within the localization sector. I was involved with several of them, addressing terminology, translation memory, segmentation, and text measurement.

TBX

Terminology exchange · ISO 30042

TermBase eXchange (TBX) provides a standard XML-based representation for exchanging terminology data between systems, with serializations for JSON and SKOS. When LISA transferred TBX to ISO for development as an international standard in the mid-2000s, I represented LISA in the process, working with Dr. Alan Melby, that led to its eventual adoption as ISO 30042. I served on the editing committee that prepared it for ISO adoption and contributed significantly to technical work on its schemas, data structures, and documentation.

I also worked as the LISA representative on the project that resulted in TBX-Basic, the simplified variant used for localization-oriented terminology exchange purposes and as a data model for many termbases. Today, TBX-Basic is maintained by Terminorgs.

TMX

Translation memory interoperability

Translation Memory eXchange (TMX) provides a common format for exchanging translation-memory data between systems. It became one of the most widely implemented localization interchange formats and remains embedded in translation technology workflows. Its limitations – which reflect processing and bandwidth constraints from the time of its creation that no longer apply – are familiar to practitioners, illustrating the compromises that often accompany widely adopted interoperability standards. Nevertheless, it is used today for tasks ranging from the exchange of translation memories to the training of neural machine translation systems.

SRX

Exchanging text segmentation

Segmentation Rules eXchange (SRX) addressed the problem of dividing text consistently into units for translation and other language-processing tasks. It provided a way to exchange segmentation rules so that different systems could reproduce the same behavior instead of making incompatible assumptions about segment boundaries. Default rules sets for many languages simplified the process of ensuring that text was correctly segmented across linguistic boundaries.

GMX-V

Text volume measurement

GMX-V addressed how to measure the volume of text consistently across tools. Word, character, and related counts are used for estimating, pricing, planning, and measuring localization work, yet different systems can produce different results from the same content. GMX-V provided a standardized way to represent those metrics. Although it did not achieve the lasting adoption of TBX or TMX, it nevertheless addressed a real interoperability problem that continues to vex localization processes.

Other Standards

Core Data Interchange Standards

I have also worked on other standards committees to deliver the core architecture needed to support interoperable processes for internationalization and localization.

XLIFF

Localizable content interchange

I also participated in the XLIFF standards committee. Where the OSCAR standards addressed resources and processing information used by localization systems, XLIFF addressed interchange of the content being localized and the information required to process it. I represented LISA on the OASIS XLIFF committee and worked to ensure that it remained broadly compatible with the OSCAR standards.

Together, XLIFF and the OSCAR standards were part of a broader move away from proprietary, tool-specific localization workflows toward an ecosystem in which content and linguistic resources could move between specialized systems.

ITS 2.0

Internationalization metadata for web and XML

I served as an editor of the W3C Internationalization Tag Set (ITS) 2.0 Recommendation, which extended ITS to support richer multilingual processing in XML and HTML-based content and closer integration with language technologies.

My primary contributions focused on the Localization Quality Issue, Localization Quality Rating, MT Confidence, and Allowed Characters data categories, although I was involved more broadly in the specification, the drafting of which was led by the team I worked in at DFKI.

The localization-quality work provides a direct link to the subsequent development of MQM. ITS 2.0 incorporated an early version of the quality framework that was being developed through the European research work that produced MQM. Although that version differed substantially from MQM’s later modular architecture, there is a definite family resemblance and it represents an early point at which the emerging quality model was expressed as interoperable metadata that could travel with multilingual content.

ITS 2.0 therefore sits at the intersection of two strands of my standards work: establishing interoperable technical infrastructure and developing shared approaches to translation-quality evaluation.

03

Frameworks incorporated into standards

Some of my contributions began outside formal standards committees but influenced their development. In these cases, work on practical or conceptual frameworks was subsequently incorporated into standards or provided a basis for later standardization.

Translation specifications

ASTM F2575

I worked closely with the team developing ASTM F2575 – particularly with Dr. Alan Melby – to develop formalized representations of translation specifications: methods for explicitly defining what a translation is intended to accomplish and the requirements against which it should be produced and evaluated. This work provided foundational material incorporated into ASTM F2575, which addresses quality assurance in translation.

My role was primarily in developing the underlying approach with Alan rather than in the formal committee process that produced the standard. The contribution was to the framework that entered the standard rather than authorship of the resulting ASTM document.

LISA QA Model

Quality framework · predecessor to MQM

I worked on development of the LISA QA Model 4.0 from 2008 through 2011. The model was one of the localization industry's established approaches to identifying and classifying translation errors and assessing translation quality. I took it from an open, but monolithic system to a flexible architecture suitable for handling various translation tasks. Unfortunately, the cessation of LISA’s operations cut this project short.

However, the work did not go to waste, and it led directly to the next stage of my research. DFKI recruited me to participate in QTLaunchPad, a European Union-funded research project that required expertise in translation quality evaluation. One of the problems we encountered was that existing models were not interoperable or flexible enough to serve for both human and machine translation. My work on the LISA QA Model delivered the flexibility required to address the range of translation-quality requirements encountered in research and industry.

The resulting work led to the development of Multidimensional Quality Metrics (MQM).

There is therefore a direct line between work I did on an established industry quality model, subsequent European research into the problem, and the development of MQM. MQM in turn moved from a research framework into industry and academic use, community governance through the MQM Council, and formal standards development at ASTM.

Across these activities, the recurring problem I have faced has been how to turn practices developed for particular organizations or purposes into shared frameworks that can work across technologies, organizations, and use cases.