Myths and truths: the translation profession

If German is one of your working languages, I invite you to hop over to my German blog to watch my latest video post in German, where I talk about a few common misconceptions about the translation profession.

Link to blog post: https://hippe-heisler-german.blogspot.com/2023/08/mythen-und-wahrheiten-der.html
 

Link to YouTube video: https://youtu.be/ftk0uF6XSio

 

Thanks for watching!

ChatGPT for translators: first steps

Is ChatGPT good at translation? What can it do – and not do – for professional translators, and how should effective ChatGPT prompts be written?

In the blog article below I will explain how to get started with ChatGPT and provide tips for effective ChatGPT prompt engineering, including examples of how it could be used by translators in their work.

 

First, though, I will share a few personal thoughts about ChatGPT in a video. Watch it here:

 

If you're receiving this blog post via email, you can watch the video on my blog here or on YouTube here.

 

Here is the link to the blog post "Why translators don't fear the machines" of 1 February 2020, which I'm referring to in the video above.

Getting started with ChatGPT

Getting started with ChatGPT is easy. Visit https://chat.openai.com/chat, click "Sign up," create an account, and use the "Send a message" field to interact with ChatGPT. (For $20 a month, you can upgrade to ChatGPT Plus, which offers access to GPT-4, larger data sets, priority access, and faster response times.)

Effective ChatGPT prompt engineering for translators

- Place instructions at the beginning of the prompt and use ### or """ to separate the instructions and context, as recommended by OpenAI.

Prompt example:

Fix all OCR errors in the following text and list any words you have corrected.

- Create simple and clear prompts, and avoid jargon. Keep in mind that ChatGPT tends to ignore anything it doesn’t understand.

- Set the context by starting your prompt with a brief statement that clearly outlines what you’re doing or aiming to achieve.

Prompt example:

I'm a technical translator. I'm translating a text from English into German in the context of artificial intelligence and machine learning for the first time. Can you provide a list of 20 frequent German technical terms in this field and a definition for each term?

- Start a new chat as often as necessary. The model will retain information within a single chat and will not propagate your input across different chats.

 

- Add one or more examples if ChatGPT doesn’t understand the instruction (but avoid overloading the engine with too many examples).

Prompt example:

Change all numbers in UK format to numbers in German format based on the following example, but do not list the numbers in UK format.
0.2291 = 0,2291

- Break down complicated instructions into multiple parts so as not to weigh down the engine with too many instructions all at once.

- Use the "Regenerate response" button, where appropriate, for more options, a differently worded response, etc.

- Whenever possible, avoid negative instructions. It is often more effective to instruct ChatGPT on what to do (rather than what not to do).



What ChatGPT is NOT

ChatGPT is not a search engine, knowledge base or encyclopedia. Keep in mind that ChatGPT will confidently provide responses, even when they are blatantly wrong, and it cannot distinguish between “facts” and made-up information. While ChatGPT typically produces coherent text, its output can be nonsensical, inconsistent, or inaccurate upon closer examination.

ChatGPT is also not a translation engine. Many texts are not suitable for machine translation; however, for any texts that are clearly suitable for machine translation, it is advisable to stick to traditional MT tools, such as DeepL, as they are generally more reliable and faster.

 

However, ChatGPT excels at carrying out certain administrative tasks on your behalf.


ChatGPT as your virtual assistant

Remember the 80/20 rule which I discussed in a previous blog post? ChatGPT can sometimes complete 80 per cent of a task, leaving just 20 per cent for you to do. Often you just need to customise its output. For example, you can seek quick answers to general IT-related queries from ChatGPT or instruct it to draft (possibly awkward) emails for you.

Prompt example:

Please write a 500-word email to a client, in which you point out that the text which I've been sent for revision will need to be retranslated as it appears that it's been machine-translated and is therefore unusable. Use a polite and formal tone. Point out that while machine translation has improved in recent years, it is still unable to match the level of precision and nuance achieved by humans translators.

