ChinAI #258: Is translation already dead in the AI era?

Greetings from a world where…

translation is an art of disappointment

…***We’ve hit a bit of a lull in paid subscriptions lately, so please consider subscribing here to support ChinAI under a Guardian/Wikipedia-style tipping model (everyone gets the same content but those who can pay support access for all AND compensation for awesome ChinAI contributors). As always, the searchable archive of all past issues is here.

Feature Translation: Did AI kill the translator?

Context: This FT Chinese op-ed (link to original Chinese), authored by a translator who draws on her experience doing English-to-Chinese translation work, provides some fascinating insights into the future of translation in the AI age.

Key Passages: To jump straight in, I like how she illustrated the differences between ChatGPT and Google Translate for translators.

• When it comes to content that is complex or has subtle connotations, AI translation will still be incoherent or full of errors. In terms of translation quality alone, ChatGPT is not superior to Google Translate, and its translations are often very rough. However, it can bring about human-computer interactions…

• …I used to call my sister to discuss things when I was working as a translator. She studies Chinese linguistics. Sometimes I understand the meaning of a piece of English, but I can't think of an authentic Chinese expression. I would explain to her what I need to express, and she would give me some inspiration. Now ChatGPT can do well at taking on the role of this linguist consultant and can be on call anytime and anywhere.

The op-ed provides two detailed examples that illuminate subtle limitations in machine translation.

• First, consider this online discussion about a long-standing misconception that a bilingual environment will cause delayed language development in children: A netizen…insisted that a bilingual environment will make children learn to speak later. I asked her to provide authoritative evidence, and she searched for a long time and posted a screenshot, which was an abstract of an academic paper. The first sentence was: "There was limited evidence to suggest that bilingual children develop speech at a slower rate than their monolingual peers”. She interpreted this sentence in Chinese as: "Although there is not much evidence, there is indeed evidence that bilingual children develop slower than monolingual children." I immediately saw the crux of the problem…Later, I entered this sentence into Google Translate and then ChatGPT. The translation that came out was indeed the same as the netizen's interpretation, which was "there was evidence (with some limits) that bilingual children develop slower than their monolingual peers." Just look at it. In this translation, it seems correct to literally translate “limited evidence” as "evidence with some limits", and the translation is also smooth. However, in an academic context, "limited evidence" does not simply mean that the amount of evidence is small, but that the amount and reliability of the evidence are so limited that it is not enough to support a certain conclusion.

• Second, translating an English poem into a Chinese poem in classical Chinese format: I really couldn't think of a suitable word for a sentence, so I asked ChatGPT for help, asking it to tell me what words meant repetition/imitation and that rhymed with the word "印 [yin4]". It listed 6 words for me, 复制、仿制, etc., none of which rhymed with "yin4". I also emphasized that I needed a word whose last word had the final rhyme "in" and which meant imitation or repetition. It again gave me a list of 7 words that either rhymed but had the wrong meaning, or had the same meaning but didn't rhyme. Finally I rearranged the rhyme (scheme) of the entire paragraph and completed the entire poem on my own.

I also really liked her comparison about how the work of translators will evolve as AI advances further.

• “If human translators are chefs, then AI translators are like a super combination of McDonald's plus pre-made dishes and cooking machines. McDonald's can provide diners with convenient, hygienic, and affordable food. Many people, especially children, also like the taste of its food. If you are in a rush for time and don’t have high requirements for the color, aroma and nutrition of food, McDonald’s is a good choice. But it would be much too naive to think that because McDonald’s exists, then other restaurants will close, or that with pre-made dishes and cooking machines, chefs all over the world will lose their jobs.”

• She continues: “Excellent human translators, like chefs at Michelin three-star restaurants, will become ‘luxury service’ providers, serving only those customers who have very high requirements for translation quality. Newcomers or people with average skills who have just entered this industry are faced with two choices: either work hard to reach the top and join the high-end market; or they can combine translation with other skills and use their creativity to explore new areas, such as specifically writing prompt words that allow AI to adjust/improve the translation, or use AI to write simple bilingual materials.”

Here at ChinAI, we are definitely not serving you three-star Michelin cuisine. I prefer to think of ChinAI translations as the soup noodles from that hole-in-the-wall shop which all the locals swear by. Keep reminding me that we shouldn’t get too big; otherwise, the influencers will infiltrate and ruin everything.

FULL TRANSLATION: In the AI era, is translation already dead?

ChinAI Links (Four to Forward)

Must-read: China x AI Reference List

The China & Global Priorities Group has put together an excellent reading list on topics at the intersection of China and AI. It’s a 32-p. google doc of helpful resources on this topic — carefully curated and organized!

Should-read: Thread on challenges to China’s domestic large AI model development

Kevin Xu helpfully unpacks three fundamental challenges, based on a slide that the Beijing Academy of Artificial Intelligence showed Chinese Premier Li Qiang during his visit, which was featured on CCTV (h/t to Bill Bishop for first flagging this).

Should-read: How Developers Steer Language Model Outputs: LLMs Explained

This series of three CSET explainers have been very helpful for me. In part 2, Thomas Woodside and Helen Toner explore fine-tuning: “a set of techniques used to change the types of output that pre-trained models produce.”

Should-read: CCID’s research report on computing service providers

Every now and then, I like to check in on what CCID (an influential think tank in the information technology space) has published lately. This February report (in Chinese) analyzes the competitiveness of China’s top 50 computing service providers.

Thank you for reading and engaging.

These are Jeff Ding's (sometimes) weekly translations of Chinese-language musings on AI and related topics. Jeff is an Assistant Professor of Political Science at George Washington University.

Check out the archive of all past issues here & please subscribe here to support ChinAI under a Guardian/Wikipedia-style tipping model (everyone gets the same content but those who can pay for a subscription will support access for all).

Also! Listen to narrations of the ChinAI Newsletter in podcast format here.

Any suggestions or feedback? Let me know at chinainewsletter@gmail.com or on Twitter at @jjding99