ProZ.com Translation Forums

The Future Of Work Now: The Computer-Assisted Translator And Lilt | @KD

Samuel Murray — Mon, 16 Feb 2026 11:03:00 +0000

Forum: Translation news
Topic: The Future Of Work Now: The Computer-Assisted Translator And Lilt
Poster: Samuel Murray
Post title: @KD

[quote]kd42 wrote:
Or does not work, or works only partially, or is misleading. Short segments, polysemy, and a dozen other phenomena, in addition to the flaws in the corpus for training forever rule out any unsupervised translation. [/quote]
Sure. Those of us who use CAT tools know that some things work more better in more ideal circumstances, and work less better in less ideal circumstances.

[quote]kd42 wrote:
[quote]Samuel Murray wrote:
...the AI suggestion automatically updates itself and suggests a new translation that works with the word that you had just typed.[/quote]
To me, this immediate response is nothing more than a show, a visual effect for the naive, because if you edited MT 20 years ago by typing the edits, you immediately got the result you needed, now you are guessing, you don't know the way the LLM will react, this is a step back from MT, but with bright fancy coloured lamps. [/quote]
I agree, this is different from traditional MT where the translator needs to respond to the MT suggestion only once, at the point that he is presented with the suggestion. This is how I edit MT too. The Lilt method keeps changing the suggestion, so the translator has to keep re-examining the suggestion.

This is different from traditional MT. But that does not mean that it is just for show. Particularly for longer, less technical sentences, this method may offer possibilities.

Personally I would have preferred that the AI don't update the suggestion automatically, but only once the translator presses a shortcut key e.g. ENTER, and if it would be possible for the translator to cycle back to a previous suggestion (and forward again) using a shortcut e.g. Alt+left arrow and Alt+right arrow.

The current implementation of the Lilt method assumes that translators edit machine translations from left to right. Some of us do, but some of us don't: we may edit something in the middle of the sentence, then at the start, then at the end, then in the middle again, etc.

Another issue with the Lilt method is that the translator's choice of edits are often determined by the other words in the suggestion. The Lilt method assumes that every translator will always produce just one ideal translation for a given sentence. In reality, I may translate an identical sentence in 10 different ways each time I get one of 10 different MT suggestions. Which words I choose to edit depends on which other words there are already present in the suggestion.

[Edited at 2026-02-16 11:10 GMT]

The Future Of Work Now: The Computer-Assisted Translator And Lilt | A school teacher from that joke

kd42 — Mon, 16 Feb 2026 10:41:00 +0000

Forum: Translation news
Topic: The Future Of Work Now: The Computer-Assisted Translator And Lilt
Poster: kd42
Post title: A school teacher from that joke

[quote]Samuel Murray wrote:[quote]kd42 wrote:
Samuel your account sounds like ... Trivia facts without conclusions and, potentially, value. [/quote]
I wrote my post to a particular audience, who already know certain facts about the industry that I felt that I did not need to repeat in order to provide the necessary context. Admittedly, newer, younger translators may not have understood what I was trying to say. (-: [/quote]
There is that Russian joke about a school math teacher who comes home furious and complains to his wife, "Kids are so dumb! I explained the theorem to them once -- they don't get it, I explained the theorem to them twice -- they don't get it, I explained a third time and even understood it myself -- the still don't get it!"

I don't think you understand well enough the principles and the consequences. Therefore, you cannot explain it fully.

[quote]Samuel Murray wrote:
if you deviate from the AI suggestion in any way, e.g. by typing a word that is different from the AI translation, the AI suggestion automatically updates itself and suggests a new translation that works with the word that you had just typed.[/quote]
Or does not work, or works only partially, or is misleading. Short segments, polysemy, and a dozen other phenomena, in addition to the flaws in the corpus for training forever rule out any unsupervised translation. To me, this immediate response is nothing more than a show, a visual effect for the naive, because if you edited MT 20 years ago by typing the edits, you immediately got the result you needed, now you are guessing, you don't know the way the LLM will react, this is a step back from MT, but with bright fancy coloured lamps. The lamps are blinking, the naive are happy.

