Research agenda
Questions shaping my work in multilingual technology.
My research centers on how multilingual AI systems should be designed, evaluated, and used when language, culture, human judgment, and business requirements interact. The areas below are not a catalog of completed projects; they are the broader questions that connect my research, applied systems, standards work, and publications.
01
Multilingual AI & quality
Evaluation · requirements · translation quality · generative AI
How can/should multilingual AI be evaluated when “quality” depends on purpose, audience, context, and the kind of transformation a system is performing and when risks are unknown or covert?
Multilingual AI is blurring boundaries between once-discrete tasks: translation, transcreation and adaptation, content generation, and quality estimation. A central part of my research asks how evaluation should change when a single system can perform all of these functions or there is no source or reference to compare it to. It also emphasizes the role of human experts in governance and risk management.
Key topics
- Quality relative to specifications.
- Human evaluation, automated evaluation, and the use cases for each.
- How MQM and related approaches need to evolve for generative AI.
- Evaluation of culturally and contextually appropriate output.
- Connections to standards and technologies
02
Multimodal AI in multilingual contexts
Speech · audio · audiovisual context · multilingual evaluation
What changes when multilingual systems must integrate speech, audio, text, and visual context to allow various needed transformations rather than treating language as text alone?
Multilingual and localization quality research has historically treated language primarily as written text. Multimodal AI makes that abstraction increasingly inadequate as content that starts in one form may transition between modalities. Speech recognition, speech generation, dubbing, subtitles, audiovisual context, and cross-modal reasoning all introduce compounding errors and require dimensions of quality that cannot be captured by text alone.
This is an emerging extension of my broader work on multilingual quality and evaluation. The central question is how existing ideas about specifications, errors, and human judgment should change when linguistic output is inseparable from timing, prosody, speaker characteristics, audio context, visual information, and all the embodied context of communication.
Key topics
- Evaluation of multilingual speech and generated audio for different applications.
- Relationships among the many forms that content can take as it crosses modalities, formats, languages, and destinations.
- Cross-modal consistency and culturally appropriate adaptation.
- Where text-based quality models transfer cleanly and where they fail.
- Relevant work in multimodal LLMs, ASR, speech understanding, and generative speech to enable meaningful quality statements tied to specifications and end-user needs.
03
Human + machine
Human judgment · automation · augmented translation · workflow design
What should machines do, where does human judgment add value, and how should that division change as technology improves?
Much discussion of AI and language work assumes that the goal is to replace humans or use their effort to correct machines. My research has instead focused on human-at-the-core approaches and how the process should change when machines gain new capabilities, where human judgment remains important, and how people can work with systems to augment their capabilities rather than being relegated to the drudgery of playing janitor for AI slop.
Key topics
- Augmented Translation and Human at the Core.
- Allocation of tasks according to comparative human and machine strengths.
- How to use AI to automate what should be automated (dangerous, dirty, or dull tasks) while freeing up humans to bring their expertise and skill to ensure that global audiences are served by content.
- How workflows should change rather than simply automate legacy processes and approaches.
- Governance, trust, oversight, and accountability for machines, translators, interpreters, and other parties.
- Research on post-localization and changing professional roles, with a focus on the future of language professionals.
04
Language, culture, economics & markets
Customer behavior · global markets · localization strategy · evidence
How do language and culture affect customer behavior, market opportunity, and the decisions organizations make about global products and content?
Organizations routinely make language and localization decisions with incomplete evidence. This strand of my work examines how linguistic, cultural, demographic, behavioral, economic, and company data can be combined to make those decisions based on firm knowledge.
The research connects directly to applied systems such as the Global Revenue Forecaster, Global CX Calculators, and the global website analysis I have developed. Although these seem like narrow applications, the underlying agenda is broader: how can we know when language support changes customer behavior, how can we measure those effects, and how can we align effort to maximize benefit for all stakeholders
Key topics
- Language preference and customer experience.
- Localization depth and how it affects individuals, e.g., what happens when someone reaches a “wall of English”?
- Market opportunity based on language, geography, vertical, and technology.
- Large-scale comparative analysis of global brands’ multilingual strategies.
- Methods for leveraging sparse or uncertain market evidence to provide reliable guidance.
Related lines