In today’s fast-paced world, speed has become one of the most sought-after service qualities, and the translation industry is no exception. A quick Google search for “fast translation services” reveals many promotional pages from companies offering such services.
Sadly, while these pages contain plenty of marketing buzzwords, they do not explain how these companies manage to provide fast translations nor disclose the potential risks involved with these services.
As such, our aim with this article is to provide a comprehensive analysis of the different fast translation methods and highlight their potential downsides.
The standard translation process
To effectively discuss fast translation, we first must understand the duration required for a standard translation process.
The conventional translation process is divided into three stages: Translation, Editing (also known as Review), and Proofreading (TEP). Different professionals handle these stages:
- The first translator translates the text from the source to the target language.
- The second translator revises this translation to enhance accuracy and professionalism.
- The third performs thorough proofreading to improve overall fluency and readability.
So, how long does a standard translation process take? Let’s consider an example. Suppose you have a document with 12,000 words:
- A translator can handle 3,000 words per day.
- An editor can revise 6,000 words per day.
- A proofreader can proofread 12,000 words per day.
Following the standard process, translating your document would take 7 days: 4 days for translation, 2 days for editing, and 1 day for proofreading.
Let’s use this estimate to explore various methods to increase the translation speed.Reducing translation stages
Skipping proofreading
Eliminating certain stages in the conventional workflow is the first conceivable method of expediting the translation process. A good candidate for elimination is the proofreading stage.
In our example, this method saves one day, which is about 14% of the total required time.
What happens if we skip the proofreading stage? Primarily, proofreading enhances the overall style and readability of the text. Therefore, two conditions must be considered to manage quality risks when skipping this stage:
- The approach is suitable for technical content such as UI strings, user guides, and online help. However, it falls short for content that needs to evoke emotions or address cultural differences, such as in marketing, literary, or cultural texts.
- The translators and editors involved must have strong linguistic skills and be capable of partially compensating for the absence of proofreading in their own work.
Skipping both editing and proofreading
Could we then consider omitting the editing stage as well? It is advised against doing so due to significant quality risks. Let’s analyze this with data:
A common benchmark for acceptable quality in the translation industry is that the overall translation error rate (E) should be less than 0.5%, meaning the total error score should not exceed 5 points per 1,000 words translated.
📃 Note that we discuss error points here, not the number of errors. In this context, each error is assigned a weighted score ranging from 1 to 10 points according to severity.
Consider achieving this standard through just translation and editing. To simplify our discussion, we’ll use the following abbreviations for error rates:
- Error rate of the translation stage: Et
- Error rate of the editing stage: Ee
According to the probability theory in mathematics, they must satisfy this equation: Et × Ee < 0.5%.
To simplify it, we can set Et = Ee.
Then, we will get Et = Ee ≈ 7% (7% × 7% = 0.49%, very close to 0.5%)
This suggests that if translators and editors maintain a 7% error rate or lower, their combined efforts can reduce the total error rate to 0.5%.
Theoretically, we could set an even more conservative error rate, i.e., 2%. However, removing editing and proofreading and relying solely on the translator would make it extremely difficult to reduce the error rate from around 2% to 0.5%.
Improving quality is not linear; the closer you get to zero defects, the harder it becomes. It is very challenging for translators alone to reduce the error rate to a fourth of its original level without the editing stage. Therefore, skipping both editing and proofreading is not a viable option.
Another option is to skip editing while retaining proofreading. However, since proofreading focuses mainly on fluency and readability, translation errors will likely go unnoticed, resulting in a quality level comparable to skipping both editing and proofreading.
Incorporation of translation technologies
Translation technologies like translation memories (TM) and machine translation (MT) significantly improve translation efficiency.
Translation memory
Since its introduction, translation memory technology has been crucial in boosting the efficiency of translation processes, especially in technical content translation. This technology focuses on reusing translations and handling repeated and highly similar text segments through technological means to accelerate translators’ workflow and reduce costs.
