I often get asked by my friends, especially the more techie friends, if I fear for the future of my profession with the remarkable advances being achieved in machine translation. I promptly reply that machines still have a long way to go before they can replace me! These computer experts would then look at me with a sardonic half-smile, or that condescending nod, as they think to themselves “she has no idea what’s in store for her; better off leaving her in her blissful ignorance”. Do they know something I don’t? Have I been so caught up in my translations that I’ve failed to notice my approaching executioner? So I thought it was high time do a bit of research into the field of machine translation.
For those not versed in the argot, it is important to point out the difference between machine translation (MT) and computer-assisted translation (CAT) tools. The latter is software to support translators during the translation process allowing them to edit, create, store and manage a translation and create translation memories. I cannot now imagine working without a CAT tool. MT, on the other hand, is the automatic translation by computer software. Here I will give a short account of the history of MT and outline its utility in the world of translation in 2021.
Some years ago, John Hutchins briefly summarised “The history of machine translation in a nutshell”, available here, broken down into the following periods.
Early days
Did you know that “it is possible to trace ideas about mechanizing translation processes back to the seventeenth century” although realistic possibilities came only in the 20th century. “In the 1930s, a French-Armenian Georges Artsrouni and a Russian Petr Troyanskii applied for patents for ‘translating machines’. Of the two, Troyanskii’s was the more significant, proposing not only a method for an automatic bilingual dictionary, but also a scheme for coding interlingual grammatical roles (based on Esperanto) and an outline of how analysis and synthesis might work. However, Troyanskii’s ideas were not known about until the end of the 1950s.”
1947-1966
A number of factors were crucial to the stimulation of interest in researching machine translation: the development of the computer, wartime successes in code breaking, the developments by Claude Shannon in information theory and the universal principles underlying natural languages. Within a few years, research was being conducted at many US universities culminating in 1954 with “the first public demonstration of the feasibility of machine translation was given (a collaboration by IBM and Georgetown University). See clip from 1954 here. Although using a very restricted vocabulary and grammar it was sufficiently impressive to stimulate massive funding of MT in the United States and to inspire the establishment of MT projects throughout the world”.
These early systems consisted of a “direct” dictionary-based approach, matching the source language to the target language word for word, with some rules for correct word order. It did not take long to recognise the limitations of this system and the interest in finding solutions to overcoming these. After a few optimistic years, disillusion grew as researchers encountered semantic barriers for which they saw no simple solutions. In 1964, the US government set up the Automatic Language Processing Advisory Committee (ALPAC) to assess progress. This committee issued a damning report in 1966 that MT was slower, less accurate and twice as expensive as human translation and that “there is no immediate or predictable prospect of useful machine translation.” It saw no need for further investment in MT research.
1966-1980
The ALPAC report brought an end to MT research in the United States for over a decade and also had a major impact elsewhere in the Soviet Union and Europe. Research did continue in Canada, France and Germany and within a few years, for example, the Systran system was installed for use by the US Air Force (1970) and by the Commission of the European Communities to translate the rapidly growing volumes of documentations requiring translation (1976).
Whereas in the previous decade in the US and the Soviet Union, MT activity had concentrated on Russian-English and English-Russian translation of scientific and technical documents. From the mid-1970s, the demand for MT came from different sources with different needs and different languages. The administrative and commercial demands of multilingual communities and multinational trade stimulated the demand for translation in Europe, Canada and Japan beyond the capacity of the traditional translation services. The demand was now for cost-effective machine-aided translation systems that could deal with commercial and technical documentation in the principal languages of international commerce.
1980s
The 1980s witnessed the emergence of a wide variety of MT system types from a widening number of countries. Apart from Systran, there was Logos (German-English and English-French); the internally developed systems at the Pan American Health Organization (Spanish-English and English-Spanish); the Metal system (German-English); and major systems for English-Japanese and Japanese-English translation from Japanese computer companies.
Throughout the 1980s research on more advanced methods and techniques continued. For most of the decade, the dominant strategy was ‘indirect’ translation via intermediary representations, sometimes interlingual in nature, involving semantic as well as morphological and syntactic analysis and sometimes non-linguistic ‘knowledge bases’.
The most notable projects of the period were the GETA-Ariane (Grenoble), SUSY (Saarbrücken), Mu (Kyoto), DLT (Utrecht), Rosetta (Eindhoven), the knowledge-based project at Carnegie-Mellon University (Pittsburgh), and two international multilingual projects: Eurotra, supported by the European Communities, and the Japanese CICC project with participants in China, Indonesia and Thailand.
1990s
This marked a major turning point. Firstly, a group from IBM published the results of experiments on a system based purely on statistical methods, and secondly, some Japanese groups began to use methods based on corpora of translation examples. Both approaches differed from earlier rule-based methods in the exploitation of large text corpora.
Also, MT on the web starts with Systran offering free translation of small texts (1996), followed by AltaVista Babelfish, which received up to 500,000 requests a day (1997).
Now in 2021
With the exponential rise of international communications, it is clear that human translators alone are unable to meet the massive demand for cheap, fast, large-scale information exchange across languages. MT has a clear advantage over traditional translation in terms of speed, scalability and cost. So why are we not all using machines to translate our documents? The answer is simple. Quality! MT is still unfit for use in all cases. It works best on technical texts with a limited or repetitive vocabulary. Raw MT (without any review by a human linguist) is not perfect but can serve a purpose for translating user-generated content, internal documentation, or in cases where fast translations are needed but accuracy is not important.
Machine translation post-editing (MTPE) combines MT with human translation giving you the speed and ability of MT engines to quickly handle large volumes of text, with the skills of trained linguists. Post-editing, a relatively new phenomenon, is the process of reviewing and adapting raw MT output to improve the quality of the translation. But MTPE too has its limitations. The poorer the quality of the raw output, the greater the post-editing required. This can lead to frustrations for both the post-editor, with more work and comparatively low payment, and for the client, as the cost saving may not be as good as expected.
This decade, therefore, we are seeing MT moving away from this high speed/poor quality output toward offering a reasonable alternative for translating low visibility content. The race is now on to achieve a competitive edge in the quality MT output with various solutions on the table.
Hybrid systems, as the name suggests, combines MT with human linguists. Neural machine translation (NMT) is an approach built on deep neural networks. Compared to previous generations, NMT generates outputs which tend to be more fluent and grammatically accurate as it evaluates fluency for the entire sentence rather than for a couple of words. Some are generic while others are trained with specific data, resulting in more accurate MT output depending on the domain. This has greatly improved the quality of the translations.
Take home
MT is extremely useful for the translation of large volumes, repetitive and non-creative texts. Producing low-cost practically instantaneous translations continues, however, to adversely affect quality. Juggling cost, quality and time-to-market continues. Human translation and MTPE remain as the gold standard for translations requiring perfect quality. For medical, pharmaceutical and regulatory affairs documents in which quality is paramount, the human translator is still necessary.
So, to my dear techie friends, the machine is not going to make me redundant any time soon!
The human linguist has still a significant role to play in the translation industry, although this is everchanging in MT-driven translation workflows.