What is Neural Machine Translation (NMT) all about?

Neural Machine Translation is an approach to machine translation that uses large neural networks to produce translations that are more natural sounding and achieve greater levels of fluency. These networks are trained on sequences (or sentences), which means they solve many of the syntactical and grammatical errors previously associated with Phrase-Based Statistical Machine Translation.

With the emergence of relatively cheap, super-powerful computers, coupled with practically unlimited storage capacity due to the emergence of “the Cloud”, we can now compute these complex NMT models in several hours.

These highly efficient models can then provide fast and fluent translations, and at an economically advantageous price. Today, over 90% of the daily traffic on the KantanMT platform is processed by our NMT services. This gives you an indication of the high regard our customers have for the efficacy of our NMT platform.

Why is there so much hype around NMT?

Simply because it provides a demonstrable improvement in machine translation outputs, resolving many of the problems associated with the statistical machine translation (SMT) model. Effectively, we have in the last two years solved many of the translation shortcomings of the SMT system; deficiencies that we’ve been trying to resolve for the last two decades! So, you can imagine how excited we are to be able to move with such speed compared to the pace of development we were lumbered with when working with SMT.

An interesting factoid about NMT is that we actually don’t train them using “whole words”, but we in fact train them using “word pieces”. And even if we don’t have parallel training data for a language combination, we can build a zero-shot network that will be capable of producing translations in these languages! It’s completely amazing what we can do once we set up the deep learning approach and throw super-computers at the problem. In Deep Learning (DL) we use highly sophisticated, multi-layered, pattern of ‘neurons’ to process huge chunks of data looking to refine the information contained within that data. The DL process can take an abstract jungle of information (word pieces), as is contained with data, and using the power of super-computation refine the data in to clearly understood language.

Can you imagine how good Neural MT will be in a further two years? Will Moore’s Law of exponential technological growth apply to Neural Machine Translation too? I believe it will. It will be amazing to see then how powerful NMT will be. It is certainly something that excites us here at KantanMT.com.

Which languages have made the greatest progress for NMT?

Any language that has a deep and complex grammatical structure can now be efficiently modelled using Deep Leaning and Neural Networks. For example, take the grammatical characteristic of the humble German verb – under normal circumstances it needs to be positioned at the end of a sentence. That would seem a straight forward enough challenge? However, SMT struggled to position the German verb accurately. To overcome this, we at KantanMT.com used advanced part-of-speech reordering approaches to improve this accuracy. This was a very complex, time-consuming and computationally intensive approach. However, NMT (because we train the engines on full sentences) almost always correctly positions that illusive German verb. This methodology also allows us to meet the challenges of languages such as Hungarian and Finnish. These are now well within our capabilities, allowing us to produce very good translation outputs using NMT.

Where do you see the translation industry in the next 5 years?

What an exciting time to be in the Localization Industry! We are on the cusp of a massive explosion in Artificial Intelligence (AI), which will impact all facets of the localization industry’s workflow and processes.

The industry will use automated translation technology to process even more content, in to more languages, and faster than ever before. Translators should not fear, as they will be the main beneficiary of this transformation. As this technological evolution grows, translators will be able to produce more words per day and consequently, significantly improve their income levels. I envisage a scenario whereby translation from scratch will be viewed as old-school and passé. The translation model will change in the same way as Computer Aided Translation (CAT) transformed the industry for the better. In the new NMT paradigm, the post-editing of a constantly improving machine translation output will be seen and accepted as the modern, progressive way of working. And the industry will be the better for it.

AI will also enable better job matching and candidate selection – so translators will be selected based on their relevant skill sets, domain knowledge and previous job performances. This is not to be feared, as essentially this is they way we choose our dentists and doctors today. AI wil

l become a driver for greater competition and increased professionalism in our industry.

I also see AI being becoming part of the project management workflow system, and the project management role.  PM systems will be expected to handle real-time translation workflows. A system that will combine automated translation and “human touch post-editing” to provide almost instantaneous results.

On the quality side, translation errors and problems will be identified by AI checkers and automatically routed for automatic recovery and fixing. The time between job arrival and completion will reduced in some cases to seconds. These “micro-jobs” will be driven by the requirement for new content to be translated in effectively “real time”. This fast system will be required for content such as blogs, wikis, live user forums, reviews, internal corporate content, help chat lines etc.

What should we expect from KantanMT in the next few months?

We’re working on a new type of Neural Network that will provide even better translation outputs than before, with a significantly reduced training time. These new networks are already in testing with one of the largest eCommerce companies; so, stay tuned for further news of this major step forward in the evolution of NMT.

Additionally, we have figured out a way of measuring the quality of an automatically generated translation. This Quality Estimation Score system was developed by KantanMT.com for the European Commission. The good news is, we shall be open-sourcing this technology in early 2019.

You’re also going to also see a new, improved version of KantanLQR that will support multi-lingual quality projects. It will give you the means to measure how individual language arcs are performing across your enterprise.

