According to the Slator 2023 Language Industry Report published in May 2023, language service providers (LSPs) consider machine translation (MT), quality evaluation, and business workflows as the most promising applications for large language models (LLMs). The same report found that LSPs are currently in the Awareness and Discovery phase of putting LLMs into production.
LLMs enable a range of functionalities relevant to the language industry, such as text adaptation via prompting, dynamic application of glossaries and style guides during translation, cleaning and maintenance of translation memories, and quality estimation.
Furthermore, LLMs may be useful for error detection, generation of variations of target text for expert linguists to choose from, automating manual tasks (e.g., tag placement), automated post-editing, and synthesizing parallel texts to train other MT models.
As LLMs reshape the language industry, one role that is bound to undergo significant evolution is that of the expert-in-the-loop post-editor.
With LLMs precisely identifying segments that require human review, post-editors can limit their efforts to reviewing specific problematic segments. This approach reduces the time spent on checking already-adequate translations, improving the efficiency of the post-editing process.
Additionally, LLMs empower post-editors by providing quality assessments and error reports, enabling them to concentrate on aspects of translation that require the most attention based on the LLM’s analysis.
Automated Only
LLMs open doors to “automated only” workflows, enabling the delivery of higher quality translation through a combination of multiple automation steps, whereby machine translation is used to generate the target text, LLMs perform quality assessment to identify errors, and LLM-enabled post-editors close the loop and correct errors.
The advances in artificial intelligence (AI) and LLMs are not only transforming existing roles but also giving rise to new roles within LSPs that require a combination of linguistic and AI-interaction expertise.
In April 2023, Slator conducted a survey among C-level LSP leaders to find out what type of roles they expect to see emerging soon.
Dave Bryant, CEO of Dotsub, emphasized the importance of roles that understand how to effectively interact with both human translators and AI-based systems. “Just as there are roles that know how to interact with human translators, there will be roles that aim to maximize the potential of AI-based systems,” Bryant said.
François Chartrand, President & General Manager of Versacom, highlighted the need for “roles that focus on designing AI-based applications and integrations.”
Slator 2023 Language Industry Market Report
Respondents to the survey suggested several potential roles, with the most frequent suggestion being “Prompt Engineer.”
Smith Yewell, Co-Founder and former CEO of Welocalize emphasized that AI will augment various digital jobs, leading to a growing demand for AI prompting and editing expertise. “Now, just about any type of digital work can be augmented with AI, so all digital jobs will change and evolve with AI prompting and editing expertise growing in demand,” he said.
These roles will play a crucial part in leveraging LLMs and maximizing their benefits in the language industry.
Jay Marciano, Director of MT Outreach and Strategy at AI-agency Lengoo and president of the Association for Machine Translation in the Americas (AMTA), echoed this sentiment during a virtual conference held on May 20, 2023. According to Marciano, linguists already have many applicable skills for jobs such as Prompt Engineer, AI Ethicist, Data Curator, or Data Scientist.
Such AI-related jobs have already emerged. In early May 2023, Super Agency RWS advertised roles for freelance linguists to support a language model training project. The role involves rewriting AI-generated user prompts to “train the model to come up with answers that are factually accurate and meet the client’s style.”
The Importance of Prompting
Prompting emerges as a crucial mechanism empowering LSPs to unlock the potential of LLMs. LLMs possess the ability to learn and apply novel concepts from just a few examples, similar to humans.
By providing instructions or examples to customize LLM output for specific tasks or contexts, prompting enables dynamic adaptation “on-the-go”, aligning translations with client glossaries, adjusting style and tone, and ensuring grammatical and phrasing choices fit the document's purpose and real-world circumstances.
This dynamic and efficient approach offers a game-changing solution for the language industry, eliminating the need for building custom models from scratch and allowing for real-time adaptation of machine translation output.
Your information will not be shared with third parties. No Spam.