The Future Of Work Now: The Computer-Assisted Translator And Lilt | @Kd

Samuel Murray — Mon, 16 Feb 2026 09:49:00 +0000

Forum: Translation news
Topic: The Future Of Work Now: The Computer-Assisted Translator And Lilt
Poster: Samuel Murray
Post title: @Kd

[quote]kd42 wrote:
Samuel your account sounds like ... Trivia facts without conclusions and, potentially, value. [/quote]
I wrote my post to a particular audience, who already know certain facts about the industry that I felt that I did not need to repeat in order to provide the necessary context. Admittedly, newer, younger translators may not have understood what I was trying to say. (-:

The Future Of Work Now: The Computer-Assisted Translator And Lilt | Lilt's tricks are not impressive at all...

Li-Hsiang Hsu — Mon, 16 Feb 2026 06:55:00 +0000

Forum: Translation news
Topic: The Future Of Work Now: The Computer-Assisted Translator And Lilt
Poster: Li-Hsiang Hsu
Post title: Lilt's tricks are not impressive at all...

That is what I'd like to say. It uses LLM and a UI to trick us. We can actually DIY a workflow to save money.
[quote]kd42 wrote:

[quote]Li-Hsiang Hsu wrote:
instantly condition the LLM
triggering coherent suggestions
fluid, dynamic impression
prompt engineering in motion
structuring the output
Lilt-style workflow
Gemini API key
Python libraries
into that same loop [/quote]
To me this does not sound impressive.
A hundred years ago the tools were less advanced, but there was a social threshold below which nobody could be regarded a translator. Same good old sh@t now. The greedy people can use AI and hungry students to output data in the desired language, but once in a while they will get sued and suffer due to a stupid error.

Here's your list converted by an NN
===========

Sure 😄 — here’s a 100-years-ago translator version with the same layout and rhythm, but historically flavored:
[quote] instantly condition the typewriter
triggering coherent dictionary thumps
fluid, dynamic ink stains
pencil annotations in motion
structuring the margin notes
card-catalog workflow
trusted fountain pen
stack of reference tomes
into that same desk loop[/quote] [/quote]

The Future Of Work Now: The Computer-Assisted Translator And Lilt | edit..

Li-Hsiang Hsu — Mon, 16 Feb 2026 06:52:00 +0000

Forum: Translation news
Topic: The Future Of Work Now: The Computer-Assisted Translator And Lilt
Poster: Li-Hsiang Hsu
Post title: edit..

...

[Edited at 2026-02-16 07:05 GMT]

The Future Of Work Now: The Computer-Assisted Translator And Lilt | Buzzwords

kd42 — Sun, 15 Feb 2026 22:03:00 +0000

Forum: Translation news
Topic: The Future Of Work Now: The Computer-Assisted Translator And Lilt
Poster: kd42
Post title: Buzzwords

[quote]Li-Hsiang Hsu wrote:
instantly condition the LLM
triggering coherent suggestions
fluid, dynamic impression
prompt engineering in motion
structuring the output
Lilt-style workflow
Gemini API key
Python libraries
into that same loop [/quote]
To me this does not sound impressive.
A hundred years ago the tools were less advanced, but there was a social threshold below which nobody could be regarded a translator. Same good old sh@t now. The greedy people can use AI and hungry students to output data in the desired language, but once in a while they will get sued and suffer due to a stupid error.

Here's your list converted by an NN
===========

Sure 😄 — here’s a 100-years-ago translator version with the same layout and rhythm, but historically flavored:
[quote] instantly condition the typewriter
triggering coherent dictionary thumps
fluid, dynamic ink stains
pencil annotations in motion
structuring the margin notes
card-catalog workflow
trusted fountain pen
stack of reference tomes
into that same desk loop[/quote]

The Future Of Work Now: The Computer-Assisted Translator And Lilt | So what?

kd42 — Sun, 15 Feb 2026 21:49:00 +0000

Forum: Translation news
Topic: The Future Of Work Now: The Computer-Assisted Translator And Lilt
Poster: kd42
Post title: So what?

[quote]Samuel Murray wrote:
tab key[/quote] Samuel your account sounds like a story by a colleague who flew to worn on, say, Madagascar for a while.
"They have XX dishes cooked from ZZ!"
"Their women often XX if ZZ!"
Trivia facts without conclusions and, potentially, value.