This article first appeared on http://www.argosmulitlingual.com in October 2018: http://www.argosmultilingual.com/blog/shaping-the-path-to-neural-machine-translation-with-tony-odowd

Today’s blog is aimed at helping the novice understand the technology that is Deep Learning (DL). To do this, I will need to discuss in-depth Linear Algebra, Statistics, Probability Theory and Multivariate Calculus. Only Joking! Nothing would turn the novice readers off than trying to hack our way through the above complex disciplines. We’ll leave that for the nerds. Today’s blog – like my last on Machine Learning – will try and use an analogy to help explain what is without doubt a very multifaceted, intricate subject to fully master.

For myself, the more I read about Deep Learning, and the more I spoke to the engineering masterminds at KantanMT the more I realised that the discipline of using a Deep Learning model bore a similarity to sculpting. Let me expand: I don’t know to whom this quote is attributed, but for me it certainly describes the methods of Deep Learning:

“The sculpture produces the beautiful statute by chipping such parts of the marble block as are not needed – it is a process of elimination.”

Indeed, I think it was no less than Michelangelo, who when asked about sculpting, said that the angel lay within the marble block; it was simply his job to release it. Michelangelo’s minimalist explanation, and the above quotation, encapsulate in its simplest form what the Deep Learning progression involves. The engineer is the sculpture. The marble block represents the huge block of dense data to be processed.  The act of processing the data is the chipping away of unwanted information by neural networks. The act of fine tuning the deep learning neural engine represents the technique of the sculptor carefully finessing the shape of the emerging form in to a recognisable figure.

In both the role of sculptor and engineer there is a vision of what the ‘fine-tuning’ activity should produce. I am confident that if you as a novice accept this simple analogy you are going someway to grasping the very fundamentals of the Deep Learning process.

As a concept, Deep Learning is less than two decades old. The origin of the expression is attributed to Igor Aizenberg, Professor and Chair of the Department of Computer Science at Manhattan College, New York. Aizenberg studies, amongst other things, complex-valued neural networks.  He came up with the concept of an Artificial Neural Network system based on that of the human neural network – the network of the human brain.

The ‘Deep’ element of the concept refers to a multi-layered processing network of neuron filters. The equivalent process in the human brain is that of information flowing through neurons connected by synapses. In the machine equivalent, artificial neurons are used to fine-tune and refine data as it is passed through the ‘engine’. The process of Deep Learning also learns from experience and can adjust its processes accordingly. In sculpting, it is the equivalent of the experienced sculptor chipping and refining the marble to release Michelangelo’s hidden angel.

Jeff Dean, a Senior Fellow at Google’s ‘System and Information Group’ – the group behind many of Google’s highly sophisticated machine learning technologies – said:

“When you hear the term ‘Deep Learning’ just think of a large neural net. Deep refers to the number of layers typically, and so this is kind of the popular term that’s been adopted by the press.”

For many novices there is a confusion around the terms Machine Learning (ML), Artificial Intelligence (AI) and Deep Learning (DL).  There need not be this confusion as the division is quite simple: Artificial Intelligence is the catch-all term to cover Machine Learning and Deep Learning. Machine Learning is an over-arching term for the training of computers, using algorithms, to parse data, learn from it and make informed decisions based on the accrued learning. Examples of machine learning in action is Netflix showing you what you might want to watch next. Or Amazon suggesting books you might want to buy. These suggestions are the outcome of these companies using ML technology to monitor and build preferences profiles based on your buying patterns.

Deep Learning is a subset of ML. It uses a highly sophisticated, multi-layered, pattern of ‘neurons’ to process huge chunks of data looking to refine the information contained within that data. It takes an abstract jungle of information, as is contained with data, and refines these in to clearly understood concepts. The data used can be clean, or not clean. Clean data is the processing of refining the pre-processed information to remove any clearly irrelevant information. Clean data can be processed quicker than data that has not been cleaned. Think of it as the human brain blocking out extraneous information as it processes what is relevant, and discards what is irrelevant. Something the human brain does every minute of every day.

But why has Deep Learning suddenly taken off so spectacularly? It is because of the ability to train Artificial Neural Networks (ANN) to a level of accuracy when trained with huge amount of data. ANN can synthesise complex non-linear processes with a high degree of accuracy. DL is also becoming predominant because of the following boosters:

• The emergence of Big Data
• The increase in computational power
• The emergence of The Cloud
• The affordable availability of GPU and TPU
• The development of DL models using open source code

Today it is estimated that Big Data provides 2.5 quintillion bytes of information per day. Now, if you are like me, you’ll will never have heard of the measure quintillion. Well, apparently, it is 1 million billion. Not that helps give it finer focus!

According to IBM:

“90% of the data in the world today has been created in the last two years. This data comes from everywhere: sensors used to gather shopper information, posts to social media sites, digital pictures and videos, purchase transaction, and cell phone GPS signals to name a few. This data is big data.”