We can examine the impact of translation memory by considering how it could impact the text blow. This text has been previously translated into the target language:
The system uses advanced algorithms to optimize data processing. Users can configure settings to customize performance according to their needs. Regular updates are essential to ensure the system runs smoothly.
Here is the updated version of the text:
The system uses advanced algorithms to optimize data processing. Users can configure settings to customize performance according to their requirements. Additionally, periodic updates are crucial to maintain system stability and introduce new features. A dedicated support team is available to assist with any issues.
When the initial text is translated, the content is stored in a database as sentence pairs, known as a translation memory. When new translations begin, these sentences can be provided to the linguists.
| ID | Source | Target |
|---|---|---|
| 1 | The system uses advanced algorithms to optimize data processing. | 该系统使用先进的算法来优化数据处理。 |
| 2 | Users can configure settings to customize performance according to their needs. | 用户可以配置设置，根据自己的需要定制性能。 |
| 3 | Regular updates are essential to ensure the system runs smoothly. | 定期更新对于确保系统平稳运行至关重要。 |
When updating the content, the new text is compared with the stored translations in the database, and discounts are applied based on the degree of match:
- The first sentence remains unchanged, so the existing translation can be directly used with minimal checking or slight modifications. Such sentences are called “exact matches” and receive significant discounts, e.g., 80%.
- The second sentence has minor updates compared to the original. Translators can modify the existing translation accordingly. These are called “fuzzy matches” and receive moderate discounts, e.g., 50%.
- The third sentence has significant updates, and the fourth sentence is entirely new; both are considered “no matches” and do not receive a discount.
After applying these TM discounts, the word count is known as the “weighted word count.” Both translation agencies and clients can use the weighted word count to estimate workloads and set pricing.
Since the weighted word count effectively reduces the number of words, it can significantly decrease the translation time. The specific impact depends on the extent of TM application and the similarity of the content, with results varying each time.
For instance, if the original word count of 12,000 is reduced to weighted 9,000 words, the translation time could be cut by a quarter.
Given the maturity of translation memory technology, it introduces very low additional risks to quality. It has been widely adopted across various translation fields and integrated into many popular computer-aided translation tools.
Machine translation technology
Over the past decade, as neural machine translation (NMT) technology has matured, machine translation has widely entered the field of professional translation services. This has led to the emergence of businesses offering machine translation services, significantly increasing translation speeds.
Machine translation only
A document can be translated almost instantly using the API of a machine translation engine. If the output from the machine translation is delivered to end users directly without any human intervention, this method is known as machine translation only.
To those unfamiliar with the target language, translations produced this way look like those from human translators. However, machine translation only poses many risks and issues and should not be used as the final output. Many translation errors that have become the subject of jokes are typically the result of this approach.
If you seek fast translation services, it is crucial to ascertain whether the provider offers machine translation-only results. Some service providers or individual translators deliver machine translation-only translations under the guise of fast translation. When clients do not verify or cannot verify the translation’s quality, they can be easily deceived.
Offering machine translation only has also proven unfeasible for translation service providers. A few years ago, a company attempted to provide ultra-fast translation services to customers, explicitly stating in their service terms that the translations involved only machine translation without human translators.
Shortly after launching the services, they received a plethora of customer complaints. Customers did not understand the limitations of machine translation and expected the quality of human translation, forcing the company to pivot, rebuild its management team, and abandon the machine translation-only service model.
Machine translation and human editing
While relying solely on machine translation is inadequate, combining the strengths of both machine and human efforts can enhance efficiency through a workflow known as machine translation post-editing (MTPE).
The process typically involves machine translation (MT) followed by human editing (PE). Some workflows also include a quality assurance (QA) step, which operates as follows:
- The original text is translated into the target language via a machine translation engine.
- Human editors then edit the machine-generated translation to correct errors and refine the text.
