Choosing Machine Translation: Balancing Quality, Speed, and Cost (China Market Example)

In language services and e-discovery decision-making, there are typically three fundamental business constraints: 1) Quality, 2) Speed, and 3) Cost.

Are you looking for a low-cost solution? A fast turnaround? Or the highest possible quality? Well, choose two.

These three factors naturally conflict with each other, requiring a careful balance when choosing. This article will explore these trade-offs, analyzing the three variables and helping you understand how to choose the best solution to maximize its value. Our first topic: Machine Translation (MT), with a focus on the Chinese market.

Why Choose Machine Translation?

Machine Translation (MT) generally refers to software-driven translation, but there are significant variations in cost, speed, and quality. The "best" option in one scenario is almost certainly not appropriate in another.

For example, let’s consider this request: “I have 550 Russian documents that need to be machine translated.”

The fact that there are 550 Russian documents is useful information, but it only provides limited help in indicating the best solution among the many MT options available. The real key lies in understanding the context and overall objectives. Once the file count is paired with the underlying context and goal, the ideal solution becomes clearer.

Let’s look at three possible scenarios:

• 550 Russian first-level relevant emails related to a tertiary case issue
• 550 Russian Excel and PowerPoint files needed in 48 hours (possibly with hidden content MT’ed if time permits)
• 550 Russian key case documents with highly nuanced language, sent by the General Counsel

Understanding the "why" behind the request helps eliminate options that ultimately won't meet the need. While there are broad categories of MT (text, formatted, post-edited), many options, enhancements, and steps can be combined to meet specific budget, time, or quality requirements.

Beyond helping narrow down the options, it’s TLS’s language consulting that enables us to actively adjust and add options within the overarching solution to better meet the translation goal.

Mapping Out the MT Solution: Quality vs. Speed vs. Cost

Quality

If the goal is to quickly understand the basic content of a set of emails, a simple text MT is a fast and cost-effective way to get the gist. By adding a little more time (and budget), that basic MT output can be improved with case-specific glossaries.

However, understanding content is not just about translating words. For certain file types, a significant portion of the meaning comes from the structure of the document itself, so any solution must account for this. For certain file types like financial data, presentations, etc., the only recommended approach is to translate from the native file format.

Typically, when providing formatted native machine translations, we do not include hidden rows, columns, or sheets. However, if the files are produced or disclosed to opposing counsel in native format, we highly recommend translating hidden content as well. This obviously adds time, but it greatly increases the quality and thoroughness of review before disclosure.

For some situations, high-quality translation is needed, but the budget doesn’t allow for a full human translation workflow. In these cases, post-edited MT can help reviewers spot details that a lower-quality translation might obscure. Post-edited MT comes in light and heavy variants. Light post-editing makes the MT understandable and factually accurate, while heavy post-editing ensures correctness and goes much further in terms of grammar, tone, style, and consistency. Heavily post-edited MT is close to full human translation in quality.

Among all MT options, post-editing is the slowest and most expensive, but it undoubtedly produces the highest quality results.

Speed

Text MT is the fastest, followed by formatted native translations. Formatted native translations with hidden content translated are slower, as there are additional steps for extracting, translating, and reinserting the hidden data. As mentioned earlier, post-edited MT is the slowest because it involves both machine translation and human review.

Knowing that machine translation is urgently needed allows us to adjust the process to meet deadlines. On the other hand, if a project is not time-sensitive, we can make suggestions that focus on quality. Of course, some steps in the workflow can be bypassed, but doing so will reduce quality.

Cost

The cost of each MT option roughly correlates with its speed, with text being the least expensive and post-edited MT being the most expensive.

Note: Costs don’t accumulate just from the machine translation process itself. For example, if a lawyer handles the review (instead of a less costly contract reviewer), there is a strong case for creating all translations as formatted natives. Compared to pure text MT, formatted natives make the review process more efficient due to their legibility.

Here’s a hypothetical cost breakdown:

• 100 hours of lawyer review x ¥3,500/hour = ¥350,000
• If formatted natives save 15 hours of review time...
• 85 hours of lawyer review x ¥3,500/hour = ¥297,500
• By increasing MT spend by ¥6,500, you save ¥52,500 in review costs.

Revisiting Our Three Scenarios…

1. 550 Russian first-level relevant emails related to a tertiary case issue: Text MT
2. 550 Russian Excel and PowerPoint files needed in 48 hours: Formatted native translations (possibly with hidden content MT’ed if time allows)
3. 550 Russian key case documents with highly nuanced language, sent by the General Counsel: High post-edited MT

In these cases, depending on the specific needs, timeframes, and budget constraints, the translation solution will vary. By slightly adjusting any of the variables, the whole solution changes.

1. 550 Russian first-level relevant emails reviewed by a senior associate: Formatted native translations
2. 550 Russian Excel and PowerPoint files needed in 48 hours at the highest quality: Low post-edited MT
3. 550 Russian key case documents sent by a General Counsel unsure about the case and seeking the cheapest option: Formatted native translations

Most cases are not composed of documents that neatly fit into either text MT or formatted native MT categories. A good workflow involves documents being analyzed by a language consultant, segmented, and routed into the MT engines and workflows that finely balance the trade-off triangle’s variables with the broader objective.

◀️：How to Develop an Effective SEO Strategy for the Chinese Market

▶️：3 Translation Methodologies for Life Sciences Documentation