It is safe to say that the amount of data available will only increase over the coming years. Institutions such as the European Union, the United Nations, the World Bank, the World Health Organisation, Social Media companies etc make huge volumes of data available daily, and in multilingual form. The importance of this resource of massive data is underlined by Andrew Ng, Chief Scientist at Baidu, China’s major search engine, who said:

“The analogy to deep learning is that the rocket engine is the deep learning models and the fuel is the huge amounts of data we can feed to these algorithms.”

The advent of Cloud Computing has allowed even small companies to have virtually unlimited storage space, and access to fantastically powerful computational power. Processors of the power of tensor processing unit (TPU) are available via Cloud computing. Some examples of Cloud computing sources would be Amazon’s Web Service, IBM’s SmartCloud or Google’s Cloud.

TPUs were developed by Google to specifically deal with the demands of ANN. Previously, graphics processing unit (GPUs) reduced from weeks to hours the machine learning process. TPUs have speeded up that process exponentially. Without this level of computing power, it is unlikely Deep Learning would be a viable technology.

Finally, Intel is reportedly developing a device called a Neural Stick which they claim will allow companies to bypass the Cloud to do their processing at a local level (i.e. non-Cloud level). This will be a boost to those companies who baulk at the security implications of processing data in a remote location. It will also increase the speed of processing as all the crunching will be done at the local level. Intel say it is their intent to make DL work “everywhere and on every device”. If they succeed, Deep Learning will expand to a huge degree. Interesting times lie ahead for Artificial Intelligence.

Aidan Collins is a language industry veteran. He is Marketing Manager at KantanMT. This article first appeared in Multilingual in the December 2017 edition: https://multilingual.com/all-articles/?art_id=2592

 

 

It is that time of year when Janus-faced we look over the year just passed and towards the year about to start to get a sense of how much progress we have made, and what progress may lie before us. It would be true to say that over the last 12 months Artificial Intelligence (AI) has become a norm in our lives and is now part of the vernacular. People have now accepted that their lives interact multiple times a day with AI, and that such technology is becoming ubiquitous within their lives.

And looking forward to 2019, what does it hold for us? Well, if we pay heed to the predictions of Nostradamus, we should brace ourselves for flooding, wars and a strike by a meteor. Nothing there to bring cheer but, assuming we survive all of that, what does 2019 hold for technology trends? Well, this is what we at KantanMT are predicting for the new year ahead:

Artificial Intelligence

We’ve seen an explosion in the use of Artificial Intelligence in the delivery of Neural Machine Translation during 2018; expect this to continue into 2019 and beyond. AI is the catch-all term to cover Machine Learning and Deep Learning. Machine Learning is an over-arching term for the training of computers, using algorithms, to parse data, learn from it and make informed decisions based on the accrued learning. Examples of machine learning in action is Netflix showing you what you might want to watch next. Or Amazon suggesting books you might want to buy.

Within the localisation industry, the use of AI in the form of Machine Translation (in several forms) has significantly improved translation quality outputs, speeded up translation of huge quantities of data and reduced the price of translation to make it economically viable.

AI refers to computer systems built to mimic human intelligence (i.e. imitating human neural abilities) and to perform tasks such as image recognition, parsing speech forms, discerning patterns from complex data sets, and informing accurate decision making. What’s more, AI can do these tasks faster, cheaper and more accurately than humans. Although AI has been around since the 1950s, it can be truly said that it has now come of age. This maturity has been propelled by the ever-increasing computational power now available in the Cloud.

According to Forbes, five out of six people use AI technology each day. These services include such things as navigation apps, streaming services (Amazon Alexa, Netflix etc), smartphone personal assistants, dating apps and even smart home devices (e.g. remote-activated home security systems). Additionally, AI is used in recommendation engines used by eCommerce sites (Amazon, Netflix etc), to schedule trains, to predict maintenance cycles and for other mission-critical business tasks.

For the localisation industry, AI will become a highly-integrated component of Machine Translation (MT) systems. The role of the human translator will continue evolving to that of an editor of MT texts, rather than translator of raw texts. In addition, pricing models will continue to move from the traditional price per word based on word volumes, to pricing on a time-measured rate.  MT will become an integral part of the standard workflow. The reality of real-time translation – driven by such technology as the Internet of Things (IOT) – will see project managers/editors managing workflows of projects required by customers who need a constant flow of updated information as needed by the IOT. MT will become part of the translation process just as much as CAT did in the past. And, as ever, evolving technology will bring with it a desire for speedier and cost-effective solutions.

Machine Learning

Machine Learning (ML) will continue to grow as a tool used by most localisation departments as the requirement for the speedy translations of large datasets continues to be a driver in the industry.

