ModernMT: A Closer Look At An Emerging Enterprise MT Powerhouse

 As one observes the continuing evolution of MT use in the professional translation industry, we see that we have reached a point where we have some useful insights about producing successful outcomes in our use of MT. From my perspective as a long-term observer and expert analyst of enterprise MT use, some of these include:

• Adaptation and customization of a generic MT engine done with expertise generally produces a better outcome than simply using a generic public MT system. 
• Working with enhanced baseline engines built by experts is likely to produce better outcomes than dabbling with Open Source options with limited expertise. While it has gotten easier to produce MT systems with open-source platforms, real expertise requires long-term exposure and repeated experimentation. 
• The algorithms underlying Neural MT have become largely commoditized and there is little advantage gained by jumping from one NMT platform to another.
• More data is ONLY better if it is clean, relevant, and applicable to the enterprise use case in focus. It can be said today that (training) data often matters more than the algorithms used, but data quality and organization is a critical factor for creating successful outcomes.
• A large majority of translators still view MT with great skepticism and see it as marginally useful, mostly because of repeated exposure to incompetently deployed MT systems that are used to reduce translator compensation. Getting active and enthusiastic translator buy-in continues to be a challenge for most MT developers and getting this approval is a clear indicator of superior expertise.
• Attempts to compare different MT systems are largely unsuccessful or misleading, as they are typically based on irrelevant test data or draw conclusions based on very small samples.
• A large number of enterprise use cases are limited by scarce training data resources and thus adaptation and customization attempts have limited success.

I have been skeptical of the validity of many of the comparisons we see of MT systems produced by LSPs and "independent" evaluators nowadays, because of the questionable evaluation methodologies used. The evaluators often produce nice graphics but just as often produce misleading results that need further investigation. However, these comparative evaluations of different MT systems can still be useful to get a rough idea of the performance of generic systems of these MT vendors. Over the last few years ModernMT has been consistently showing up amongst the top-performing MT systems in many different evaluations, and thus I decided to sit down with the ModernMT team to better understand their technology and product philosophy and understand what might be driving this consistent performance advantage. The level of transparency and forthcoming nature of the responses from the ModernMT team was refreshing in contrast to other conversations I have had with other MT developers.

The MT journey here began over 10 years ago with Moses and Statistical MT, but unlike most other long-term MT initiatives I know of, this effort was very translator-centric right from its inception. The system was used heavily by translators who worked for Translated and the MT systems were continually adapted and modified to meet the needs of production translators. This is a central design intention and it is important to not gloss over this, as this is the ONLY MT initiative I know of where Translator Acceptance is used as the primary criterion on an ongoing basis, in determining whether MT should be used for production work or not. The operations managers will simply not use MT if it does not add value to the production process and causes translator discontent.  Over many years the ongoing collaboration with translators at ModernMT has triggered MT system and process development changes to reach the current status quo, where the MT value-add and efficiency is clear to all the stakeholders. The long-term collaboration between translators and MT developers, and resulting system and process modifications are a key reason why ModernMT does so well in both generic MT system comparisons, and especially in adapted/customized MT comparisons.

Thus, translators who actively use the ModernMT platform do so most often through MateCat, an open-source CAT tool that ties together MyMemory (a large free-access shared TM repository with around 50 billion words in it) together with ModernMT or other MT platforms. MT is presented to translators as an alternative to TM on a routine basis, and corrections are dynamically and systematically used to drive continuous improvements in the ModernMT engines. Trados and other CAT tools are also able to seamlessly connect to the ModernMT back-end but these systems may see less immediate improvements in the MT output quality. However, this has not stopped ~25,000 downloads of the ModernMT plugin for Trados on the SDL Appstore. Translators who do production work for Translated are often given a choice of using Google instead of ModernMT but most have learned that ModernMT output improves rapidly from corrective feedback and that collaborative input is also easier, and thus tend to prefer it as shown in the surveys below. Over the years the ModernMT product evolution has been driven by changes to identify and reduce post-editing effort rather than optimizing BLEU scores as most others have done. 