- Quality assurance personnel review the final text and assign a quality score. If the score is low, revisions are necessary.
How much time does this process save?
- Machine translation consumes minimal time and is not a significant factor.
- Since machine translation quality typically falls short of professional human translation standards, editors require more time to make corrections. We can assume an editing efficiency of 5,000 words per day.
- Quality assurance generally involves spot checks, estimated to take about 0.2-0.4 days, assumed here as 0.3 days.
For 12,000 words, the required time is 2.4 + 0.3 = 2.7 days.
There is a proofreading stage in the TEP process, which could also be included in MTPE. However, since proofreading was not included in the earlier calculation, the original translation time should be considered 6 days (excluding proofreading from TEP). Therefore, using MTPE saves approximately 3.3 days compared to the original process—about 55% of the total time.
The quality of the initial machine translation is crucial in the MTPE approach. If the quality is poor, post-editing efficiency will be significantly reduced. Many editors complain that editing a machine-translated text is more cumbersome than translating from scratch.
The engines specially trained with tailored corpora significantly outperform generic engines. Therefore, it is recommended that MT engine training be organized before adopting the MTPE workflow.
Although MTPE has been successfully applied in many scenarios, the quality of MTPE translations is generally inferior to that of purely human translation processes.
Light post-editing
MTPE can be further categorized into two types:
- Full post-editing involves thoroughly checking and correcting all issues.
- Light post-editing (Light PE) focuses only on obvious errors, ignoring minor issues, thus requiring less time.
Let’s assume Light PE can process 10,000 words per day. If QA is not considered, for the same 12,000 words, the translation workflow would be shortened to 1.2 days, saving 4.8 days, or 80% of the time.
However, since Light PE only addresses obvious errors, some may remain, potentially decreasing quality. In scenarios where high quality is required, Light PE is not suitable.
Adjusting translation management
Innovations in translation management can also speed up translation.
Concurrent translation and editing
Concurrent translation and editing involve translating and editing simultaneously to reduce overall translation time.
Traditionally, editing could only begin once the entire translation was complete. However, with this method, the editing process can be completed shortly after the translation is finished by allowing editors to start reviewing text sections as soon as they are translated.
The concurrent translation and editing process without proofreading should take the following amount of time:
- Four days for a 12,000-word translation.
- An additional half a day to complete the editing process.
The total time required is 4.5 days. Compared to the original timeline of 6 days, it saves 25% (or 1.5 days).
The risk with this approach is that editors may not approach their tasks holistically. Instead, in a rush to meet tight deadlines, they may focus excessively on specific content sections instead of the whole, thus leading to a lack of translation coherence. It’s a trade-off you’d have to consider carefully.
Additionally, concurrent translation and editing necessitate an online translation platform capable of supporting this feature. The platform must enable translators to send translated segments to the editors in real time.
Crowdsourced translation
Crowdsourced translation is a good option for large volumes of content that are not highly demanding in quality and require rapid completion.
The essence of crowdsourcing is to use a large number of translators (not necessarily professional ones) to achieve the goal of rapid translation.
Effective crowdsourcing requires strong organizational and management skills and a dedicated platform. For more details on crowdsourced translation, please refer to “Beginner’s Guide to Crowdsourced Translation.”
If we use crowdsourced translation for the example of 12,000 words, the time will be spent on two items:
- Setup time: Including platform setup and team preparation. Let’s take 1.5 days for this case.
- Translation time: This translation could be completed in just 1 day.
It’s 2.5 days in total, saving 3.5 days, or 58%, compared to the original 6 days (as with MTPE; proofreading will not be considered here).
Multiple professional translators
Given the unstable quality of crowdsourced translation, employing multiple professional translators for one translation project is certainly feasible. However, additional considerations like coordination and style consistency must be made when splitting the project among many translators.
For the 12,000 words:
- Assign two translators to reduce translation time from 4 days to 2 days.