ML is a subset of Artificial Intelligence: with ML, computers are automated to learn to do something that they are not initially programmed to do. So, ML is an over-arching term for the training of computers to use smart algorithms to automate actions, to parse complex data and to learn patterns from this learning thus enabling the computer to make informed decisions based on this accrued knowledge. Machine Leaning can be broadly broken down into two types of learning – supervised and non-supervised learning.

For supervised machine learning, the training data is pre-labelled and consists of an aligned input data set and desired output data set. For example, an input data set could be a translation memory. An ML algorithm analyses the training data and maps how to convert future inputs to match the learned, desired output data sets.

Unsupervised machine learning is like supervised machine learning; however, the input data sets are not pre-classified or labelled. The goal of unsupervised machine learning is to find hidden structures in the unlabelled data.

So how does this impact the localisation industry? Well, suppose you want to build a translation system to translate from Zulu to French, without any Zulu-French training data? The answer is, you can combine both supervised and unsupervised approaches to achieve this. You can use an English-Zulu data set in combination with an English-French data set and using unsupervised machine learning, the system can learn how to translate from Zulu into French.

This approach is commonly referred to as ‘Zero-Shot’ machine learning – expect to hear more about this in 2019 for machine translation systems for long-tail languages.

Blockchain

I know what you’re thinking – why have we put Blockchain into this blog? Sure, isn’t that technology used only for Cryptocurrencies such as Bitcoin? Well you’re correct; while blockchain is most widely known as the technology behind Cryptocurrencies, it offers security that is useful in many other ways.

In simple terms, blockchain can be described as data you can add to, but not take away from or change. These ‘blocks’ of data can be ‘chained’ together to create incredible secure data repositories. Not being able to change any previous blocks is what makes it so secure.

This enhanced security is why blockchain is used for cryptocurrencies. It is also why it will play a significant role in localisation where it will used to protect information such as a client’s financial details, and to protect and preserve translation memories; especially in TMs used in distributed translation workflow scenarios.

Edge Computing

Cloud computing has now become mainstream: most of all global companies now rely on this centralised hosting structure for machine learning and powerful computational power. This Cloud market is dominated by just a few gigantic companies such as Amazon, Microsoft, Google, and IBM. However, now that we’ve been using Cloud Computing for some time, companies have realised that accessing all data from a central repository introduces a time-delay latency, which in turn slows down the delivery of services which can, in turn, increase costs. The “round trip” made by Cloud-based data is seen by many of today’s companies as a hindrance to their business growth.

Technology stands still for no man, and so, for many, the Cloud has reached its peak as a service for some technologies. The Cloud will continue to be used to analyse and process huge swathes of data, but the advent of the Internet of Things (e.g. connected security systems, electronic appliances, vending machines, automated lighting etc), where data processing needs to be high speed, if not real time, demands a different model. So, the logical, and necessary next move is to move this data processing to the Edge. The Edge simply means that data processing is moving from a centralised, far away location to a geographical site closer to the data source. The advent of powerful computer chips which allows such processing to be done locally has expedited this move to the Edge. Indeed, many of today’s Cloud setups automatically look to place the processing of data at the optimum Edge site for that data’s requirements.

So, Edge Computing solves the latency problem by simply moving the data processing closer to home. Closer to home means less time spent uploading and downloading data. Instead of the – centralised storage model, which has hitherto driven AI, companies are moving their data into the “local community” to be processed. This move will undoubtedly make data access much faster and facilitate the growing demand for real-time computing.

How will this impact on localisation: well, in 2019 we can expect to see the Edge model used in domain-adapted machine translation systems, and distributed translation workflows that are designed to meet the increasing demand for data distribution in real-time.

Summary

We are on the verge of an explosion in the use of AI. An explosion of the very thing that drives many of the vital cogs within the localisation business. This change will redefine many key roles and bring about the reconfiguration and automation of everyday workflow tasks. The inevitable growth of AI, and the implementation of such things as machine learning, will fundamentally re-shape how companies manage translation workflows; the very engine of their work process. Real-time translations will become the norm, where it is required.

We also predict that changes will happen at a human level; for example, the role of the translator will change from that of translator of raw text to that of editor of huge volumes of high-quality MT-produced text. We also believe this will be a beneficial change, allowing translators to increase their capacity and so increase their income. In 2019, we predict that the overall transformation effected by the advent of AI at all levels of the industry will bring with it an increased velocity of production, an improved efficiency in the delivery of translations, and a reduction in the cost of translating huge volumes of data.

We hope you all have a very successful 2019!

Tony O’Dowd CEO, KantanMT

This article first appeared in Multilingual in January 2019: https://multilingual.com/localization-tech-predictions-2019/

 

As the localization industry strives at a fast pace to integrate Machine Translation into mainstream workflows to increase productivity, reduce cost and gain a competitive advantage, it’s worthwhile taking time to consider which type of Neural MT provides the best results in terms of translation quality and cost.

This is a question that has been occupying our minds here at KantanMT and eBay over the past several months. The fact is, Neural MT comes in many variants – with the different models available yielding remarkably different quality results.

