Google Translate: Assessing Its Widespread Use; The Role of Human Translators in a Technologically Centered World

Translation services have been a critical part of connecting an increasingly globalized world with the need for communication across various languages. Thus, machine translation, which uses technological patterns to calculate an appropriate translation from one language to another given an input, has become increasingly popular in attempts to address this increasing translation need. With this increasing usage of machine translation, it is important to understand where machine translation falls short. These shortcomings can help to better understand how to make appropriate improvements to popular machine translation software and utilize their capabilities effectively while limiting adverse side effects of their usage.

My technical research focuses on proposing improvements to the dataset, algorithms, and interface of Google Translate, the translation software most closely associated with the concept of machine translation. Google Translate varies in accuracy across different language pairs, and hiring translators to generate more training examples for datasets can help to address these inequalities. Bias has also been found in translations produced by Google Translate, such as with favoring particular gendered language signifiers for occupations. Algorithms to combat unfairness in datasets such as Fair-SMOTE, which rebalances the dataset based on protected attributes, could be incorporated into the process of generating training examples for language translations. In terms of the interface of Google Translate, the design is simplistic in order to be as user-friendly as possible. However, this results in a perception of perfection or objectivity with the translation, which may not be the case for some longer sentences, paragraphs, and text requiring particular contextual information to understand. Thus, interface improvements could include an option to view additional likely translations as well as adding contextual tags to narrow the possibilities for translations. Machine translation, due to its ease of use and perception of reliability, must be kept to the highest standards of quality possible. Accuracy must be prioritized in order to affirm the perception of reliability, but shortcomings must be acknowledged and addressed to ensure that users of machine translation technology use it in an effective manner.

My STS research focuses on the role of human translators as machine translation technology continues to grow in its capabilities and usage. Human translators are currently more equipped to handle the intricacies of generating an appropriate translation due to the ability to specialize in language pairs and better understand contextual information, including cultural context. However, the ease and convenience of machine translation could encourage clients of translation services to use machine translation for their needs. Thus, machine translation and human translators will likely combine forces in the future, with new roles and adaptations for translators emerging. This includes an expansion of the existing machine translation post-editor role, where translators edit machine translation output to ensure standards for translation quality are met. Other such roles could include generation of challenging training examples and a translation consultant position to assess appropriate machine translation models to be used for a particular purpose.

Improving translation services will continue to be an important goal for an interconnected multilingual world, so improving machine translation and its usage will continue to be important. Despite the different research focuses for my technical and STS research, both result in the same idea that improving machine translation requires looking at two important angles. One angle is focused on how machine translation training data is generated. Incorporating fairness algorithms and using human translators to generate training instances both serve as examples. The other angle is centered on the idea that clients should interact with machine translation in an optimal manner. This can be seen in interface improvements incorporating context and alternate translations as well as in expanding the machine translation post-editor role. These ideas can work in tandem. For instance, post-editors could use a list of alternate translations to speed up the process of finding the best translation for a particular purpose. Post-editors could also use contextual tags to generate a more useful translation. Overall, machine translation is a product of technological innovation that works for an inherently social purpose, as it exists to connect speakers of different languages from across the world, so considering its usage necessarily requires examining how it interacts with society at large.

School of Engineering and Applied Science

Bachelor of Science in Computer Science

Technical Advisor: Briana Morrison

STS Advisor: Sean Murray

Technical Team Members: Siddharth Premjith