The Future Of Work Now: The Computer-Assisted Translator And Lilt | MTPE is just the same as TRA-from-scratch except only for a lower rate

Rıdvan Patat — Fri, 13 Feb 2026 11:52:00 +0000

Forum: Translation news
Topic: The Future Of Work Now: The Computer-Assisted Translator And Lilt
Poster: Rıdvan Patat
Post title: MTPE is just the same as TRA-from-scratch except only for a lower rate

[quote]Li-Hsiang Hsu wrote:

...isn’t that they’re using MT - it’s that their MT is just not good. If you’re gonna ask for lower MTPE rates, at least give me a decent MT text to work with. Something that actually performs - like the paid options or good LLMs. Better than what I can rig up myself. But honestly, the MT quality they’re serving is pretty disappointing - for the most part.

I just roll my own AI-assisted setup - and I end up with better AITPE than what most agencies deliver. That is, if the client’s OK with AI-assisted translation.

I always pre-treat the text before it ever hits the LLM - using POS tagging and NER to mask sensitive information and keep client data confidential. It takes some extra prep work, sure. But here’s the thing: most agencies don’t even bother.

[quote]Daniel Frisano wrote:

[quote]Li-Hsiang Hsu wrote:

Clients and agencies routinely supply poorly machine-translated bilingual files... [/quote]

From what I've seen, many 'translators' deliver hastily-edited poorly machine-translated files too, charging for the real thing...

I've heard many 'colleagues' complain of clients using MT, but then they use MT themselves. Like 80% of those reading this post.

We have no one but ourselves to blame, folks.
[/quote]

[Edited at 2026-02-12 13:29 GMT]

[Edited at 2026-02-12 13:53 GMT] [/quote]

I've been accepting MTPE tasks recently due to lack actual, proper translation work.

and in my experience so far, MTPE is just the same as TRA-from-scratch except only for a lower rate.

The Future Of Work Now: The Computer-Assisted Translator And Lilt | Where is the cost cutting?

Peter Motte — Fri, 13 Feb 2026 10:38:00 +0000

Forum: Translation news
Topic: The Future Of Work Now: The Computer-Assisted Translator And Lilt
Poster: Peter Motte
Post title: Where is the cost cutting?

[quote]Mario Chávez wrote:
Translation agencies that didn't use sound workflows for their CAT tools and other technologies had either to fold or lose clients. Cost-cutting translation agencies (many of our clients) forced some of us to cut rates in order to remain competitive.
[/quote]

I often wonder where the cost-cutting is.
Is it because the tools can speed up the process?
Or is it because the tools can be used to put workers under pressure to lower their costs?
Yesterday I got a demand from somebody who asked me to work at 15 USD/hour. I flatly refused, of course.
Today somebody want 14 / hour. And he didn't even mention the currency.

The Future Of Work Now: The Computer-Assisted Translator And Lilt | What gets me...

Li-Hsiang Hsu — Thu, 12 Feb 2026 13:22:00 +0000

Forum: Translation news
Topic: The Future Of Work Now: The Computer-Assisted Translator And Lilt
Poster: Li-Hsiang Hsu
Post title: What gets me...

...isn’t that they’re using MT - it’s that their MT is just not good. If you’re gonna ask for lower MTPE rates, at least give me a decent MT text to work with. Something that actually performs - like the paid options or good LLMs. Better than what I can rig up myself. But honestly, the MT quality they’re serving is pretty disappointing - for the most part.

I just roll my own AI-assisted setup - and I end up with better AITPE than what most agencies deliver. That is, if the client’s OK with AI-assisted translation.

I always pre-treat the text before it ever hits the LLM - using POS tagging and NER to mask sensitive information and keep client data confidential. It takes some extra prep work, sure. But here’s the thing: most agencies don’t even bother.

[quote]Daniel Frisano wrote:

[quote]Li-Hsiang Hsu wrote:

Clients and agencies routinely supply poorly machine-translated bilingual files... [/quote]

From what I've seen, many 'translators' deliver hastily-edited poorly machine-translated files too, charging for the real thing...

