The Potential Benefits and Challenges of Machine Translation

Abstract

Comparison of both machine translation (MT) and human translation is a familiar topic to discuss some roles of AI technologies in the future. Back in the 1950s, computer scientists started solving the problems of a language translator. Since that time, the quality and availability of machine translation have been improved in many countries. Recently, because of advanced translation software and the spread of digital devices, the demand for introducing into workplace or language learning classes is increasing drastically. If MT produced a human-like translation, human translators would be most affected by the improvement of MT. This report introduces how people should or would deal with MT.

Introduction

The idea of inventing MT first appeared in the seventeenth century, but it became a reality only at the end of the twentieth century. MT is fully automated software that can translate source content into target languages. The past ten years since 2018 have been a very fruitful period for MT research and development. He claims that developing the spoken language MT systems (the process by which conversational spoken phrases are promptly translated and spoken in target language) has made great progress in improving speech analysis technology to produce many working systems (2018). The development suggested a new growth era for MT. The success of MT research enhanced the service of the World Wide Web, usually free. Owing to this, a number of people started to use such web to translate foreign words or texts and to make sentences in other languages. However, fully automated translations are far from being perfect. Although MT has become more familiar, Use of MT is limited under some circumstances such as negotiation or conference. This report argues that if MT has the potential to enhance English learners’ performance and will replace human translators or coexist with them.

Literature Review

Applying MT in Education

It is becoming more common to use MT to translate simple words or sentences when people learn a second language. Therefore, some students use electric dictionaries or devices such as iPad. In Japan, most students from elementary schools to universities are required to learn English as a second language. Although Japanese is the only language needed in their daily life, they normally study English “for about 1,000 hours”. Therefore, they tend to use MT to avoid struggling with the second language. Adults also study English to communicate with foreign people in English or to get higher pay. Although the performance of MT is still poor in terms of complex texts, it is a fact that MT can reduce much time and effort. MT could be a tool that enhances the performance of foreign language learners effectively. The reason why Japanese people learn English sounds weak. Gally claims that “children, in particular, would be unconvinced by the assertion that they should study English”. To increase their motivation, some schools allow them to use MT in English learning classes. Ahlan, A. et al (2014) argues that the introduction of MT is seen as “fuel that increases the speed of proper academic activities”. According to Ugwuadu and Jota (2013), with information technology such as web translation software, students performed better in foreign language classes than students who did not use it.

Will MT Be Superior to Human Translators

It is often said that if MT became competitive of human translators, they will struggle to get their jobs. At the moment, there is no possibility that MT will replace human translators because Nirenburg (1989) claims that implicit meanings of texts or slangs are not interpreted and recognized by MT. The complexity of the text increases the likelihood of more errors in MT. Free MT such as Google translation tends to make direct translations, and they confuse people. Khan (2019) points out that MT cannot understand cultures. There are many different cultures which are strong and unique, but MT is not programmed to reflect those culture into translation. Only human translators can comprehend what words such as cultural texts could mean. Luong et al (2015) found out that MT performs particularly poorly on rare words such as dialects. To deal with them, MT needs to be trained frequently. Knowles and Koehn (2017) assert that amounts of training data lead to better performance of MT. By repeating training, MT learns to offer alternatives to users and correct errors by itself. On the other hand, Wu et al (2016) say that some recent papers claim that MT can almost produce human-like translations or MT got closer to the average of human translators. However, “standard for human translators is too high for MT”. Ahrenberg (2017) claims that the limitations of MT are not only temporary but also inherent in the business.

How Can Human Translators Coexist With MT

As globalization grows, it is more likely that more foreign people will visit your country. Thus, opportunities to communicate with them will increase. MT is the most useful tool to deal with such a situation, if you do not speak or know foreign languages. Takeda (2019) say that MT could assist foreigners living or working through all their day such as when they visit hospitals or need to fill in a paper in a foreign language. Meanwhile, human translator or human interpreter cannot always accompany with foreigners. At times, the convenience of MT is superior to not only human translators but human interpreters. Khan (2019) says that many translation companies in UAE apply machine translation to help their human translators, to cut costs and to increase the effectivity of works. However, as mentioned before, the present MT needs to be improved in terms of accuracy and quality of correction. Eksiripong (2019) asserts that if a company relies on the accuracy of MT and introduces it into report or negotiation, directors could not speculate the loss of trust or profit when MT made errors. Denkowski (2019) argues that MT can reduce the amount of work required from human translators, and human translators’ feedback to MT could accelerate the speed of MT improving.

