Multilingual AI Round Table Terminology Corner
The seventh edition of the Multilingual AI Round Table, entitled “Transition in Localization – From Automation to Agentic AI”, was a truly rewarding experience. Since the first round table in 2022, these events have fostered valuable connections and knowledge sharing among language industry professionals, computer scientists, and decision makers. It is inspiring to see such diverse expertise coming together to navigate and shape the integration of AI in our respective fields.
The term list we are sharing here is a direct result of collaboration from those who attended the round table. The detail of this glossary is a reminder that the advancement of AI is a collective endeavor and relies on the insights and wisdom of many. Building successful AI solutions, especially in the context of this transition, requires and benefits greatly from interdisciplinary collaboration – a principle this round table has consistently championed.
Working Glossary
agentic AI, noun
Definition: Agentic AI is a class of artificial intelligence that focuses on autonomous systems that can make decisions and perform tasks without human intervention. The independent systems automatically respond to conditions to produce process results.
Source: [Wikipedia]
Comment: Agentic AI is AI that has a purpose and that can carry out its purpose and actions without explicit instruction or rules [Arle Lommel, CSA Research, (Multilingual AI Round Table, LocWorld53, 2025)]
Comment: An agentic AI typically contains multiple AI agents in a workflow path [Marina Pantcheva, RWS (Multilingual AI Round Table, LocWorld53, 2025)]
AI agent, noun
Definition: AI agents are intelligent applications designed to automate tasks and enhance human productivity. They can analyze information, make decisions and take actions to achieve specific goals, freeing up time and resources for more strategic work.
Source: [Databricks, A Compact Guide to AI Agent]
Comment: AI agent: A workflow where
– an LLM with reasoning capability which decides which actions to take
– it has access to external systems (tool)
– It has an external (archival) memory and an internal memory, so it can remember past interactions[Source: Marina Sánchez Torrón, Smartling (Multilingual AI Round Table, LocWorld53, 2025)]
Comment: From another perspective, rather than a workflow, an AI agent is a single AI unit in a wider workflow. That is, an agentic system contains AI agents. An AI agent is an augmented AI model that has access to external tools, databases, and memory. [Source: Marina Pantcheva, RWS (Multilingual AI Round Table, LocWorld53, 2025)]
Comment: The above two comments provide two angles to approach AI agents: as a workflow in itself or as a single unit in a wider workflow system. [Source: Peng Wang (Multilingual AI Round Table, LocWorld53, 2025)]
AI assistant, noun
Synonym: virtual assistant,
Source: [Wikipedia]
LLM, noun (acronym)
Acronym for: large language model (fullForm)
Definition: A machine learning model designed for natural language processing tasks, especially language generation [Wikipedia]
Source: [Wikipedia]
RAG, noun (acronym)
Acronym for: retrieval augmented generation (fullForm)
Definition: A technique that enables large language models (LLMs) to retrieve and incorporate additional information than the prompt itself. [revised definition based on Wikipedia]
Source: [Wikipedia]
Comment: The definition provided by Wikipedia can be improved. The information does not have to be necessarily new. It is any external information that is not contained in the prompt itself. [Source: Marina Pantcheva, RWS (Multilingual AI Round Table, LocWorld53, 2025)]
CoT prompting, noun (acronym)
Acronym for: chain-of-thought prompting
Definition: A technique that allows large language models (LLMs) to solve a problem as a series of intermediate steps before giving a final answer
Source: [Wikipedia]
MQM, properNoun (acronym)
Acronym for: Multidimensional Quality Metric – fullForm
Definition: A framework for analytic translation quality evaluation (TQE) for assessing QE accuracy
Source: wmt-qe-task.github.io, https://themqm.org/
ESA, noun (acronym)
Acronym for: error span annotation (fullForm)
ICL, noun (acronym)
Acronym for: in-context learning (fullForm)
TQE, noun (acronym)
Acronym for: translation quality estimation (fullForm)
TQT, noun (acronym)
Acronym for: translation quality technology (fullForm)
Comment: I’m unaware if this term was used prior to us using it at Unbabel in 2020 and beyond. Source can be further investigated. [Source: Alon Lavie, Phrase (Multilingual AI Round Table, LocWorld53, 2025)]
QPS, noun (acronym)
Acronym for: quality performance score (fullForm)
Comment: Term coined by Phrase for marketing purposes of its proprietary QE technology. [Source: Alon Lavie, Phrase (Multilingual AI Round Table, LocWorld53, 2025)]
QE, noun (acronym)
Acronym for: quality estimation (fullForm)
NLP, noun (acronym)
Acronym for: natural language processing (fullForm)
NMT, noun (acronym)
Acronym for: neural machine translation (fullForm)
Definition: machine translation that uses an artificial neural network to predict the likelihood of the next word in a sequence of words. [revised definition based on Wikipedia]
Source: [Wikipedia]
Comment: The more precise formulation is “the likelihood of the next word in a sequence”. [Source: Marina Pantcheva, RWS (Multilingual AI Round Table, LocWorld53, 2025)]
MTQE, noun (acronym)
Acronym for: machine translation quality estimation (fullForm)
WMT, properNoun (acronym)
Acronym for: Workshop on Machine Translation
Definition: A primary conference event for machine translation and machine translation research. (revised definition based on https://machinetranslate.org/wmt)
Comment: The conference is held annually in connection with larger conferences on natural language processing.
