The recent acceleration in the refinement and reliability of machine translation is leading some observers to speculate that machines will practically eliminate the need for expensive human translation, including in the corporate market, wiping out tens of thousands of jobs in product and service localization, publishing, marketing and a range of other fields – even as demand for translation explodes.
However, this is a false extrapolation from current success. While high expectations for the latest improvements are fully justified, the idea that machines will destroy language services as an industry and drastically reduce the need for translation and globalization teams does not hold up. There are several reasons:
- The bar for successfully translating the language used in companies is substantially higher than for consumer applications.
- Even within the agency, the bar is being raised ever higher, for these reasons:
- More languages and dialects have to be handled.
- More specialized vertical markets, such as the legal system and healthcare, have to be served.
- There is a growing need for more types of documents on special horizontal media, for example documents describing decisions, requirements and systems.
- Translating the functional aspects of a product (for example menus and documentation) is a specialized practice, one to which consumer approaches to translation are not easily applied.
- Current translation technologies will not keep improving indefinitely at this pace. The biggest recent advances came from using huge amounts of already translated material to train translation models that can translate similar content in future. Many companies are much more specific in terms of context and discipline, and also have smaller volumes of already translated data for these narrow contexts. These are technical challenges that AI algorithms are only now beginning to address, and the transfer of new technologies – if new research and development work is not required – must reach the next level in promoting commercial value.
- The number of languages that can be translated profitably grows with the new low-cost, AI-supported approach. So although the cost of translating priority languages may fall, the volume of emerging priority languages keeps growing. These less-translated languages have less data, which makes the automation problem harder.
The number of languages and language pairs now handled by the most advanced translation platforms is only a small fraction of the languages spoken even in developed countries. Yet translating content into more than 40 languages, supported by the largest language service providers (LSPs) and enterprise software vendors, has until now been difficult or impossible to justify on cost: for most companies, the cost and time needed to add just one new language for a product have been measured in millions of dollars and a good many years.
These barriers are about to be broken down by combining the level of efficiency enabled by cloud-based platforms with the productivity gains of translators who use machine translation.
While some of the “long-tail” language markets (languages with no recorded content) are growing fast, only a few will ever have enough revenue to justify the cost of deploying the current non-AI technology model locally. However, with platform economics, AI-driven efficiency and lower costs, translation agencies will be able to recover the money and time invested in adding new languages while keeping translation prices affordable for a much larger number of customers. The economics will also allow agencies to significantly increase the volume of translation they can handle, helping to sustain revenues and margins even as prices fall.