How AI is helping learning: A Case Study on automatic translations

    The use of Artificial Intelligence technologies is established in many areas. Most of us are probably using AI every day, at work and at home. Alexa, Cortana and other virtual assistants are just a few examples of popular AI applications that we already use. Even Facebook and Google use AI as well as machine-learning to recognise pictures of our friends in our photos and prioritise email. AI makes it possible for machines to learn from experience, to adjust to new inputs, and perform human-like tasks. Taking these cues, it can also be applied to learning.

    Here, we will discuss one of these applications. Namely, automatic translations. Translation AI-based translation engines are extremely useful and exciting applications. In this context, AI allows the accurate detection of a language of any text string. AI also enables one to customise text translations using previous translations. The historical machine-learning technique was Statistical Machine Translation.

    This would find the best possible translation of a word by grabbing just a few words around it to understand the context. In about 2011 came Neural Machine Translation (NMT) and with its emergence, a lot of issues were solved.

    How does it work?

    NMT translates a word based from the whole sentence and not just from a couple of surrounding words. Consider the inside of the human brain, comprising 7,000 neural networks that process information given to them by our senses. The NMT model is based on the same concept. During the translation process each sentence is passed through a channel or layers to evaluate each word to determine the characters. This approach provides a more accurate translation. NMT utilises ‘deep-learning’ techniques to constantly learn from the ecosystem created. Creating an ecosystem contains a minimum of 10,000 sentence-pairs of the specific targeted languages. A custom translator will create a model from the sentence pairs. This approach provides a more accurate translation.

    NMT utilises ‘deep-learning’ techniques to constantly learn from the ecosystem created. Creating an ecosystem contains a minimum of 10,000 sentence-pairs of the specific targeted languages. A custom translator will create a model from the sentence pairs. This model will contain an organisation’s ‘lingo’, standard of translation, and also provide more examples of complete sentence structures.

    Advantages:
    • The speed and quality of translation
    With NMT the ability to learn linguistic rules on its own from statistical models, results in a quality translated document in seconds

    Cost
    How NMT works is not simple, but the ability to train a central tool, and deploy, are very cost-efficient. The cost savings also come with the quality of translation. With a higher-quality translated document you will see a great cost saving when handing these translated documents to a localisation company to have those documents processed.

    Accuracy
    Research* by Bentivogli et al (2016) focused on NMT vs Traditional Machine Translation and found NMT to have:
    • 70% less verb placement errors;
    • 50% less word order errors;
    • 19% less morphological errors; and
    • 17% less lexical errors.

    CASE STUDY:

    FORTUNE 100, TOP 10 LARGEST GLOBAL SOFTWARE FIRM

    The authors have proprietary research that demonstrates the benefits, in terms of costs reduction (of between 30% and 50%), for multi-national organisations. This has resulted in a proven 80% reduction in production time for global workforces with multiple language needs. This includes translation of those languages with different character sets.

    Background, scenario and need

    The focus organisation is a Fortune 100 Top 10 largest global software firm. It has multiple global locations and a workforce that speaks 29 different languages, including those with different character sets.

    Each year the company has a course ‘roll-out’ to its global sales organisation, which takes six months to develop in English.
    The organisation needs to deliver the course to the (Asia pacific) market, and the amount of time to translate in the traditional way (six months) has impacted its ability to deliver in a timely fashion.

    Result

    With consultation, a solution was developed using emerging translation technology that shortened the time from six months to one month, and saved the organisation a significant amount of money in the process.

    While using NMT it is possible to see a cost saving of anywhere between approximately 30% and 50%. This of course is different for each organisation, and also depends on the particular desired language(s)/character sets.
    The authors, while using proprietary UI, and the fast processing time of NMT, have seen a reduction of approximately 80% time-savings.

    Conclusion

    The application of AI to e-learning content is not just a cost-saving solution; it also presents a whole new way of looking at learning itself.
    Apart from the quality of learning, AI provides a valuable solution for training in highly-dynamic industries. Companies that need to continuously update their content, will benefit from adaptive-learning environments once machines have the ability to accurately predict how course material needs to improve and change.
    Intelligent learning environments can also analyse data across all personalised training instances, to recommend improvements and highlight inefficiencies that would not be possible otherwise.

    * Reference: Bentivogli et al (2016), Neural versus Phrase-Based Machine Translation Quality: a CaseStudy

    A version of this article first appeared in Learning Technologies magazine: https://view.pagetiger.com/dfkwwlx/17/page1.htm

    For further information, get in touch on info@exactls.com and/or download our E-book.