Discussion

According to Gally (2017), the quality of MT between Japanese and English offered for free by Google improved surprisingly and suddenly in late 2016. Until then, the evaluation of MT in Japan was not good, but the significant improvement has encouraged more people to use Google translation. Although Google translation produces some ambiguities, it has become an educational tool for English learning by being applied to some Japanese schools. Therefore, Google translation enhanced a student’s image of MT. The electrical dictionary has been used earlier than MT during English classes, but Japanese students have started using MT more because of its quantity of functions such as producing translated sentences and a wide range of foreign words online. Japanese is the national language of Japan, and English is taught as kyoyo, which can be said as “general knowledge and character development”. In fact, the English level of Japanese people was ranked 35th out of 72 countries. Compared with Asian countries, only Tajikistan and Laos were ranked below Japan. It seems that the Japanese think they do not urge to study English, so Japanese students tend to use MT to reduce self-thinking time and effort. As a result, the use of MT in Japan does not enhance students’ performance.

Meanwhile, in other foreign countries, MT has also been part of foreign language classes and has gained popularity. In 2017, Marang conducted a study to research the frequency of use of Google translation with a group of 25 students of the English department. The result found out that almost all students (90%) have used Google translation. There are 50% of the students who often use it, 30% of students sometimes use it, and 10% of students always use it. The 25 students were labeled as ‘good students’ by relying on their class teachers’ judgement and their test scores. The students stated that google translation can perform translation tasks very quickly. They also said that if users can notice an error made by MT, at least they can get a hint to complete their tasks. Thus, MT can help both students of poor performance and high performance, and foreign language learners who have good language competency can make the most of MT.

Maulidiyah (2018) points out that MT could provide more advantages than disadvantages in terms of education and business, however, when it comes to comparing human translators, MT is far from human-like quality. Contextual errors are major problems of most computer-assisted translation. MT is programmed to translate a target language directly, ignoring the order of context. Compared with an electric dictionary, MT’s capability of translating more than two words and producing sentences in other foreign languages is a strong point. However, that is also a weak element, because the wrong order of translation in context leads users to confuse. The main section which MT is used for is writing, so MT is used to translate works or articles of literature. If many translation errors were made in articles, users can hardly understand the content. MT tends to struggle with words that have different meanings. There are also a number of words that have deeper meanings behind each word and phrase. Deeper meanings can be translated such as cultural meanings or dialects. Unique cultures are associated with words, and they are distinct to some languages and are implicit to other languages. Translating words into other languages requires human translators to be careful of taboo and value differences. Incorrect or unfamiliar words can be aggressive to readers or listeners. MT needs to be improved in order to understand the slight difference in word meanings. Khan (2019) claims that MT should be allowed to have creativity such as AI, then it can produce matching sentences. That means MT needs to get close to a human translator’s flexibility and fluency. At present, MT is not a threat to human translators.

In the future, MT will make significant progress, then people need to know how to deal with MT. In terms of saving time and money, MT has got popularity and trust from its some users. Because of the convenience of MT, more and more companies have applied it to their offices to reduce amounts of workers’ tasks. As globalization keeps expanding, some companies have come to handle more foreign languages. At that time, human translators help directors of companies select the most appropriate MT software. Takeda (2019) says that usage of MT would become common for enterprises that handle technical and business documents and people would edit text translated by MT. Nowadays, MT has evolved from rule-based and statistics-based translation (two kinds of MT which have engines based on the software’s analysis of language and substantial amounts of data) to neural-based translation in which AI is programmed. Neural MT provides high quality and affordable translation, and it has been developed by human translators’ knowledge and advice. Advanced MT has been introduced into human translators’ office by many translation companies in the UAE to reduce their burden. Therefore, human translators have already coexisted with MT, by developing the quality of MT on their own. It seems that MT is a rival or a job stealer for human translators, however, they are working together for their effectivity and evolution.

Conclusion

MT will be surely improved by being programmed AI and be increasingly an important part of education and business. MT will be able to complement not only human translators but human interpreters in certain situations. However, it will be very long until MT replaces them because conversation or article has many complexities. Despite that MT is still incomplete, it has spread all over the world. Using MT can be described as picking up one appropriate meaning from enormous amounts of words, thus MT requires us to have sharp management and judgement skill. At times, users of MT need to be suspicious of translation MT produces, because if students accept all MT works, they will get negative habits in learning a language and lack attempts to read or to write sentences in other languages. Meanwhile, if workers rely on MT, that may lead them to overlook errors of MT. In conclusion, users of MT need to use it wisely and to understand fully the benefits and risks of MT. It is vital to have translation literacy to deal with the drawbacks of MT or to make the most of MT.