- Still assign only one editor to ensure style consistency, taking 2 days for editing.
The original timeframe was 4 + 2 = 6 days; splitting it among two translators results in a timeframe of 2 + 2 = 4 days, saving 2 days, or 1/3 of the time (as with MTPE, proofreading will not be considered here).
This arrangement considers style and consistency. It is relatively conservative but ensures lower quality risks, and it is highly advisable. Theoretically, more translators could be employed.
For example, by assigning four translators to the project, they could complete the translation in one day; two editors could finish editing in one day, saving four days or two-thirds of the time.
However, such an arrangement is inappropriate as it significantly increases risks, as described in “The Mythical Man-Month,” which states that simply expanding the workforce does not proportionally decrease time.
Early involvement
In some translation projects, it has been observed that inviting the translation team to participate earlier can also “increase the translation speed,” achieving significant effects with minimal extra input. Consider the following example:
A translation company received a client’s request to translate 100,000 words within one month with high-quality standards. After discussions with the client, the translation company decided to use multiple translators combined with machine translation followed by editing. This approach barely met the deadline and posed certain quality risks.
In this example, the content needing translation was written by technical writers and only sent to the translation company after completion. However, if the client had involved the translation team earlier, the situation might have looked different:
- As soon as the technical writers completed a portion of the content (e.g., 20,000 words), it could have been sent to the translation company to start it two months earlier.
- As the writers continued, reaching another milestone (e.g., 15,000 words), the content could be sent for translation. The translation company could then update the content using translation memory, ensuring only new content was accounted for in the workload.
- Once the technical writing was completely finished, the translation company could complete the work earlier, with half a month being sufficient in this example.
With these adjustments, the total translation time was 2.5 months for the 100,000 words, providing ample time and better quality control. The time from the completion of technical content writing to the translation was reduced from one month to half a month, saving 50% of the time.
Since technical writers might change previously completed content, the overall translation workload generally increases slightly. Typically, content changes are less than 10%, and many modifications involve changing only a word or a few words in a sentence. Applying a translation memory system would help reduce the translation workload of these changes. Thus, the total additional word count remains below 5%.
So, less than 5% extra effort yields better translation results and faster translation speeds. This method has already been accepted by many large companies globally.
Moreover, clients should also invite translation companies to participate early in the international design and development stages of software products. This speeds up localization, saves costs, and ensures better quality control, achieving multiple benefits.
Key takeaways
Through the descriptions above, you now have an understanding of various methods to achieve rapid translation. Refer to the table below for a quick overview of each method discussed:
| Method | Time saved | Quality Risk | Use Case | Prerequisites |
|---|---|---|---|---|
| Skip proofreading | ≈14% | Medium-low | Technical content translation | Strong professional skills in translators |
| Apply translation memory | ≈25% | Low | High repetition in translation content | Previous translations stored in the translation memory |
| Full MTPE | ≈55% | Medium-low | Good results from machine translation | Specially trained machine translation engine |
| Light MTPE | ≈80% | High | Content with lower quality requirements | Specially trained machine translation engine |
| Concurrent translation and editing | ≈25% | Medium | Lower overall coherence requirements in translations | Translation tools support concurrent translation and review features |
| Crowdsourced translation | ≈58% | High | High-speed but lower quality requirements | Crowdsourcing translation management capabilities and platform tools |
| Multiple professional translators | ≈33.3% | Medium-low | Suitable for most translation scenarios | Multiple professional translators, capable of coordination |
| Early involvement | ≈50% | None | Content in development | Allows early involvement of the translation team |
Keep in mind that new methods may emerge in practice. As such, testing them carefully before implementation is advisable to minimize risks.
If you need rapid translation services, we recommend discussing detailed and viable solutions with your translation service providers or industry experts.
This ensures that while meeting the rapid translation needs, quality risks are effectively managed, considering costs and feasibility as well.
For more questions about rapid translation, feel free to contact us.