Overview of Neural Network Types

 The main models of Neural MT are:

• Recurrent Neural Networks (RNNs) – these have been designed to recognize sequential characteristics of data and use the detected patterns to predict the next most likely sequence. Training happens in a both forward and backward direction; hence, the use of the descriptor recurrent. RNNs have been the predominant neural network of choice by most MT providers.

Fig 1: Image Courtesy of Jeremy Jordon

• Convolutional Neural Networks (CNNs) – these are the main type of networks used in computer image processing (e.g., facial recognition and image searching) but can also be used for machine translation purposes. The model exploits the 2D-structure of input data. Training process is simplified and CNNs require less computational overhead to compute models.

Fig 1: Image Courtesy of Jeremy Jordon

• Transformer Neural Networks (TNNs) – The predominant approach to MT is based on the recurrent/convolutional neural networks model of connecting the encoder and decoder through an attention mechanism. However, the Transformer Neural Networks model uses only the attention mechanisms aspect (e.g., contextual characteristics of input data). This completely avoids using the recurrence and convolution structures of the other models. This has the effect of simplifying the training process and reducing the computational requirements for TNN modelling.

Fig 3: Image Courtesy of “The Illustrated Transformer” by Jay Alammar

The eBay NMT Experiment

To determine which model yields the best translation outcomes, eBay and KantanMT collaborated and set up a controlled experiment using the KantanMT platform, which supports all three types of Neural Models.

The language arc English => Italian was chosen, and the domain defined as eBay’s Customer Support content. Each Kantan model variant was trained on identical training data sets which consisted of:

• eBay’s in-domain translation memory
• eBay’s glossaries and lists of brand names
• Supplementary KantanLibrary training corpora

The Test Reference Set was created by the eBay MT Linguistic Team by sampling the eBay Translation Memory to mirror its segment length distribution (e.g., 10% short segments, 30% medium and 60% long).

To provide a comprehensive comparison and ranking of the performance of different models, the translation outputs from the following systems were included in our joint experiment:

• Kantan TNN (Transformer Neural Network, customized)
• Kantan CNN (Convolutional Neural Network, customized)
• Kantan RNN (Recurrent Neural Network, customized)
• Bing Translate (Transformer Neural Network, generic)
• Google Translate (Transformer Neural Network, generic)

Human Translation (HT) was also included in this comparison and ranking to determine how neural machine translation outputs compare to translations provided by Professional Translators.

The evaluator was an eBay Italian MT language specialist with domain expertise and experience in ranking and assessing the quality of machine translation outputs.

The following Key Performance Indicators (KPIs) were chosen to determine the comparative fluency and adequacy of each system:

• Fluency = Fluency determines the translation follows common grammatical rules and contains expected word collocation. This KPI measures whether the machine translation segment is formed in the same way a human translation would
• Adequacy = Adequacy measures how much meaning is expressed in the machine translation segment. It measures whether the machine translation segment contains as much of the meaning as if it were translated by a human

Each KPI was rated on a 5-star scale, with 1 star being the lowest rating (i.e., No Fluency) and 5 stars being the highest rating (i.e., Human-Level Fluency).

KantanLQR was used to manage the assessment, randomise and anonymise the Test Reference Set, score the translation outputs, and collate the feedback from the eBay MT linguist.

The Results

Our Conclusions

The Custom Kantan Transformer Neural Network (Kantan TNN) performed the best in terms of Fluency and Adequacy. It outperformed RNNs in terms of Fluency by 9 percentage points (which is statistically significant), and 11 percentage points in terms of Adequacy. While there is still some way to go to achieve near-human-level quality (as depicted by the HT graphs), Transformer Neural Networks provide significant improvements in MT quality in terms of Fluency and Adequacy, and they offer the best-bang-for-your-buck in terms of training time and process simplification.

Since this blog was first published, comparative analysis was also carried out for English=>German, English=>Spanish and English=>French language combinations and in all cases Kantan TNNs out-performed CNNs, RNNs, Google and Bing Translate.

As the localization industry strives at a fast pace to integrate Machine Translation into mainstream workflows to increase productivity, reduce cost and gain a competitive advantage, it’s worthwhile taking time to consider which type of Neural MT provides the best results in terms of translation quality and cost.

This is a question that has been occupying our minds here at KantanMT and eBay over the past several months. The fact is, Neural MT comes in many variants – with the different models available yielding remarkably different quality results.

Overview of Neural Network Types

 The main models of Neural MT are:

• Recurrent Neural Networks (RNNs) – these have been designed to recognize sequential characteristics of data and use the detected patterns to predict the next most likely sequence. Training happens in a both forward and backward direction; hence, the use of the descriptor recurrent. RNNs have been the predominant neural network of choice by most MT providers.