References

1. Ahlan, A. and Shehu, Y. Balogun, N. (2014). Information technology enhances students’ Academic Performance. A case of University of Ilorin. Research Gate
2. Ahrenberg, L. (2017). Comparing Machine Translation and Human Translation [online] p24. Available from https://www.aclweb.org/anthology/W17-7903/ [Accessed 21 August 2019]
3. Asian Absolute. (2017). How to Pick the Right Machine Translation Software for Your Business [online]. Available from https://asianabsolute.co.uk/blog/2019/02/20/how-to-pick-the-right-machine-translation-software-for-your-business/ [Accessed 24 August 2019]
4. Denkowski, M. and Neubig, C. Stronger Baselines for Trustable Results in Neural Machine Translation. [online]. Available from https://www.aclweb.org/anthology/W17-3203/ [Accessed 23 August 2019]
5. Eksiripong, P. Machine Translation- What is it, Why Use It. [online]. Available from https://asianabsolute.co.uk/blog/2019/05/31/machine-translation-what-is-it-why-use-it-and-how-far-its-come/ [Accessed 23 August 2019]
6. Gally, T. (2017). Machine Translation and English Education in Japan. [online]. Available from http://park.itc.u-tokyo.ac.jp/eigo/KJEE/009/043-055.pdf [Accessed 21 August 2019]
7. Kaponen, M. (2015), How to Teach Translation Post-Editing? Experience from a Post-Editing Course. [online]. Available from https://pdfs.semanticscholar.org/2321/e3f395ef5f37f7ede60bd8dbf3d3587aa8cd.pdf p.5. [Accessed 25 August 2019]
8. Khan, S. (2019). Machine Translation vs Human Translation-Which one Wins? [online]. Available from https://scoopempire.com/machine-translation-v-s-human-translation-which-one-wins/ [Accessed 21 August 2019]
9. Koehn, P. (2017). Six Challenges for Neural Machine Translation: Cornel University. 68(7).
10. Luong, M. Pham, H. and Manning, D. (2015) Effective Approaches to Attention-based Neural Machine Translation [online] Available from https://arxiv.org/abs/1508.04025 [Accessed 24 August 2019]
11. Maulidiyah, F and Malang, P (2018). To Use or Not to Use Google Translate in English Language Learning [online]. Available from https://jlt-polinema.org/?p=996 [Accessed 25 August 2019]
12. Nirenburg, S. (1989). Machine Translation: Theoretical and Methodological Issues. Cambridge: Cambridge University Press 104(32)
13. Schwartz, L. (2018). The History and Promise of Machine Translation: Department of Linguistics 63(4) [online]. Available from http://dowobeha.github.io/papers/history18.pdf [Accessed 25 August 2019]
14. Shunji, I. (2015). How Long Does It Take for Japanese Speakers to Learn English: CiNii articles. [online]. Available from https://ci.nii.ac.jp/naid/110007138660 [Accessed 23 August 2019]
15. Somers, H. (2018). Machine Translation Latest Developments [online]. Available from https://personalpages.manchester.ac.uk/staff/harold.somers/Mitkov-book-chapter.pdf [Accessed 24 August 2019]
16. Takeda, K. (2019). How Should Coexist with AI? [online] Available from https://english.rikkyo.ac.jp/closeup/dn4ddm0000001ao0.html [Accessed 23 August 2019]
17. Tsuboya, I. (2017). Why Japanese Have Trouble Learning English [online] Available from https://www.japantimes.co.jp/opinion/2017/10/29/commentary/japan-commentary/japanese-trouble-learning-english/ [Accessed 23 August 2019]
18. Ugwuadu, I. and Jota, T. (2013). Survey on the Statistical Machine Translation: Cambridge University Press 422(43)
19. Wu, Y. Schuster, M. Chen, Z. Le, Q. Norouzi, M. (2016). Google’s Neural Machine Translation System: Bridging the Gap between Human and Machine Translation [online] Available from https://arxiv.org/pdf/1609.08144.pdf [Accessed 24 August 2019]