In a world defined by global connectivity through using information technologies, Machine translation is one of our most important tools in creating a linguistic bridge. This tool can instantaneously translate spoken or written words between different languages, promoting understanding and unity. As we know, translation plays a crucial role in facilitating cross-linguistic communication between individuals, businesses, and nations. The increasing demand for efficient and effective translation in today’s interconnected world highlights Machine translation’s critical role in facilitating networking, trade, and diplomatic engagement. It demonstrates the mutual benefit of artificial intelligence and computational linguistics, establishing smooth cross-cultural communication. In this blog post, we will explore the various aspects of machine translation, including its development, effects on the work of the present-day translator, and the ethical questions created by Artificial Intelligence in the field of dubbing and voice-over.
To begin, let’s take a look at the types of machine translation available to us. To start, Rule-Based Machine Translation (RBMT) was one of the earliest versions of machine translation. This model dating back to the early 1970s relies on a bilingual dictionary and predefined linguistic information or rules that specify how a term or phrase should appear in the target language. While this works for certain language pairs and domains, it is not adept at recognising subtle expressions and is prone to idiosyncrasies. PROMPT (Pioneering Research on Machine Translation) is one of the more well-known RBMT technologies, and you can purchase it to aid in your translations in software such as SharePoint. Statistical Machine Translation (SMT) was introduced in the 1990s, incorporating statistical models and bilingual corpora in order to produce a translation. With this approach, translation quality was greatly enhanced and a wider range of language pairs and domains were possible. SMT adopted the principle that if people tend to translate in a particular manner, the machine should follow suit. While this works in some cases, the model still faces challenges in handling complex linguistic structures.
The final phase brings us to Neural Machine Translation (NMT), a system that I have wrote about already on this page. See my presentation on DeepL to find an in-depth rundown of how NMT works. NMT employs artificial neural networks and deep learning. This revolutionary model has significantly improved the quality of translations by accurately interpreting complex linguistic nuances and patterns. In contrast to the RBMT and SMT models, Neural Machine Translation has the capacity to learn and enhance its performance with each translation task, continuously improving its translations based on its newly acquired knowledge. It can seamlessly comprehend context, correctly translate idiomatic expressions, and generates fluent and natural sounding output.
Google Translate is the most popular model of NMT that people of all ages and backgrounds use daily. Let’s dive into this particular tool to examine in detail the workings of Neural Machine Translation. Google Translate’s implementation of NMT in 2016 has significantly elevated the quality of its translations. The complex neural network that supports Google Translate’s NMT is made up of numerous layers of interconnected nodes. Nodes represent the basic components of the network’s architecture. They are finite computational units. Also called neurons, these nodes function as computational algorithms in the network, processing and sending data. To help the network learn and adapt to particular tasks, each node has a set of variables, such as weights and biases, which are changed during the training phase.
In the case of Google Translate’s NMT, the nodes are essential to process the input data, which in this case is the text that needs to be translated from the source language. The nodes form layers, each of which performs a different computation-related function. This allows the network to identify intricate linguistic structures and patterns. Through a process called forward propagation, data transmits through the interconnected nodes from the input layer, where source text is initially inputted, towards the output layer, where the translated text is then subsequently produced. Backpropagation, the repeated adjustment of each node’s weights and biases during its initial conditioning, enables the neural network to optimise its parameters and improve its capacity to detect complex language patterns. As you can see, these interconnected nodes play a vital role in developing accurate and contextually meaningful translations.
Now, the burning question: how are machine translators impacting the work of a human translator? Are we going to have a Terminator scenario on our hands in the next 50 years? I hope not. Human translators are using MT systems more and more, and working alongside these systems have become a characteristic of modern translation processes. Human translators are using MT technologies to boost their productivity, efficiency and creativity. These tools simplify the initial translation process and allows a more efficient workflow. Equipped with the assistance of a MT, translators may quickly produce rough translations, freeing up their time to concentrate on fine-tuning the final product. This can be particularly helpful in situations when a lot of information needs to be translated quickly.