 

 

ChatGPT also excels at extracting data or terminology from sources that have been chosen for use in translations. This feature can save you from the tedious task of sifting through lengthy documents and manually compiling glossaries yourself. Here the "Act as..." ChatGPT hack can prove useful.

Prompt example:

I want you to act as a text-based Excel sheet. Please extract 10 keywords from the English text below and list them in table format, with a column with the English terminology on the left and a column with the corresponding in-context German translations of those keywords on the right.

 

 

The quality of the ChatGPT output varies: sometimes it's good, sometimes it’s not. Therefore any ChatGPT response should always be carefully reviewed before it is used. It is therefore advisable to rely on conventional approaches for terminology research in translation, such as consulting specialised dictionaries, blogs or other online resources, using Google search operators, or discussing terminology with humans.


This blog article provides an introduction to ChatGPT for translators and includes tips for effective ChatGPT prompt engineering, along with a short video in which I share personal thoughts about ChatGPT.

Vlog #2: Does machine translation impact human translators?

As ChatGPT has just arrived on the scene, I've chosen artificial intelligence and machine translation as a topic to talk about in another vlog.

 

Does the widespread use of machine translation these days impact the work of human translators?

Is artificial intelligence a threat to the translation profession?

And do human translators use machine translation and tools like ChatGPT in their work?

 

Watch my latest video post, in which I'm sharing some thoughts on these and other questions:

Here are the blog posts which I'm referring to in the vlog:

4 April 2022: My 10 translation workflow stages

17 February 2022: The translation workflow stage that should never be omitted

1 February 2020: Why translators don’t fear the machines

12 December 2020:  The hallmarks of a good translator

30 April 2019: Human translation simply explained 

27 January 2018:  Machine translation in human translation workflows

12 September 2017: DeepL: Tool or threat for translators? 

 

Thanks so much for watching my second vlog, and do stay tuned!

Vlog #1: Hello and thank you!

Welcome to my first vlog!

Video blogging (or vlogging) has been on the rise in recent years, so I've decided to jump on this bandwagon, too.

I have therefore just had a go at creating my first vlog, which you can watch here: 

Here are the posts about Google search operators and AutoHotkey, which I've mentioned in my vlog:

3 March 2021:  Must-know Google search operators for translators (part 1)

18 March 2021:  Must-know Google search operators for translators (part 2)

 

9 November 2022: Getting started with AutoHotkey

1 October 2020: Working more efficiently with AutoHotkey (part 1) 

25 July 2021: Working more efficiently with AutoHotkey (part 2) 

15 January 2022: Working more efficiently with AutoHotkey (part 3)

Note that I'm a complete newbie to vlogging, video creation and video editing, so please bear with me as I'm teaching myself the ropes and learning how to do this.

Do stay tuned for further blogs and vlogs!

Christmas 2022 donation to Sustrans

In the same minimalist vein as in previous years I have once again, instead of spending money on Christmas cards, made a Christmas donation to a charity this year.

 

 

As a regular user (and lover) of the Bristol-to-Bath cycle path (as mentioned in previous blog entries here or here), I’ve donated to Sustrans, a British charity and the custodian of the National Cycle Network. Sustrans, which is headquartered in Bristol, provides and cares for a UK-wide network of signed cycling and walking paths, connecting cities, towns and countryside, where everyone can thrive.

 

The Bristol-to-Bath cycle path was constructed on the track bed of the former Midland Railway, which closed for passenger traffic at the end of the 1960s. Between 1979 and 1986, the railway line was converted into the railway path. The delights encountered on the path include public artworks, disused railway station buildings, spectacular views of the countryside and attractive resting places.

 

The Bristol-to-Bath cycle path was constructed
on the track bed of the former Midland Railway

Sustrans provides and cares for a UK-wide network of signed
cycling and walking paths, connecting cities, towns and countryside

Donations to Sustrans, which will contribute towards creating safe and accessible paths and spaces for cyclists, walkers and wheelchair users, can be made here.

 

May I take this opportunity to wish all followers of my blog a serene, peaceful Christmas and a productive, healthy new year!