Open Source, Multilingual AI and Artificial Neural Networks : The new Holy Grail for the GAFA

Since 2016, there has been a sharp increase in open source machine translation projects based on neural networks or Neural Machine Translation (NMT) led by companies such as Google, Facebook and SYSTRAN. Why have machine translation and NMT-related innovations become the new Holy Grail for tech companies? And does the future of these companies rely on machine translation?

Never before has a technological field undergone so much disruption in such a short time. Invented in the 1960s, machine translation was first based on grammatical and syntactical rules until 2007. Statistical modelling (known as statistical translation or SMT), which matured particularly due to the abundance of data, then took over. Although statistical translation was introduced by IBM in the 1990s, it took 15 years for the technology to reach mass adoption. Neural Machine Translation on the other hand, only took two years to be widely adopted by the industry after being introduced by academia in 2014, showing the acceleration of innovation in this field. Machine translation is currently experiencing a golden age of technology.

From Big Data to Good Data

Not only have these successive waves of technology differed in their pace of development and adoption, but their key strengths or “core values” have also changed. In rule-based translation, value was brought by code and accumulated linguistic resources. For statistical models, the amount of data was paramount. The more data you had, the better the quality of your translation and your evaluation via the BLEU score (Bilingual Evaluation Understudy, the most widely used algorithm measuring machine translation quality). Now, the move to Machine translation based on neural networks and Deep Learning is well underway and has brought about major changes. The engines are trained to learn language as a child does, progressing step by step. The challenge is not only to process exponential data (Big Data) but more importantly to feed the engines the most qualitative data possible. Hence the interest in “Good data.”

Continue reading

Artificial Intelligence: And You, How Will You Raise Your AI?

[This article originally appeared on Kirti Vashee’s Blog]

This is the final post for the 2017 year, a guest post by Jean Senellart who has been a serious MT practitioner for around 40 years, with deep expertise in all the technology paradigms that have been used to do machine translation. SYSTRAN has recently been running tests building MT systems with different datasets and parameters to evaluate how data and parameter variation affect MT output quality. As Jean said:

” We are continuously feeding data to a collection of models with different parameters – and at each iteration, we change the parameters. We have systems that are being evaluated in this setup for about 2 months and we see that they continue to learn.”

This is more of a vision statement about the future evolution of this (MT) technology, where they continue to learn and improve, rather than a direct reporting of experimental results, and I think is a fitting way to end the year in this blog.

It is very clear to most of us that deep learning based approaches are the way forward for continued MT technology evolution. However, skill with this technology will come with experimentation and understanding of data quality and control parameters. Babies learn by exploration and experimentation, and maybe we need to approach our continued learning, in the same way, learning from purposeful play. Is this not the way that intelligence evolves? Many experts say that AI is going to be driving learning and evolution in business practices in almost every sphere of business.

Continue reading

SYSTRAN at the Digital Forensics and Analysis Summit

Digital SummitOn October 16-17th, SYSTRAN and its partner Relativity will be participating in the Digital Forensics & Analysis Summit as sponsors and exhibitors. The Digital Forensics & Analysis Summit is a two-day forum that will gather international experts from around the world in Abu Dhabi to share best practices on how technology is used in their forensics department to extract evidence that is able to stand up in trial.

Since information governance, forensics and eDiscovery procedures face mounting pressure from the growth of Electronic stored Information, legal standards and rules governing digital investigation requirements have also contributed to the rise in litigation and associated legal costs.

Within this environment, documents written in languages other than English, including data collection, processing and reviewing can pose major challenges, especially when ensuring the mandatory confidentiality of those procedures, as these typically forbid online translation. Organizations need to search by keyword and find relevant documents and emails in the appropriate languages while controlling costs and maximizing productivity. Therefore time-intensive human translation is usually not an option and the need for viable machine translation solutions becomes all the more apparent.

Continue reading

Digital platforms and the post-language economy

This article was originally published on ITProPortal Digital platforms and the post-language economy by Denis Gachot.

The world can get even smaller, and new technology is making that happen.

When we imagine international companies, we think large, publicly traded conglomerates that have substantial resources and funds to facilitate operations on opposite ends of the globe. But that is changing.

Already the Internet has shrunk the world so that small companies now rely on software engineers in Pakistan and marketing agencies contract with graphic designers in the Philippines. But the world can get even smaller, and new technology is making that happen.

Today, the new frontier is language. Individuals have a plethora of platforms that allow them to access consumers all over the globe and work with other companies in faraway places – if only they could speak the same language. In an ironic twist, language has turned from something that first facilitated human cooperation and growth, to something that currently impedes our ability to work together.

