Thoughts on Machine Translation - Intellectual Property Advent Calendar 2021 -

Introduction

This article is a submission for the "Intellectual Property Advent Calendar 2021", organized by Patent Salon. I am involved in patent translation (Japanese translation). I have summarized my current thoughts after actually using machine translation. By the way, when I say machine translation here, I am referring to what is called AI translation or automated translation, such as Google Translate or DeepL. Please note this.

Impressions of Machine Translation

First, my impressions of machine translation. I feel that translation accuracy is improving year by year. While it depends on the field, I think things that are capable of practical use are emerging. For example, I feel that it can be used sufficiently for manuals, legal documents, patents, and medical texts. However, one should always use machine translation with the assumption that mistranslations exist. You should know its limits through personal experience.

There are many such machine translation tools released. If you are going to use translated text as a product, I think it is better to use a dedicated engine specialized for each field. In fact, I compared general-purpose engines (like Google Translate and DeepL) with dedicated engines (like T-4OO and Japio-AI translation), and I got the impression that the translation accuracy of the dedicated engines was better. Since dedicated engines are almost certainly paid services (and reasonably expensive), it is a pity that not everyone can introduce them immediately.

However, I think that if an advanced translator masters it, it can become a formidable (useful) tool. Depending on the item, you can use a custom engine matched to your own translation targets, so in a sense, you are creating a degraded copy of yourself, right? This means it reproduces your own habits, so I think it might be very easy to handle. But as a warning, if you create a custom engine with inappropriate translation pairs, the translated text will become messy (I have experienced this). Therefore, I think it also has an aspect of being a tool for advanced users.

Translators and Machine Translation

I think there are various opinions about using machine translation as a translator, but I am a fan of machine translation (lol). I think it is amazing technology (interesting technology), and I believe it is very useful for our lives and businesses, especially for those other than translators (generally speaking, the negative aspects might be greater for the lives and businesses of translators...).

If you are going to use machine translation, you first need to know the habits of the machine translation. I think it is better to check what kind of mistakes it makes, look at its tendencies in translation methods, and check its compatibility with yourself for each translation engine, each Japanese translation, and each English translation. As I said earlier, you should use a dedicated engine for translation. This is because there are fewer mistranslations compared to general-purpose engines, and it would be even better if you could customize the dedicated engine based on personal or corporate specifications. Perhaps it might be better to use different engines for different types of translated texts. I feel that there are pros and cons depending on the field, such as Company A's engine for patents and Company B's engine for medical texts (there are also personal preferences). By the way, I think general-purpose engines can also be quite useful. First, try using them in practice. To put it extremely, even if there are many mistranslations, if it is easy for you to correct (easy to handle), then that engine is better, right?

Also, as perhaps the most important thing, there is the processing of the translated text before putting it into machine translation. It actually has a significant effect on translation accuracy. This is the work called pre-editing. What do you do? You prepare the text to make it easier for the machine translation to process. For example, for simple things, it might be deleting unnecessary spaces or line breaks, not omitting the subject for English translation, or changing hiragana to kanji. For advanced things, it might be editing the original text while considering the habits of the machine translation. This is something that cannot be done without knowing the habits of the machine translation. Since translation accuracy can change significantly with this pre-editing, I would like to work on it as thoroughly as possible.

Also, just to be clear, machine translation always makes mistakes, so you must check the translated text output by the machine translation from beginning to end. For personal use, such as wanting to know the content of a certain original text to some extent, I don't think you need to go that far, but it is necessary if you are creating translated text as a product. Therefore, when using the output results of machine translation, I think you should explore the usage that suits you, including how to use it.

Users/Learners and Machine Translation

Moving away from the translator's perspective, as for what I think from the perspective of users and learners, I wish I had the high-precision machine translation that exists today when I was in university. I went to a science university, but most people hate English (right?). I was no exception. Leaving aside the question of "Why are you translating?" here, if there had been high-precision machine translation, the psychological and physical hurdles to English might have been lowered, and I think I could have increased my opportunities to come into contact with English. If that had happened, my skills in gathering information from English-speaking countries might have become more skillful, and I sometimes think that I might have lived a different life.