In contrast to most MTPE experiences, the individual translator experience here is characterized by the following:

• A close and symbiotic relationship between a relevant translation memory and MT, even at the translator UX level
• An MT system that is constantly updated and can potentially improve with every single interaction and unit of corrective feedback
• Immediate project startup possibilities as no batch MT training process is necessary
• Translator control over all steering data used in a project means very straightforward control over terminology and term consistency, mirroring the latest TMs and linguistic preferences
• Corrective feedback given to the MT system is dynamic and continuous and can have an immediate impact on the next sentence produced by the MT system
• One of very few MT systems available today that can provide a context-sensitive translation 
• Measurable and palpable reduction in post-editing effort and translator UX compared to other MT platforms
• Continuing free access to the CAT tool needed to integrate MT with TM, and interact proactively with MT with the option to use other highly regarded CAT tools if needed. 

 

Memory here refers to user input data TM and glossaries to tune the generic system to the needs of the current translation task

Instance-Based Adaptation

ModernMT describes itself as an "Instance-Based Adaptive MT" platform. This means that it can start adapting and tuning the MT output to the customer subject domain immediately, without a batch customization phase. There is no long-running (hours/days/weeks) data preparation and pre-training process needed upfront. There is also no need to wait and gather a sufficient volume of corrective feedback to update and improve the MT engine on an ongoing basis. It is learning all the time. 

Rapid adaptation to customer-unique language and terminology is perhaps the single most critical requirement for a global enterprise, and thus this is an optimal design for enterprises that works optimally with their specialized and unique content. This is also true for LSPs too, for that matter. ModernMT can adapt the MT system with as little as a single sentence, though the results are better if more data is provided. The team told me that 100K words (10-12,000 sentences)  would generally produce consistently good results that are superior to any generic engine. The long-term impact of this close collaboration with translators who provide ongoing corrections, feedback on critical requirements to improve efficiency and process workflow, and careful acquisition of the right kind of data, results in the kind of relative performance rankings that ModernMT now regularly sees as a matter of course. One might even go so far as to say that they have built a sustainable competitive advantage. 

I have always felt that a properly designed Man-Machine collaboration would very likely outperform an MT design approach that relies entirely on algorithms and/or data alone. We can see this is true from the comparative results of the large public MT portals who probably have 100X or more of the resources and budget that ModernMT does. The understanding of the translation task and resulting directives that ongoing translator feedback brings to the table is an ingredient that most current MT systems lack. Gary Marcus and other AI experts have been vocal in pointing out that machine learning and data alone is not the best way forward and more human steering and symbolic knowledge is needed for better outcomes.  

   

Special Features

ModernMT is a context-aware machine translation product that learns from user corrections. There has recently been growing interest in the MT research community to bring a greater degree of contextual awareness to MT systems and ModernMT has also been investigating implementing capabilities around doing this. The current production version has an implementation of this already, and this feature continues to evolve in speed, efficiency, and capability.

The ModernMT Context Analyzer analyzes an entire document text to be translated in milliseconds before producing a translation. This analysis seeks out and identifies the distinctive terminology and intrinsic style of the document. This information is then used to automatically select the most suitable private translation memories loaded by the user for that particular document. This results in the engine selecting the translation memory inventory that best reflects the right terminology and writing style. It is precisely this inventory that the MT engine leverages to customize the output in real-time, for each and every sentence of the document.

As translators at Translated working with ModernMT regularly have the ability to compare the MT output with that of Google Translate, the developers monitor translator preferences on an ongoing basis. This ensures that translators are always working with the MT output that they find most useful and that developers understand when their own engines need to be improved or enhanced. The following charts are based on feedback from translators during production work and show a very definite preference for the rapidly improving ModernMT engine output. This preference is seen in internal translator assessments working in production mode rather than just a selective test set, and this has also been confirmed by independent third-party assessments with both automated scores and human evaluations. They all consistently show that ModernMT customizations regularly outperform most others in independent comparative evaluations. The forces driving this superior performance are the result of design philosophy and long-term man-machine collaboration that cannot be easily replicated by others.

 

Recent comparative assessments done by independent third parties also confirm this preference using different evaluation methods that include both human and automated metrics as shown below. It is not unreasonable to presume that this performance advantage will remain intact for at least the short term.

Data Privacy

In response to a question on data privacy, Davide Caroselli, VP of Product, ModernMT responded: "Any content sent to ModernMT, whether a “TMX” memory or an MTPE correction from a professional translator, is saved in the user’s private data area. In fact, only you will be able to access your resources and make ModernMT adjust to them; in no way will another user be able to utilize that same inventory for his/her system, nor will ModernMT itself be able to use those contents, other than to exclusively offer your personalized translation service.