Fig 1: Image Courtesy of Jeremy Jordon

• Convolutional Neural Networks (CNNs) – these are the main type of networks used in computer image processing (e.g., facial recognition and image searching) but can also be used for machine translation purposes. The model exploits the 2D-structure of input data. Training process is simplified and CNNs require less computational overhead to compute models.

Fig 1: Image Courtesy of Jeremy Jordon

• Transformer Neural Networks (TNNs) – The predominant approach to MT is based on the recurrent/convolutional neural networks model of connecting the encoder and decoder through an attention mechanism. However, the Transformer Neural Networks model uses only the attention mechanisms aspect (e.g., contextual characteristics of input data). This completely avoids using the recurrence and convolution structures of the other models. This has the effect of simplifying the training process and reducing the computational requirements for TNN modelling.

Fig 3: Image Courtesy of “The Illustrated Transformer” by Jay Alammar

The eBay NMT Experiment

To determine which model yields the best translation outcomes, eBay and KantanMT collaborated and set up a controlled experiment using the KantanMT platform, which supports all three types of Neural Models.

The language arc English => Italian was chosen, and the domain defined as eBay’s Customer Support content. Each Kantan model variant was trained on identical training data sets which consisted of:

• eBay’s in-domain translation memory
• eBay’s glossaries and lists of brand names
• Supplementary KantanLibrary training corpora

The Test Reference Set was created by the eBay MT Linguistic Team by sampling the eBay Translation Memory to mirror its segment length distribution (e.g., 10% short segments, 30% medium and 60% long).

To provide a comprehensive comparison and ranking of the performance of different models, the translation outputs from the following systems were included in our joint experiment:

• Kantan TNN (Transformer Neural Network, customized)
• Kantan CNN (Convolutional Neural Network, customized)
• Kantan RNN (Recurrent Neural Network, customized)
• Bing Translate (Transformer Neural Network, generic)
• Google Translate (Transformer Neural Network, generic)

Human Translation (HT) was also included in this comparison and ranking to determine how neural machine translation outputs compare to translations provided by Professional Translators.

The evaluator was an eBay Italian MT language specialist with domain expertise and experience in ranking and assessing the quality of machine translation outputs.

The following Key Performance Indicators (KPIs) were chosen to determine the comparative fluency and adequacy of each system:

• Fluency = Fluency determines the translation follows common grammatical rules and contains expected word collocation. This KPI measures whether the machine translation segment is formed in the same way a human translation would
• Adequacy = Adequacy measures how much meaning is expressed in the machine translation segment. It measures whether the machine translation segment contains as much of the meaning as if it were translated by a human

Each KPI was rated on a 5-star scale, with 1 star being the lowest rating (i.e., No Fluency) and 5 stars being the highest rating (i.e., Human-Level Fluency).

KantanLQR was used to manage the assessment, randomise and anonymise the Test Reference Set, score the translation outputs, and collate the feedback from the eBay MT linguist.

The Results

Our Conclusions

The Custom Kantan Transformer Neural Network (Kantan TNN) performed the best in terms of Fluency and Adequacy. It outperformed RNNs in terms of Fluency by 9 percentage points (which is statistically significant), and 11 percentage points in terms of Adequacy. While there is still some way to go to achieve near-human-level quality (as depicted by the HT graphs), Transformer Neural Networks provide significant improvements in MT quality in terms of Fluency and Adequacy, and they offer the best-bang-for-your-buck in terms of training time and process simplification.

Since this blog was first published, comparative analysis has also been carried out for English=>German, English=>Spanish and English=>French language combinations and in all cases Kantan TNNs out-performed CNNs, RNNs, Google and Bing Translate.

To be called a veteran is a dubious honour, and as sure a sign of aging as greying hair. As I sit here looking back on my almost three decades in the translation/localisation industry there is little doubt that I am one of its veterans, something of which I am proud.

We are only days away from the United Nation’s International Translation Day on September 30^th. The UN describes this international day of recognition as:

“International Translation Day is meant as an opportunity to pay tribute to the work of language professionals, which plays an important role in bringing nations together, facilitating dialogue, understanding and cooperation, contributing to development and strengthening world peace and security.”

The date of 30 September was chosen because it is the feast of St. Jerome, the Bible translator who is considered the patron saint of translators. St. Jerome was a priest from North-eastern Italy, who is known mostly for his endeavour of translating most of the Bible into Latin from the Greek manuscripts of the New Testament. He also translated parts of the Hebrew Gospel into Greek.

The UN encourages all member states, international and local organisations and individuals to celebrate the occasion to raise the awareness of the importance professional translation has in this modern world. A world that has been shrunken by the advent of technology that now puts instant, worldwide access in the hands of many of the world’s citizens.

Given this call to celebrate and commemorate, I thought it a good moment to write a blog reflecting on the journey this industry has taken since I first joined a small company called Softrans (subsumed later by the newbie, Berlitz), all those years ago.