Technology may finally be ready to abolish that barrier forever. It is somewhat remarkable that in 2017, more than 20 years after widespread use of the Internet began, we still rely almost exclusively on humans to translate language in commercial formats. But translation bears all of the earmarks of those functions that artificial intelligence ought to be capable of replicating, and a technology called Neural Machine Translation (NMT) does just that.

Contextual translation ability

By leveraging its contextual translation ability alongside its deep learning functions, NMT has achieved historic results in the journey to a post-language economy. In a side-by-side comparison with human translators, in a technical domain translation for English-Korean, SYSTRAN’s NMT translations were preferred 41 per cent of the time. That success is achieved by advancing language translation beyond rule-based translation methods.

Before NMT, machine translation models – known as rules-based or ‘phrase-based’ – were only able to reference five to seven words at a time when determining meaning. Each language pair has its own linguistic challenges, but it made the translation for certain languages, like Japanese, more challenging because you need to know the entire sentence to put all of the words into context.

Advertisement
Let’s say a colleague forwards you an urgent email, and it includes this sentence: パリに出張の時に私はCEOに会いました.

With a phrase-based machine translation, you would receive this output: ‘I met Paris in the CEO trip doing business.’ With NMT, you would get: ‘I met the CEO when I was in Paris on a business trip.’

In Japanese, main verbs are placed at the end, so you need to reference the end of a sentence to make sense of the phrases within it. NMT processes the entire sentence (and soon paragraph) from end-to-end without intermediate stages.

That is why context is so important. The effect of machine translation being able to better understand context results in a huge jump in BLEU score (the industry measurement for accuracy). SYSTRAN’s Pure Neural Machine Translation (PNMT) program has seen increases in all 61 language pairs and where we see the biggest increase is in Asian languages. For some languages, we saw a jump of 200 per cent in BLEU score.

Gisting

With machine translation, we have a metric called ‘Gisting,’ as in ‘you get the gist.’ In addition to this metric, we test whether or not a user can solve their problem with the translated output. Were they able to search a FAQ and customise a piece of software with the answer? Were they able to search a digital database of products and images and find what they were looking to purchase? If yes, then they got the gist.

“Gisting requires extensive post-editing. NMT has moved us into fluency. What fluency allows is the ability to read and understand so that you no longer need to post edit,” says Ken Behan, V.P. of Sales and Marketing. NMT is allowing us to focus on ‘meaning’ and ‘fluency’ scores. With fluency, we ask if the translation sounds like a native speaker wrote it. With fluency and meaning, we can ask:

Were they able to understand a review and make an educated decision? Were they able to read the manual specs and assemble a piece of heavy machinery? Were they able to find a product, read the description and purchase what they were expecting?

Refer to the English translations above. With the first sentence, did you get the gist? Yes! You could infer someone went to Paris on something business related.

With the second, did you understand the meaning and the fluency? Yes, you can understand what kind of trip it was and what happened.

New solutions

Neural Machine Translation will further advance traditional MT solutions and create new ones in communication, customer support, e-Learning, eDiscovery, compliance and user-generated content to name a few. Also, early-adopting linguists using NMT are already increasing their productivity.

Similar to the human brain, the neural machine translator learns through a process in which the machine receives a series of stimuli over several weeks. This development, based on complex algorithms at the forefront of Deep Learning, enables the PNMT engine to learn, generalise the rules of a language from a given translated text, and produce a translation close to human levels of competency.

You can think of NMT as part of your international go-to-market strategy. In theory, the Internet erased geographical barriers and allowed players of all sizes from all places to compete in what we often call a ‘global economy,’ But we’re not all global competitors because not all of us can communicate in the 26 languages that have 50 million or more speakers. NMT removes language barriers, enabling new and existing players to be global communicators, and thus real global competitors. We’re living in the post-internet economy, and we’re stepping into the post-language economy.

The difference between previous language technologies and what NMT can do today is remarkable, and business leaders should take note. Just ask yourself who you would want to do business with: the guy that says, “I met Paris in the CEO trip doing business,” or the one that says, “I met the CEO when I was in Paris on a business trip.”

SYSTRAN’s team is setting private meetings for an exclusive view of the PNMT concept. For more information, contact Craig Stern at craig.stern@systrangroup.com and to set up a meeting, click here.

Related Links

SYSTRAN PNMT DEMO

Positive Feedback from PNMT Beta Tester

This article was originally published on ITProPortal Digital platforms and the post-language economy by Denis Gachot.