However, the balance between human and machine translator is delicate, and the human translator must make sure to keep an eye out for cultural nuances or idiomatic expressions that the computer may miss (and the computer can keep an eye out for when the translator messes up the conjugation of a tense because they didn’t get enough sleep the night before). Take a legal document as an example of a text that a human translator has to translate. Even though MT can provide a draft translation quickly, a human translator is still necessary to guarantee that legal terminology is accurate, understand complex legal details, showcase contextual sensitivity and tailor the translation to the target language’s legal context. This complementary relationship between human and machine is crucial and in my opinion, only further aids a human in their work as a translator.
The entertainment and media industries are currently experiencing an evolutionary shift with the introduction of Artificial Intelligence into dubbing and voice-over applications. AI has proven significant in producing synchronised translations with exceptionally natural-sounding voices due to its capacity to imitate human speech patterns and intonations. In the field of dubbing, AI algorithms carefully examine the audio in the source language audio, detecting grammatical complexities and emotive signals. The translated scripts are then synchronised with one other, guaranteeing the dubbed content is accurate and authentic. This technology speeds up the dubbing process, cutting expenses and production time while improving the materials accessibility for a wide range of viewers. AI developments are also beneficial when it comes to creating human-like voices for voice-over applications. Again, this speeds up the translation of material and simplifies the production process.
As voice-over and dubbing become more reliant on AI, a number of ethical issues arise that must be taken into account. The first thing that comes to mind for me is: “are people’s jobs safe?”. The shift towards AI voice-overs and dubbing highlights the possible unemployment of dubbing professionals and real voice actors. Moreover, the responsible use of AI voices is being talked about more and more. There are serious ethical issues with consent and the digital replication of a voice without explicit permission. An additional degree of complication is introduced by the possible exploitation of AI generated material, heightening worries about identity, theft, false information, and the decline in public trust in audio-visual media.
It’s important to examine how AI-generated material affects society and culture. Efficiency may come at the expense of the unique tone and emotional resonance that real human voices offer when telling stories. A proper balance between technology and maintaining the integrity and authenticity of human’s natural creativity is crucial if we want to withhold a safe and responsible relationship between humans and their use of AI. Respect for creative professions, cultural sensitivity and general standards in the entertainment industry can ensure that we continue our good relationship with Artificial Intelligence and Information Technologies.
It is evident that Machine Translation has revolutionised the translation world, encouraging a balanced alliance between human translators and technology. The invention of Neural Machine Translators, and the incorporation of AI in dubbing and voice-over applications signify how efficient and convenient the world around us has become. While machine translation offers the human translator lots of free time to brush up on the little details of their work, Artificial Intelligence poses a real threat for those with jobs in voice acting or dubbing. We must ensure responsible use of technology so it aids us in our everyday lives, and doesn’t end up being more of a curse rather than a blessing in the world of communication.
Bibliography
Baños, Rocío. “”Key challenges in using automatic dubbing to translate educational YouTube videos.” Linguistica Antverpiensia, New Series–Themes in Translation Studies 22 (2023).
Hearne, Mary, and Andy Way. “Statistical machine translation: a guide for linguists and translators.” Language and Linguistics Compass 5.5 (2011): 205-226.
Shiewen, Yu and Bai Xiaojing. “Rule-based machine translation.” Routledge encyclopedia of translation technology. (2014): 186-200.
Wu, Yonghui, et al. “Google’s neural machine translation system: Bridging the gap between human and machine translation.” arXiv preprint arXiv:1609.08144 (2016).
Yuan, Zichen, and Haina Jin. “The application of machine translation in automatic dubbing in China: A case study of the feature film Mulan.” Linguistica Antverpiensia, New Series–Themes in Translation Studies 22 (2023).