On reflection, the average translation office of yore was truly a low-tech, primitive environment compared to today’s office full of super technology and computational power. To give you an idea, here’s a list of technologies we didn’t have, way back then:

• CAT Tools
• Machine Translation (of any calibre)
• Email (although it came soon after)
• The Internet
• The Cloud
• Memory Sticks
• Windows (although it arrived soon after)
• Software development tools such as Catalyst
• HELP authoring tools
• Automated software testing tools
• Workflow systems

The modern localisation professional must look at this list and scratch their heads in wonderment as to how we managed to produce anything. Well, here’s a list of tools we did have at our disposal:

• Fax machines
• 5 ¼ inch 1.2 mb floppy disks (you read that correctly)
• DOS (do you remember C:>?)
• 24-hour delivery services (replaced by the internet!)
• Internal translation teams
• Internal desktop publishing teams
• Internal engineers (with few tools)
• Manual software and help testing
• Manual Software building
• A choice of maybe three DTP packages

And as if that was not bad enough, the machines at our disposal in those days were slower than a wet Sunday. In Softrans, the average translator had use of an IBM 286. A machine that had a whole 10 MB hard drive (that’s not a typo), 128K of RAM (ditto) and a four colour CGA screen. The RAM could be boosted to a massive 640K. Oh, and to move it you needed someone to lift all 80 lbs of it. The operation system was DOS. When you turned it on, instead of a shiny looking Windows GUI you saw C:>. And off you went typing your commands to get started.

But technology does not stand still, and soon we were able to upgrade some of our translators to the IBM 386 model. Yoo-hoo – now we had screaming fast computers. It had a 256K RAM and a huge 40 MB of hard disk space. They also came with the first version of Windows, a revolutionary development in GUI.

They were much fought over by the translators. And I say translators, as the early localisation office had basically two main production departments; the translation department and the desk top publishers. There was also a small project management team and a small engineering group. Project management was still a discipline in its early development, without any of the tools on today’s project manager’s desk.

The engineers were more IT support people than the software engineers that are so needed today. As the industry developed, customers saw that shipping translated documentation and non-localised software and Help did not make much sense. So, they started asking for the Help and software to be translated. Consequently, the engineering team began to grow and evolve in to the engineering discipline that we have today. The engineers then had few of the tools we have today. Software engineering tools such as Catalyst did not exist in any commercial sense. Some of the bigger software development companies had their own internal tools, but it was the mid-1990s before a commercially developed tool – Catalyst – was made available to the engineers in companies like Softrans.

Translation too was done in a rudimentary way. The translator worked in MSWord, translating directly from the English hardcopy source document. They relied on dictionaries and self-created glossaries to keep them right. They would create the glossaries by laboriously identifying key terms in the English and then, in a table, translate the term in to their local language. Each translation team had an editor whose job was to ensure continuity of terminology across the project. The editor also had the task of keeping glossaries updated when the reviewed copy came back from the customer. Very few customers had their own glossaries. Indeed, most customers did not have a translation/localisation department. As a consequence, their translation would be managed by groups as diverse as marketing, internal training, or simply an individual tasked to run the project.

Because the only means of external communication was by fax, post or overnight courier it meant that translation companies had to keep a core team of translators in-house. Softrans, for example, had French, Italian, German and Spanish (FIGS) translators. It was like a mini UN. So too with DTP. Today, in the era of the home office a translator or DTP person can be in Timbuctoo but so long as they have access to the internet they can be usefully productive. Not then, everything had to be housed under one roof, which caused high overheads for companies as they needed to supply desk space, heating, lighting etc.

Today, jobs are delivered by the pressing of a key and off zips the project. Or, the work is done on the customer’s network without the files every having to moved from their control and monitoring. No project is too big as the Cloud offers an endless amount of storage space. In the Softrans’ day a project was delivered on a mass of 5.25-inch disks. Each had a capacity of 1.2 MB. So, a large project took a very large number of disks. These were usually shipped by overnight courier. On occasions, the disks got corrupted in transit meaning a new set would have to be sent, so causing a delay of days on something that was badly needed by companies with product ready to ship to meet a particularly tight marketing window.

The first time I realised that the industry was developing a tool to help speed up translation was when Softrans was chosen to trial a new product called a Computer Aided Translation tool. That tool was called Trados. It was about 1994 when I first came across it. I remember translators falling in to two categories; those who feared that their jobs would become redundant, and those who were cynical about the possibility of technology making their job easier. But as it was introduced over time, both groups began to see the power of Trados. It very soon became an essential part of the translation process. No translator today would even consider translating a large project without using a CAT tool. Even more so, no customer would accept you translating from scratch every time and charging them for new words in every project!