Source: https://machinetranslate.org/wmt
Comment: Change the “The main event” to “A primary conference” in the definition. [Source: Alon Lavie, RWS (Multilingual AI Round Table, LocWorld53, 2025)]
LTP, noun (acronym)
Acronym for: language technology platform (fullForm)
BPMN, properNoun (acronym)
Acronym for: Business Process Model Notation
ZSL, noun (acronym)
Acronym for: zero-shot learning (fullForm)
Definition: a machine learning technique where a model can classify data it has never seen before by using semantic information about the categories. [Wikipedia]
Source: [Wikipedia]
non-AI automation (noun)
Definition: Automation based on rule-based deterministic systems using scripts
Source: Marina Pantcheva, RWS (Multilingual AI Round Table, LocWorld53, 2025)
single LLM pass (noun)
Definition: A single prompt which is passed to a language model to generate output in one go without iteration.
Source: Marina Pantcheva, RWS (Multilingual AI Round Table, LocWorld53, 2025)
augmented LLM (noun)
Definition: A single LLM enhanced with retrieval-augmented generation (RAG), APIs, memory storage, external plugins.
Source: Marina Pantcheva, RWS (Multilingual AI Round Table, LocWorld53, 2025)
agentic workflow (noun)
Definition: multiple LLMs and tools working together in a pre-orchestrated logic path.
Source: Marina Pantcheva, RWS (Multilingual AI Round Table, LocWorld53, 2025)
parallelization(noun)
Definition: A process where input is provided to a number of LLMs and the intermediate results from these are then aggregated by an LLM Aggregator which provides a final output
Example: LQA score calculator, multimodal content production
Note: A human may be needed to evaluate (samples in) the individual checkpoints
Source: Marina Pantcheva, RWS (Multilingual AI Round Table, LocWorld53, 2025)
routing(noun)
Definition: A process where input is provided to an LLM Orchestrator which analyses and chooses which LLM should process the input and provide the output
Example: language detectors, sentiment detector, age-based content customization
Note: A human may be needed to evaluate (samples in) the individual checkpoints
Source: Marina Pantcheva, RWS (Multilingual AI Round Table, LocWorld53, 2025)
agentic system (noun)
Definition: A system where LLMs dynamically plan, choose tools, and adjust strategies to meet a goal. Autonomous, goal-directed behavior
Comment: A process where an LMM receives input and decides which LLMs to apply and then the result is then provided by a LLM synthesizer
Comment: Hard to predict & control, guardrails needed
Example: autonomous research agents
Source: Marina Pantcheva, RWS (Multilingual AI Round Table, LocWorld53, 2025)
We invite you to download a copy of the Multilingual AI Working Glossary below:
Join us in Monterey for the next Multilingual AI Round Table to connect with leading minds in localization and AI. The term list we are sharing highlights how interdisciplinary collaboration can help you build successful AI solutions. Don’t miss your chance to help shape the future of intelligent language solutions. Seats are limited to 25 and fill up quickly for this event.
Disclaimer: Copyright © 2021-2025 The Localization Institute. All rights reserved. This document and translations of it may be copied and furnished to others, and derivative works that comment on or otherwise explain it or assist in its implementation may be prepared, copied, published, and distributed, in whole or in part, without restriction of any kind, provided that the above copyright notice and this section are included on all such copies and derivative works. However, this document itself may not be modified in any way, including by removing the copyright notice or references to The Localization Institute, without the permission of the copyright owners. This document and the information contained herein is provided on an “AS IS” basis and THE LOCALIZATION INSTITUTE DISCLAIMS ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION HEREIN WILL NOT INFRINGE ANY OWNERSHIP RIGHTS OR ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
