Companies Are Starting to Launch GPT-4-Powered Machine Translation Services

Newswire GPT4 Machine Translation

Canadian newswire service TheNewswire (TNW) announced on April 18, 2023, that it has integrated GPT-4 into its workflow, allowing for English-Quebecois French machine translation of press releases.

In a press release on Yahoo!Finance, the company called the integration “unprecedented” in the newswire industry, and quoted TheNewswire CEO Pat Beechinor as explaining that the technology will help introduce publicly traded companies to “massive new investment audiences.”

The Calgary, Alberta-headquartered company also emphasized that “over 7.5 million Canadian Francophones” will also now be able to read press releases in their preferred language. 

The company cites one of its managers, Dominic Gray, who oversees language integration at TNW, as being impressed with GPT-4’s translation accuracy.

Indeed, at least for high-resource language pairs, the quality of ChatGPT-generated translation has piqued a flurry of both excitement and concern for the language industry.

And researchers at a number of institutions have churned out papers on further improving ChatGPT’s machine translation (MT), whether by using more effective prompts or by adjusting specific parameters, such as temperature.

TNW has not provided details about how the integration will function. It is unclear, for instance, whether the service will be provided as a given, or only at a client’s request, and what quality assurance, if any, TNW will include. 

GPT-4’s fairly restricted API, compared to more established MT platforms, makes it all the more remarkable that given the choice, TNW decided to use a model that was originally designed not for MT specifically. Furthermore, GPT-4’s latency and cost are both much higher than dedicated machine translation providers, suggesting that TNW made the move as much for its PR value as for its technical suitability. 

In addition to MT, TNW plans to use GPT-4 to generate headlines on social media platforms, including Facebook, Twitter, and LinkedIn.