And here I am still loitering with intent inside the localisation industry. This time in the marketing department (previously, I managed translators, DTP groups and engineers) of one of the most innovative technology groups in the industry KantanMT.com. Kantan is at the cutting edge of machine translation technology. Kantan started with Statistical Machine Translation, before riding the wave that is Neural Machine Translation. MT is a technology that has an amazing exponential growth. The advent of the Cloud, super-computers and huge volumes of translated data has fed the industry like petrol to a fire. And, as with all ‘new’ technologies – for example CAT – the cynics are out there saying it cannot be done. The translators are decrying MT as a sorcery that is going to disappear their industry. But I am too long in this industry not to realise that MT is a technology that the translation/localisation needs to lift in to the next level of development. The industry profile is not longer one of small translation companies like Softrans with its $1 million turnover, but companies that are now valued in the billions.

The business world has contracted, and every major company sees its market as global. They need to overcome the challenges that language brings, and they want to do this in the speediest, most cost-effective way. Machine translation allows this to happen. The market will not allow MT to disappear. You only need look at the companies now committed to MT – Amazon, Google, eBay, Facebook; to name but a few, to see that MT is fast becoming the norm. The translation paradigm is shifting fast. But translators will not disappear; their jobs will just take on a different dimension. They will become more editors than translators. MT will become an integral part of the standard workflow. It will become part of the process just as much as CAT did in the past. We are seeing exciting new developments in translation/localisation. It is difficult to predict what is coming around the corner for the industry; but unlike the IBM 286, IBM 386 and floppy disks, it is not going to go away!

Enjoy International Translation Day!

Aidan Collins is a language industry veteran. He works in the marketing department at KantanMT.

The author of this book is Founder and Executive Chairman of the World Economic Forum. That is the forum where the movers and shakers of the world meet at the Swiss alpine location of Davos every year to discuss geo-global trends across a range of political, social and economic areas. In this slim book Schwab seeks to predict how impending technological changes will impact on our life; socio, political and economic.

The author argues that the fusing of the political, physical, digital and biological worlds will have a transformative impact on all facets of human existence. This will range from the way we live our lives, the manner in which we will work, the reconfiguration of economic models, the products we sell and even, how long we will choose to live.

The author outlines the drivers of this revolution and cautions the business readers to “get on board” as we are “already reaching an inflection point in [technological] development as they build on and amplify each other in a fusion of technologies”. In an appendix he conveniently lists what are termed 23 “Deep Shift” technologies; those must likely to impact the way we live. This list is headed up by implantable technologies, courtesy of nanotechnological developments; the widespread use of digital currencies such as Bitcoin, driven by Blockchain developments; and the surreal prediction that neurotechnology will allow for humans to have artificial memories implanted in their brains.

Of course, all of the above can not happen without society acquiescing (or one would hope so). The author does discuss how the revolution will throw up challenges on all fronts as to the ethics, morality and legality of many of the putative changes. He warns that society will be in a state of rapid change as the fusion of technologies will create an exponential growth that will make this revolution a much shorter and deeper period of impact that the societal revolutions than mankind has ever before experienced.

Throughout the book Schwab posits a benign view on the power of the Fourth Industrial Revolution. He argues that its power can be used for good if harnessed by careful, democratic control by “good leaders” and “decision makers”. Of course the obverse is also a possibility, although the author does not discuss that likelihood.

The challenge to societies’ leaders will be to learn how to harness for good the changes, but controlling and curtailing them when they venture in to possible unethical or illegal terrain  (as experienced  already with the whole hoovering up of people’s data for selling on to others).

Schwab sets this revolution in a historical context referring to the previous societal upheavals such as the industrial revolution. Each revolution did a lot to transform society, and not always for the better

The first revolution was 1760-1840 and was triggered by the construction of railroads and development of steam power. It heralded the beginning of the mechanical age. The second started in the late 19th century and was the beginning of mass production, factory workplaces, the production line and mass employment in often poor conditions. The third, and most recent, was caused by the Digital Age. The development of semiconductors, mainframe computing and the emergence of personal computing harnessing the internet.

The author concedes that there are those who argue that what he heralds as a fourth revolution is no more than the outworking of a more advanced part of the third industrial revolution. Schwab holds his ground and says the fourth industrial revolution began in 2000 when technologies began to converge, Artificial Intelligence became a reality and robotics made huge advancements. He also argues that the ubiquity of small, integrated technology available to all at low cost, plus the conquering of the language barrier through the use of machine translation, has made the global market available to all who choose to exploit it.

Disappointingly, for a book that deals with such a diverse range of concepts and technologies, it lacks an index. This was probably the result of the book having been the product of the compiling of a series of papers originally written for other forums. The book is heavy on jargon and management speak. It is, nonetheless,

a slim volume that is fairly accesible to the average reader. Schwab leaves you in no doubt that the Fourth Industrial Revolution is capable of creating a dystopian world of cloud power, AI, implanted brains and robots. The question – which Schwab hints at rather than elucidates on, is whether we as humans should meekly adapt to all technology, or whether as a society we say: “hold on; thus far and no further, thank you.”

Aidan Collins is a language industry veteran. He works in the marketing department at KantanMT.