Lesson 1: Human or Automatic Translation, or both?
Learning Objectives
- Evaluate the advantages and limitations of different translation approaches: automatic (NMT and LLM), human, and hybrid.
- Understand professional post-editing methodologies and error types specific to automatic translations.
- Analyze how translation approach choices impact localization quality and search engine optimization.
When choosing a translator, what are your options?
Sometimes localization is treated as an automatic approval activity, which is unfortunate because word-for-word translation can negatively affect SEO rankings, uptake, loyalty, and the like. This brings us to an important question: when choosing a translator, what are your options?
You can opt for automatic translation or human translation, or a hybrid approach that combines both. Within automatic translation, you can choose between Neural Machine Translation (NMT—systems like Google Translate and DeepL) and translation generated by LLMs (in its most basic form, prompting a system like ChatGPT to generate translations).
If your approach combines automatic translation with human translation, professional translators can help you prepare effective prompts when working with LLMs. For both types of automatic translation, professionals should review the translation outputs to ensure they convey your intended messages.
Post-Editing Automatic Translation
When post-editing automatic translation, professional evaluators verify the output following methodologies such as full post-editing and light post-editing. Common errors they check for include misinterpretations, bias, and cultural appropriateness.
Methodologies for verifying NMT are described in this guide from TAUS. Using LLMs for automatic translation introduces error types not seen with NMT, such as hallucinations.
⚠️ Important Consideration
When using translations produced by NMT or LLMs that have not been post-edited, keep in mind that these systems produce convincing content that makes errors harder to detect. This is why working with professional translators is recommended.
Questions for Consideration
As you reflect on the choice between human and automatic translation, consider the following questions:
When is automatic translation without human post editing appropriate?
Given the limitations of word-for-word translation, in what situations is automatic translation with no professional post-editing appropriate, if any? What types of content are appropriate for automatic translation? What types of content must be professionally translated? Take into consideration factors such as audience, intent, the consequences of translation errors, and the impact of word-for-word translation on SEO, adoption and loyalty.
NMT versus LLMs
Neural machine translation (NMT) works differently than LLM-generated translation. NMT is trained on word-for-word translations to perform translation, where LLMs are trained on a broader range of content. For LLMs, translation is an emergent ability, but with greater flexibility comes a new type of error: hallucinations. In what sitations would you select NMT over translation produced by LLMs, and vice versa? When responding, take into account factors like the content type, the need for consistent terminology use, adaptations that need to be made to tone, and the risks associated with errors that are common in each type of automatic translation.
The Value of Professional Post-Editing
Hybrid approaches in which human translators post-edit automatic translation output are very common in the language industry. Still, with so many readily available automatic translation tools available, skipping the post-editing can be tempting. How would you explain why a hybrid approach with human post-editors is valuable to a person who wants to use automatic translation "as is" for high impact content?
Looking Ahead
In the next lesson, we'll look at Translation Labels being developed by ASTM F43 to help end users understand the confidence/risk level of the content they're reading. Following that, we offer some guidance on how to choose a qualified human translator or translation agency to perform translation.
📚 Expanding Your Localization Toolkit
While we've highlighted a first major decision to be made—human or automatic translation—forthcoming LocEssentials courses will dive deeper into the technical aspects of automatic translation systems:
- Selecting a Neural Machine Translation Engine: Learn how to evaluate and choose NMT systems, and understand automatic quality estimation and additional metrics.
- Selecting an LLM for Translation: Explore the emerging field of LLM-generated translation, understand capabilities and limitations, and learn evaluation frameworks for selecting among LLMs.