The eighth edition of the Multilingual AI Roundtable, entitled “AI 2.0: From automated translation to end-to-end global content enablement”, was held in Monterey, California. Prior to the in-person event, a survey was sent out to past participants to collect questions and assess expectations.

1. Emerging Landscape of Industry Players

One observation concerns the fact that survey respondents and roundtable participants started to consider names and concepts of industry players in a new light. This tendency indicates rapidly evolving changes. Some respondents are particularly eager to better understand the power dynamics between the players in a “value map” or “stakeholder map”. For example, a respondent asked to create a value map to show ”… which companies and providers work towards a substitution of human translators, and which ones see just another role for the linguists?”

Currently, there are new players joining, established players who are taking on new roles, and players being replaced or removed due to the introduction of AI agents. This roster change is a demonstration of the uncertainty surrounding the entire industry’s transformation. It provides a lens through which to glimpse the beginning of a seismic shift.

2. Will translation remain a valid task for humans?

In analyzing the proposed/discussed names and concepts, there is a tendency that the task of translation is diluted or removed from human workloads altogether. For example, Matt Singer shared that the MT company Lilt used the name Expert Verifier to refer to conventional linguists. This idea was echoed by the proposal of Marina Pantcheva (RWS). Cultural Experts must be included in this list. These new concepts and names were mentioned, with the majority of translation work being increasingly handed over to machines. In light of this fact, humans need to identify new territory where they can add real value.

Interestingly, “de-translation” appeared as an unspoken consensus in the responses and discussions. None of the proposed roles includes the word “translation”. Project Managers are managing beyond translation, thus this word is not in the title. The same is true for the Translation Management System (TMS). For example, in the discussion Kirill Solovjov (ContentQuo) mentioned, current TMS solutions are not yet ready to help companies fully manage all aspects of multilingual AI, as “they focus too much on T in TMS, that is, translation.”

Furthermore, respondents and participants frequently brought in players outside the scope of translation, including non-localization leadership, business leaders, as well as those who deal with governance.

3. Sketching a Stakeholder Map

From the survey results and the roundtable discussions, altogether 20 names and concepts were extracted. The author used two dimensions to interpret these roles: (a) functional categories, and (b) human vs. machine. Below is a description of the five major functional categories.

(1) Cultural-linguistic & Factual Grounding: involves the process of connecting the output of a model to verifiable sources of information[1].

(2) Project & Organizational Governance: covers localization and non-localization leadership, legal matters, as well as broader organizational activities.

(3) Data & Infrastructure: encompasses data resources and providers that support multilingual AI systems, as well as the infrastructure patterns used to host and manage these resources.

(4) AI & Automation: includes both AI system owners, model engineers, and automation providers who design, deploy, and maintain intelligent systems.

(5) Orchestration & Integration: focuses on AI agents and their abilities to collaborate across systems and workflows, supported by interoperability protocols.

Below is a list of these industry roles and their classification: 

Furthermore, a stakeholder map (see below) was created. In this map, human roles are in oval boxes, colored pink, whereas machine roles are in rectangular boxes, colored blue.

Please Click to Enlarge

You may also download a copy at this location.

Visually, one might think human influence is still dominating, based on the higher number of blue/oval boxes (human player) versus pink/rectangular boxes (machine player). However, the goal of automation and AI systems is always to learn from humans and scale up manual efforts. With more tasks being accomplished by humans, sufficient data and knowledge accumulation will facilitate the advancement of machine systems in every aspect of human role. Once a system becomes mature, it can be used to take over partial, if not all, human duties. For example, AI agents fall into two categories, which includes (1) agents as workers, such as translation agents, and (2) agents as orchestrators, such as workflow management agents. As a result, new human tasks/roles will continue to emerge as machine capabilities advance

4. Keeping up to Date: Designing New Role Portfolios for Humans

To discuss more on how to orchestrate agentic AI and drive AI enablement, join Dr. Peng’s next Multilingual AI Master Class starting February 4, 2025.

Acknowledgements

I would like to express my sincere gratitude to the roundtable advisory board members (speakers): Olga Beregovaya (Smartling), Alon Lavie (Phrase/CMU), and Marina Pantcheva (RWS), as well as speakers: Klaus Fleischmann (Kaleidoscope), Sheriff Mohammed Issaka (African Languages Lab), Erik Vogt (Argos Multilingual), and Matt Singer (LILT).

I would like to thank Megan Reid for helping me improve the writing of this article, as well as understanding how a reader without a technical background would feel about the content.

Footnotes:

[1] By default, computer systems have “the symbol grounding problem”. That is, the interpretation of linguistic symbols is parasitic on the meanings in the mind of the interpreter, rather than intrinsic to the computer system that analyzes the symbols. To address this problem, humans must use, evaluate, and correct machine generated results.