I've heard many 'colleagues' complain of clients using MT, but then they use MT themselves. Like 80% of those reading this post.

We have no one but ourselves to blame, folks.
[/quote]

[Edited at 2026-02-12 13:29 GMT]

[Edited at 2026-02-12 13:53 GMT]

The Future Of Work Now: The Computer-Assisted Translator And Lilt | The killer feature is a real-time "Edit-as-Prompt" loop

Li-Hsiang Hsu — Thu, 12 Feb 2026 13:00:00 +0000

Forum: Translation news
Topic: The Future Of Work Now: The Computer-Assisted Translator And Lilt
Poster: Li-Hsiang Hsu
Post title: The killer feature is a real-time "Edit-as-Prompt" loop

where your edits instantly condition the LLM, triggering coherent suggestions instantly as you type. This creates a fluid, dynamic impression far beyond the static suggestions of traditional CAT tool.

To demystify: what’s really happening there is prompt engineering in motion - regenerating text and structuring the output directly in the UI.

I actually DIY’d a Lilt-style workflow using just a Gemini API key and Python libraries. But here’s the cool part: we can use TMs into that same loop and keep them updated.

[quote]Samuel Murray wrote:

The killer feature is that if you deviate from the AI suggestion in any way, e.g. by typing a word that is different from the AI translation, the AI suggestion automatically updates itself and suggests a new translation that works with the word that you had just typed. I speculate that if you insert the AI translation into the editing box, and then start editing it, the AI suggestion box would also update with new suggestions that make use of your edits.

[url removed]

I would love to test drive this, but it is only accessible to corporate clients. Which kind of makes sense, since each "translation memory" is an LLM. [/quote]

The Future Of Work Now: The Computer-Assisted Translator And Lilt | What Lilt is

Samuel Murray — Thu, 12 Feb 2026 07:21:00 +0000

Forum: Translation news
Topic: The Future Of Work Now: The Computer-Assisted Translator And Lilt
Poster: Samuel Murray
Post title: What Lilt is

Let me save you some time: Lilt is an online translation interface for AI. It is a "CAT tool" only in the sense that it segments the text and presents you with one segment at a time. From what I can tell, it doesn't have a translation memory, but instead uses the translator's translations to train its own AI (although it does remember each translator's individual translation style and preferences).

The UI consists of three boxes: the source text, a box where you type your translation, and an AI suggested translation of the source text. You can type the translation yourself and/or you can "type" the translation by pressing the tab key. Each time you press the tab key, it inserts the next word from the AI translation.

The killer feature is that if you deviate from the AI suggestion in any way, e.g. by typing a word that is different from the AI translation, the AI suggestion automatically updates itself and suggests a new translation that works with the word that you had just typed. I speculate that if you insert the AI translation into the editing box, and then start editing it, the AI suggestion box would also update with new suggestions that make use of your edits.

[url removed]

I would love to test drive this, but it is only accessible to corporate clients. Which kind of makes sense, since each "translation memory" is an LLM.

The Future Of Work Now: The Computer-Assisted Translator And Lilt | Ehm

Daniel Frisano — Wed, 11 Feb 2026 23:33:00 +0000

Forum: Translation news
Topic: The Future Of Work Now: The Computer-Assisted Translator And Lilt
Poster: Daniel Frisano
Post title: Ehm

[quote]Li-Hsiang Hsu wrote:

Clients and agencies routinely supply poorly machine-translated bilingual files... [/quote]

From what I've seen, many 'translators' deliver hastily-edited poorly machine-translated files too, charging for the real thing...

I've heard many 'colleagues' complain of clients using MT, but then they use MT themselves. Like 80% of those reading this post. Or maybe 90%? More like 95%?

We have no one but ourselves to blame, folks.

[Edited at 2026-02-12 13:21 GMT]

The Future Of Work Now: The Computer-Assisted Translator And Lilt | Undeniable progress?

Mario Chávez — Wed, 11 Feb 2026 21:53:00 +0000

Forum: Translation news
Topic: The Future Of Work Now: The Computer-Assisted Translator And Lilt
Poster: Mario Chávez
Post title: Undeniable progress?

