The difference between AI translation and machine translation
Generative artificial intelligence (AI) and neural machine translation (NMT) are two technologies that have revolutionised the field of machine translation. But how do they actually work, and do they produce equally good results? In this blog post, we’ll explain the basics of these technologies and describe how we at Textforum combine them to improve machine-translated texts in three clear ways.
How does NMT work?
Neural machine translation uses neural networks to translate texts from one language into another. Instead of just translating individual words and stringing them together, NMT can translate entire sentences and understand context. The resulting translations are more coherent and sound more natural. Today, NMT is the most commonly used machine-translation method.
How does generative AI work?
Generative AI is a different technology. It uses advanced machine-learning algorithms, like generative adversarial networks (GAN) and transformer models. GAN let two neural networks ‘compete’ with each other: one creates content, the other assesses the content. This allows generative AI to create authentic texts from scratch.
GPT models are a special kind of generative AI. These models are trained by feeding them large amounts of data, which teaches them to generate coherent and contextually correct content. Because they understand context and the way things are linked, they can improve machine translation and produce translations that sound more human.
What does that look like in practice?
At Textforum, we can directly integrate both NMT and generative AI into our translation systems. The first step in every machine-translation project is to thoroughly prepare all files, to enable the best possible translation. Apart from NMT and generative AI, we also use unique translation memories and term databases we’ve created for our clients, based on previous translations we’ve done for them. These memories take precedence over machine-translation suggestions, because they contain translations and terms the client has already approved. Afterwards, we choose which machine-translation engine to use and translate the text. Sometimes, we ask an AI engine to review the resulting translation sentence by sentence and suggest ways of improving the text. We can either automatically accept these AI suggestions, or bring in a human proofreader to post-edit the result.
What are the advantages of this process?
- Context is truly taken into account: When generative AI is integrated into an NMT system, translations can benefit from the transformer models’ ability to understand and generate sentences that contextually make sense. The resulting translations are more accurate and better integrated into their context. Generative AI is also better at dealing with idioms and cultural expressions, because it has been fed massive amounts of varied language data. The resulting translations are more likely to maintain and reflect the original text’s cultural nuance and expressions.
- The translation is tailored to your target group: By providing the GPT model with certain information (so-called ‘prompts’) about your target group’s way of speaking and its linguistic style, generative AI’s translations tend to be a better fit for the intended audience, which makes communication more effective.
- New terms are quickly picked up: Generative AI can swiftly learn to use new terms and expressions when it is fed updated data. This is particularly useful in sectors in which existing terms are constantly finetuned, or new terms are regularly coined.
To sum things up, both generative AI and NMT can be used to machine-translate texts. By combining the two, however, we harness their unique strengths to produce machine translations that are as fluent and natural-sounding as possible. When you let us handle the entire translation process, we can even make sure the result features your company’s specific terms and expressions.