Hello.
Thank you for inviting AILIA, the Language Industry Association, to lean in on these important conversations. Here is a quick brief about AILIA.
It was created in 2003, with a mission to increase the visibility of the language industry, promote and advocate for the language industry, increase competitiveness, act as a spokesperson for the industry, share information, and support high standards for quality, as with the development of the national standard CGSB 131.10. All the members of the board, our working board, are all volunteers.
Upon receipt of the request to speak today, we asked the entirety of the AILIA membership for their comments with regard to a few questions, and I will be presenting the responses to these questions.
These were the questions: What has been the impact of machine translation on your business, marketplace, and the Translation Bureau? What has been your experience working with the Translation Bureau? All of these comments are an amalgamation of all of those.
First, we encountered misguided beliefs that machine translation with post-editing can deliver results that are equivalent to professional translation. What we are seeing more and more in the industry is a belief that machine translation with post-editing can deliver results that are equivalent to professional translation or leveraging translation memory tools, which is not true. Clients are prepared to embark on this high-risk technological adventure completely uninformed. No matter how one uses it or looks at it, machine translation remains high-risk technology, for several reasons.
First and foremost, there is no actual translation or communication involved. There is a gross misinterpretation of the term “machine translation” by the public and how it works. As we know, machine translation works strictly by statistical matching, without any human comprehension, knowledge of language, or validation of meaning. As a result, machine translation tools generate all kinds of unpredictable errors and nonsensical output that can also be downright offensive.
We encountered a lack of understanding of the real implications of post-editing. Post-editing involves people reviewing and trying to improve the raw machine translation output, which is extremely tricky because of all the unpredictable mistakes and nonsense, things that you would never encounter in the work of human translators. The industry is faced with a complete lack of understanding by the public of the time involved in post-editing with uncontrolled use of machine translation.
In these discussions, we've heard about the pre-editing and post-editing required to successfully work with machine translation. This information is not in the sphere of understanding, even for many who are experienced translators in the industry. In other words, post-editing is a very different exercise from professional revision. It is a type of work for which there is little or no training. So very different is this work that there is now an ISO standard being prepared, a copy which I have provided to the greffier so that he can circulate it. It's at the DIS stage. I can't provide it to you. Only the link for consultation is available.
It's interesting how it is described. The section about pre-production processes begins with the concept of whether the source language content is suitable for machine translation, differing very much from other tools. Because the subsequent post-editing, combined with machine translation efficiency, depends on the machine translation system, language combination, domain, style, and source language content, this is very complex, which is what I am trying to convey. There is a 20-page standard coming out about it.
It goes further to specify that requirements for post-editing must be identified, documented, and made accessible to the post-editor at the onset of the project. Post-editors must be informed of the level of estimated usefulness of the machine translation output.
The output must also be easily identifiable to the post-editor by way of indication marks to correctly distinguish between machine translation and output from other sources. That is how very different and difficult it is to use pure, raw machine translation.
There were a lot of comments.
We see requests for post-editing often turning into complete retranslation in the marketplace. Our most recent experience shows this happens primarily because most machine translation output is useless. It needs specialists at the onset, specialists using it, specialists programming it, and specialists inputting the information, which brings with it the complexity of explaining to clients that it's not faster or cheaper. This is a necessary burden for the industry in general.
The bottom line is that we're dealing with high risk that brings few efficiency gains when the need is for real translation.
The major strategic mistake is that language experts are made to work downstream, where they must find and fix machine-generated errors and nonsense.
We fully understand the challenges of a nearly exponential rise in translation and multilingual communications needs in society and government as a whole. However, we strongly believe that the self-serve use of machine translation by Government of Canada employees who are not language experts, without any supervision or validation by language experts or stringent policies to prevent the sharing or circulation of machine translation output, entails unacceptable risks that would outweigh the expected benefits and convenience, all of which machine translation does not competently address.
What can we say about machine translation in the hands of the public?
It should be used only for gist translations, such as those Google has long provided. When someone gets the gist of a text they don't understand, it doesn't mean that it's going to be as useful as they think it will be. It can help them make decisions on whether texts are to be translated if the content of the text is such that they want to translate it, or they'll need it, or it's pertinent. Again, it's a bit of a fallacy. We can fall into not understanding sufficient content, even with machine translation.
The tool developed by the Translation Bureau lends itself to misuse and misinterpretation. It's relatively unknown to the industry. I had to knock on the door and ask to be let in to have a demonstration.
From what I saw, it seems to be very basic. It has no bitext ability and no pre-analysis content. It has only a Google-like statistical matching at best. For gisting of the content it's viable, but what is to say that civil servants who have this tool will not use it in their communications?
We've heard all about the phantom units all over the federal government. What is to say that if they're not respecting the actual law, they'll respect the intended use of this tool?
The message or image being sent to the public with this machine translation is that machine translation is now government approved.
As for our recommendations, the first is to educate. Educate parliamentarians, civil servants, and federal employees to start with. Continue to do so with all documentation on the Translation Bureau website. We're presuming this will not go away. Educate them on the philosophy, the concept, and the ultimate uses of machine translation. Educate them about the profession of translators, about revisers, and about post-editing. Educate the Translation Bureau about respecting their intended audience, about the gains and risks, and about creating access without chaos.
Inform the public. We must decide on the value we place on our language heritage and the quality of language with which we want to address our citizens.
We are convinced that any imposed reliance on machine translation post-editing by the Government of Canada is a hazardous strategy that would risk turning a world-renowned area of expertise built over decades into low-quality mass production work unworthy of Canada's proud cultural heritage, its government, its citizens, and its enviable place among the world's developed nations.
It would amount to the gradual destruction of a strong economic sector for Canada involving thousands and thousands of high-skilled jobs. It would be replaced with a low-paying, low-value industry that is likely to move entirely out of the country over time to parts of the world where wages reflect poor standards of living.
To set the record straight, put language experts at the forefront of the process. As we've heard, tools intended to be used for real communication must be put in the hands of professional translators. Therefore, what is required is to put language experts in control upstream of the entire process where they can analyze content and make optimal decisions before processing.