There are phrases (such as “undeniable progress”) which, if found by a university professor in a paper or thesis draft, would deserve a correction.

We are translators, yes, but we are expected to operate with demonstrable facts, not opinion, let alone regurgitated marketing pieces.

On the other hand, the phrase “cutting costs” is a demonstrable fact, it can be independently checked. Translation agencies that didn't use sound workflows for their CAT tools and other technologies had either to fold or lose clients. Cost-cutting translation agencies (many of our clients) forced some of us to cut rates in order to remain competitive.

MTPE is a double-edged sword: like all technology-based solutions or approaches, it can be a boon if properly designed and implemented. Or not.

Fuzzy matches, on the other hand, require more finesse and experience. For instance, I've built a solid TM from translating hundreds of similar medical reports over 10-12 months. However, a frequent phrase hits 99% because the date is different. In other contexts or types of documents, 99% may be treated as 100% because the difference lies in tags, not content. My point? I have to review each 99% match.

Generative AI will come and go. Not all translators and translation agencies may survive the untrammeled advance of well-moneyed AI engines and platforms and their intrusive nature in human workflows, but the future as painted by tech bros, uber capitalists and investors is never as shiny and fulfilling as they promise.

[quote]Li-Hsiang Hsu wrote:

The rise of MT, CAT tools, and now AI has brought undeniable progress—streamlining workflows, boosting productivity, and cutting costs. Yet these efficiencies have consistently failed to translate into increased revenue for translators.

MTPE is a joke. In its current form, it often feels like a pretext. Clients and agencies routinely supply poorly machine-translated bilingual files, demanding reduced rates, even when the raw MT output fails to accelerate the process or improve quality. It is less a tool for collaboration and more a lever to drive down compensation.

Similarly, the requirement to translate against fuzzy matches below 85-90% is rarely justified by any real time savings. Nevertheless, it is imposed to secure lower rates, disregarding the negligible gain from such minimal matches.

While AI can genuinely enhance translation speed and quality, it is already being used as the latest justification for cutting rates. Once again, translators see the benefits of technological advancement harnessed by others, while they themselves are left with diminishing returns. [/quote]

The Future Of Work Now: The Computer-Assisted Translator And Lilt | It's capitalism.

Eugenio Garcia-Salmones — Wed, 11 Feb 2026 08:39:00 +0000

Forum: Translation news
Topic: The Future Of Work Now: The Computer-Assisted Translator And Lilt
Poster: Eugenio Garcia-Salmones
Post title: It's capitalism.

It's capitalism. That money that used to be distributed among 6,000,000 translators now ends up in the hands of 15 large companies. Even if the reality is as you describe, it's even more so in a language as rich as Chinese, with such an enormous tradition. Machines will never be able to translate Chinese correctly, but 90% of the work that will come to Chinese translators in the future will be MTPE.

The Future Of Work Now: The Computer-Assisted Translator And Lilt | Helps to reduce costs, not to increase translator's revenue

Li-Hsiang Hsu — Wed, 11 Feb 2026 07:49:00 +0000

Forum: Translation news
Topic: The Future Of Work Now: The Computer-Assisted Translator And Lilt
Poster: Li-Hsiang Hsu
Post title: Helps to reduce costs, not to increase translator's revenue

The rise of MT, CAT tools, and now AI has brought undeniable progress—streamlining workflows, boosting productivity, and cutting costs. Yet these efficiencies have consistently failed to translate into increased revenue for translators.

MTPE is a joke. In its current form, it often feels like a pretext. Clients and agencies routinely supply poorly machine-translated bilingual files, demanding reduced rates, even when the raw MT output fails to accelerate the process or improve quality. It is less a tool for collaboration and more a lever to drive down compensation.

Similarly, the requirement to translate against fuzzy matches below 85-90% is rarely justified by any real time savings. Nevertheless, it is imposed to secure lower rates, disregarding the negligible gain from such minimal matches.

While AI can genuinely enhance translation speed and quality, it is already being used as the latest justification for cutting rates. Once again, translators see the benefits of technological advancement harnessed by others, while they themselves are left with diminishing returns.