In addition, I think that I might have been able to learn the English reading and writing necessary for business efficiently, and acquired the skills to handle foreign language work that would be necessary in the future. What I thought when I started translating is that, for example, to master Japanese translation, there is no choice but to repeat reading actual English sentences and translating them into Japanese, relying on basic grammar and vocabulary. Trial and error in a practical format is the best. I think high-precision machine translation is optimal as an auxiliary tool to promote that trial and error.

However, there is something you must be careful about when using machine translation. Since machine translation always makes mistakes, an attitude of believing the output results as they are is not acceptable. You need to carefully examine it sentence by sentence. For example, you should examine whether the translation is correct according to your own grammar and vocabulary skills, or compare it with the translation you thought of yourself. Furthermore, it would be good if you could refer to similar texts translated by professionals. If you cannot judge at this time, I think it is better to just skip it and keep moving forward. For now, even if you don't understand, I think you should keep fighting!

I said it roughly for now, but this methodology is something I am currently experimenting with, and I would like to refine it while sharing information from now on. Oh yes, as for what I think, I think it is better to separate "English reading and writing necessary for business, etc." and "translation" for the time being. For example, I think there are cases where you see a translated Japanese text and think, "I understand the meaning, but it feels like strange Japanese." If it is sufficient for understanding the text but insufficient as a text to show others, I think it is fine to stop at the understanding level if you are just reading it, but what do you think? (lol)

Business and Machine Translation

Now, let's move on to business. First of all, it must be a major premise that machine translation cannot be a substitute for a translator. Incredibly, there are people who think it can be a substitute for a translator, but that is impossible for now. I will say it again and again, machine translation always makes mistakes, so you should not use the translated text as is. Even if you use the machine-translated text almost as is, the person who used it must guarantee the quality.

And isn't the person (company) using it the only one who benefits from using machine translation? That benefit is an improvement in translation processing speed (shortening delivery time). There are companies that often think that costs will go down if they introduce machine translation, but that is not necessarily the case. Isn't it depending on how you do it? Costs only go down if it is completed internally, and if you outsource it, you should not put machine translation in between. As I have been saying, machine translation always makes mistakes. You must check everything from beginning to end. "It's machine translation, so if you just correct it a little, you can get a translation. So the translation fee can be cheap, right?" That is an illusion. This is something you can understand if you experience "translation" yourself, but it is a point that is hard to get people to understand...

So, there are MTPE projects. This MTPE is a translation project where you post-edit (PE) the text translated by machine translation (MT) and deliver it. By the way, post-editing is the work of correcting the translated text. MTPE projects are set with lower compensation compared to regular translation projects. I think the majority of the reason is "because it requires less effort." But, in reality, while the processing speed might become faster, the effort does not change. If you are not careful, it is not rare for the effort to increase (in short, when the translated text is terrible). Therefore, the more advanced the translator, the stronger the tendency to dislike MTPE projects. Also, many people dislike it because they say their own translation ability will decline due to the influence of machine translation.

Regarding this MTPE project, I personally feel it is not suitable for outsourcing. To make the most of machine translation, I believe it is necessary at the very least to understand the quirks of the translation engine being used, for the translator to use an engine they are comfortable with, and to use a custom engine adapted to the specific project. Fundamentally, with MTPE projects, I don't know what translation engine is being used (they won't tell me), nor do I know if it's an engine that is easy for me to use. I believe machine translation is something that only shows its true potential when it is thoroughly customized for an individual or company and mastered by them. Therefore, being handed something of unknown origin is just a nuisance. In reality, since I am editing something of unknown origin that is full of mistakes, the translation fee should be higher, right? It might be nice if they would at least let the translator freely use the translation engine that was used, but well, that's probably impossible (laughs).

Therefore, I believe that if you are going to use machine translation, you should establish a system that can be completed internally. In the case of companies, if you are introducing machine translation to utilize foreign language documents and information, there are not many things to consider, but when it comes to creating translated materials, the hurdle rises all at once, doesn't it? I imagine it is surprisingly difficult because you have to prepare both people who can translate in-house and a management system for machine translation. However, once the system is in place, I think the speed at which translated materials are completed and the cost reduction effect compared to outsourcing will be significant.

In conclusion

I have written down my thoughts on machine translation. I like machine translation. Through machine translation, I would like to aim for collaboration with AI. And I would also like to think about machine translation literacy and convey it to everyone. Therefore, these days I am thinking that I would like to pioneer machine translation education and consulting businesses, aiming for the healthy spread of machine translation while verifying its usefulness through practice.