💁‍♂️ ChatGPT drove the whole world crazy. No sooner had OpenAI released version 4, than all AI hell broke loose:

😃 Duolingo introduced Duolingo Max, a subscription option where an AI interlocutor can explain to the user why a particular answer is correct or wrong and just chat with them.

😃 GPT-4 has become available on Bing immediately. As it turned out, it had already been secretly used there earlier.

😃 An AI chatbot will appear even in CrowdIn. The beta version is currently being tested.

😃 Paul Filkin is trying to construct regular expressions using GPT—and on the 5th attempt, AI managed to create a query that adds thousand separators to multi-digit numbers.

* * *
What’s happening now is a hype of the first days of fun with an exciting toy. But it will take some time for this toy to have a powerful impact on our work. Let’s remember Amara’s law:

We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run.

#duolingo #crowdin #regularexpressions #microsoft

📤 A quick reference on how to create files of different types in memoQ (the file is assumed to be already translated).

Right-click any file (or several selected files) on the Project home tab in the memoQ interface, then select what you need:

😃 To create an MQXLZ file: Export > Export Bilingual > memoQ XLIFF > check the Save a compressed file box.

😃 To create an MQXLIFF file: Export > Export Bilingual > memoQ XLIFF > uncheck the Save a compressed file box.

😃 To create an MQBACK file: right-click the memoQ file > Deliver/Return. Please note: this command works only in an offline project! In an online one, the command will upload the file to the server and close it in the current role, so don’t take offense.

😃 To convert a memoQ file to the Trados Word format (as if it was translated into Word using Trados 2007): Export > Export Bilingual > Trados DOC/RTF.

😃 To create a clean file: Export > Export (Stored Path).

#memoQ #techtips #tradosstudio #word

💻 News from the translation software market:

😃 CrowdIn has released a desktop version for Windows, Mac, and Linux.

😃 Update 8.0 for Across and even several updates for this update have been released.

😃 DeepL for Windows is now available in Microsoft Store.

😃 For those who localize subtitles: Subtitle Edit v3.6.12 has been released. There are a lot of changes, they are listed here.

#softwarefortranslators #across #deepl #videotranslation #subtitleedit

📧 This post is not about translation, but most of us use email addresses in our work, and here are some useful links related to them:

😃 Have I Been Pwned? is the most famous service that helps you check if your email address is on the list of mailboxes whose passwords have been figured out by bad people. Check your email addresses right now.

😃 How to quickly determine whether you are receiving emails from a regular or suspicious address: enter it on the Simple Email Reputation website. The service quickly checks whether the address appears in various block lists (this is a bad sign), whether it is used for registration of user profiles on social networks (this is a good sign), etc.

😃 Some tips on how to avoid being treated with suspicion or contempt because of your email address.

#usefullinks

🖍 By default, Trados Studio checks spelling using the free Hunspell dictionaries, but you can enable the Word spelling module in Trados Studio instead. This is how to do it:

😁 In Trados Studio, choose File > Options. The Options dialog box appears.

😁 In the left pane, select Editor, and then, in the list that appears, select Spelling.

😁 In the drop-down list on the right side of the Active Spell Checker dialog box, select MS Word Spell Checker.

We recommend checking the Check spelling as you type checkbox: Trados Studio will highlight spelling mistakes in red as you type. Below you can also specify whether spelling mistakes in locked segments, 100% matches, and CM/PM segments are to be ignored.

This and other posts are also available on our blog.

#tradosstudio #techtips #spelling

🤖 What you need to pay attention to when using machine translation (MT):

😃 Meaning: This is the most important problem of machine translation and the dividing line between it and human translation. MT easily produces complete nonsense at the same speed as the seemingly well-thought text.

😃 Omissions and additions: MT may simply throw out some parts of the source text or invent something that was not there. This mistake is also made by humans, but for other reasons. Technical translators working in CAT tools most often make such mistakes when they are in a hurry and thoughtlessly confirm fuzzy matches, mistaking them for 100% matches.

😃 Terminology: The same term can be translated by MT differently even within the same sentence. For a technical translator, this is a disaster. They try to get rid of it by “training” MT engines. But this requires a large amount of text in a particular terminological field, at least 500,000 segments translated by a human. In most cases, they do not exist at all.

😃 Gender and sex: This is the area where MT services are extremely politically incorrect. MT will mix formal and informal ways of addressing people. The woman from the previous sentence can be easily transformed into a man in the next one. If you use machine translation, make sure that your agent, including the inanimate one, does not change its gender unnoticed (“Maria picked up the watch. He was beautiful.”). Particular attention should be paid to personal pronouns (he, she, his, hers, etc.) and other parts of speech that have morphological signs of gender.

😃 Short sentences: They have many flaws because MT lacks context. Translation in the absence of context is a challenge for humans as well.

😃 Letter case and punctuation: Sometimes, the result of machine translation changes as soon as you change the case of the first letter of the sentence, add or remove a period, etc. Most MT services have successfully overcome this flaw, but it has not disappeared completely.

😃 Lower quality if neither the source nor the target language is English: English is the language of international communication and, most often, the native language of MT service developers. English datasets used to train their neural MT networks (NMT) are the largest in size. For a rare language pair, such as Hungarian—Swahili, there is simply no dataset of sufficient size to train a neural MT network. Therefore, machine translation in such language pairs often occurs through the mediation of English: Hungarian → English, then English → Swahili. That is, double machine translation is performed, and the number of all the defects mentioned above doubles. They are trying to fight this drawback, too.

😃 Humor, wordplay, allusions, hints, allegories: For machine translation, this is still terra incognita: it simply does not identify them and translates them literally. MT is not good at humor and wit.

#machinetranslation #editing #english