Topics In Demand
Notification
New

No notification found.

Handling Language Translation using Generative AI
Handling Language Translation using Generative AI

854

0

“Translation is a creative, generative process which may result in many different translations which are all equally good (or bad).”

                                   - Christian Federmann, Principal Research Manager on Language Translation, Microsoft

I will start this blog with a story. I was working at a firm as a Data Analyst. We got a project for a large Spain-based telecommunication organization. They had a manual ticketing process where the Network B2B team classified the tickets manually. Due to this, at times, the tickets got routed to the wrong teams. This impacted the overall SLAs and the quality of the resolution for their corporate customers. The ticket data was in form of conversations between the customer and the ticket agent. Since it was a Spain-based organization, the data was a mix of English, Spanish, and a couple of other languages as sometimes the ticket gets routed through multiple countries for resolution. Moreover, it is quite common to see that many times the words do not exist in the other language. For instance, it’s very difficult to translate the word “nice” into different languages. “Put” cannot be translated into German. Portuguese call everything – bread, dough, cake mix, batter, and pastry –"massa". Not going far, let’s just consider our rich Indian culture with multiple and diverse languages. The Tamil language has 12 stages of flower development starting from bud formation (Arumbu) until the appearance of withered fresh flowers (Pommal). Similarly, in Hindi, we have multiple words to express a single emotion. We use “Tum, Tu, and Aap” in different contexts as per age group, and courtesy. However, in English, there is just one word “You” to replace all three. Such things make translations even more complex.

The challenge we faced was during the translation of the data from multi-lingual to English as most of the Natural Language Processing (NLP) models work in English. Another challenge was the context. Machines do not understand in what context a person has written/spoken something while translating. Languages are complex, and words can have different meanings based on the situation they are used in. For example, the term “bank” can be used to describe either a financial institution or the river’s edge. Similarly, taking literal meanings for idioms can be different from the original meaning. 

Another instance can be when the translation machine algorithm was applied at scale on Meta’s platforms. Facebook mistranslated a post by a Palestinian man from “good morning” to “hurt them,” which led to his arrest by Israeli police. So, you can see that even a small number of errors can produce disastrous results.

This shows that machines lack the same level of cultural knowledge, understanding, and experience as human translators. However, with the advancement in the field of Generative AI, there have been few developments that are minimizing such challenges.

Let’s understand some of the language-translation tools in the market –

  1. Language Weaver – It is an enterprise-level machine translation that enables global organizations to manage multilingual content and workflows in real-time at scale. It helps you improve collaboration between teams, increase productivity, and go to market faster internationally.
  2. Yandex Translate – It is a platform that enables the translation of text or web pages in almost 100 languages. It can take inputs from documents, voice, images, and websites. It works similarly to Google Translate.
  3. Amazon Translate – It is a neural machine translation service that delivers fast, high-quality, affordable, and customizable language translation. The best part is it treats each word as an independent entity and remembers the connections between words within the same clause which ultimately increases the accuracy and reliability of the software.
  4. Unbabel – It is a language operations platform that facilitates effective cross-cultural client communication. It combines the best blend of machine and human translation to provide a consistent multilingual customer experience, grow to new markets and build trust around the world.

Still, there is a long way to go in terms of improvement in accuracy and reducing the nuances of translations. This is very unlikely in the short term as accuracy requires specific domain training data, and terminology, which is currently not available on the world wide web data used to train ChatGPT. Even though the machines cannot master the translation, they can be extremely helpful in automating and increasing the efficiency of translation processes.


That the contents of third-party articles/blogs published here on the website, and the interpretation of all information in the article/blogs such as data, maps, numbers, opinions etc. displayed in the article/blogs and views or the opinions expressed within the content are solely of the author's; and do not reflect the opinions and beliefs of NASSCOM or its affiliates in any manner. NASSCOM does not take any liability w.r.t. content in any manner and will not be liable in any manner whatsoever for any kind of liability arising out of any act, error or omission. The contents of third-party article/blogs published, are provided solely as convenience; and the presence of these articles/blogs should not, under any circumstances, be considered as an endorsement of the contents by NASSCOM in any manner; and if you chose to access these articles/blogs , you do so at your own risk.


images
Unnati Kohli
Technical Storyteller

Unnati is currently pursuing the flagship PGDM program at Management Development Institute, Gurgaon (2022-24). Now, she is a Strategy & Analytics Intern at Deloitte USI. Prior to this, she has 37 months of work experience as a Data Analyst in TCS in Telecom Network Domain. She also worked as a Technical Storyteller with Bloggers Alliance.

© Copyright nasscom. All Rights Reserved.