In addition, ModernMT uses state-of-the-art encryption technologies to provide its cloud services. Our data centers, employee processes and office operations are ISO 27001:2013 certified." 

On-Premise Capabilities

While the bulk of the current ModernMT customer base works with the secure cloud deployment, the team at ModernMT has also defined a range of on-premise deployment capabilities for those enterprises that need the security, control, and assured data privacy needs that characterizes some National Security, Financial, Legal, and Healthcare/Pharma industry requirements. The open-source foundations of much of the ModernMT infrastructure should make it particularly interesting to US Government Intelligence and Law-Enforcement agencies seeking large-scale multilingual data processing capabilities for eDiscovery and Social Media Surveillance applications.  

Given that ModernMT is a continuous learning MT platform that learns with each correction, dynamically, there is a requirement for more GPU infrastructure than some other on-premise solutions in the market. However, there is a strong focus on computational efficiency to minimize the IT footprint needed to deploy it on-premise, and based on information provided to me, their capabilities are quite similar to competitive alternatives both in terms of hardware requirements and software pricing. Hardware costs are linked to throughput expectations with more hardware required for high throughput requirements. As with most machine learning-intensive capabilities, only enterprises with competent IT teams could undertake this as an internal deployment, and most LSPs and localization departments will see a lower total cost of ownership with the cloud deployment. 

Enterprise Readiness  

As ModernMT has evolved from the localization world it is already optimized for MT use cases where there is a significant need for a machine-first human optimized approach. More and more we see this model as being a preferred approach for the exploding volumes of localization content.  The Localization Use Case is possibly the most challenging MT use case out there, as it requires very high-quality initial output that translators are willing to work with where it can be proven that the MT enhances productivity and efficiency. Localization use cases demand the highest quality MT output from the outset compared to eDiscovery,  social media surveillance, eCommerce,  customer service & support use cases which are all more tolerant of lower MT output quality on much larger volumes of data. Very few MT developers have had success with the high-quality and rapid responsiveness needs of the localization use case and many have tried and failed. This is why LSP adoption of MT is so low. ModernMT's success with the challenging localization use case, however, positions them very well for other MT use cases as their growing success with these other use cases proves.

The ASTW case study illustrates the success of ModernMT in Intellectual Property (Patents) and Life Science focused translations, where the ease of customization for complex terminology and morphology, the ability to learn continuously and quickly from corrective feedback, and superior MTPE experience compared to other MT solutions has quickly made it a preferred solution. 

"ModernMT is currently our favorite MT engine, especially in patent translations and in the Life Science sector, because it proves reliable, efficient, qualitatively better than its competitors, easily customizable and advantageous in terms of cost."

Domenico Lombardini, CEO ASTW

We see that eCommerce giants understand the positive impact of translating huge volumes of catalog and user-generated CX content has on driving international revenue growth with the examples of eBay, Amazon, and Alibaba. ModernMT is now the MT engine driving the multilingual expansion of Airbnb web content and is translating many billions of words a month for them. User-generated content influences future customers, and there is great value in translating this content to drive and grow international business. Interestingly ModernMT began this initiative with almost no translation memory and had to perform specialized heuristic analysis on Airbnb content to build the training material.    

ModernMT has reached this point with very little investment in sales and marketing infrastructure. As this builds out and expands I will be surprised if ModernMT does not continue to expand and grow its enterprise presence, as enterprise buyers begin to understand that a tightly integrated man-machine collaborative platform that is continuously learning, is key to creating successful MT outcomes. I am aware that many other high-profile enterprise conversations are underway, and I expect that most enterprise buyers who evaluate the ModernMT platform will very likely find it is a preferred, cost-efficient way to implement large-scale MT solutions in a way that dramatically raises the likelihood of success. 

Future Directions

Davide also mentioned to me that his team is very connected to the AI community in Italy, and have been experimenting with GPT-3 and BERT, and will continue to do so until clear value-added applications that support and enhance their MT product emerge. ModernMT has a close relationship with Pi Campus and thus has regular interaction with luminaries in the AI community e.g.  Lukasz Kaiser who will be speaking about improvements in the Transformer architecture later this month.

The team also showed me demos of complex video content that had ModernMT-based automated dubbing from English to Italian injected into it. Apparently, Italy is one of the largest dubbing markets in the world. Who knew? Since my wife speaks Italian, I showed her some National Geographic content on geology, filled with complex terminology and scientific subject matter that she was shocked to find out had been done completely without human modification. The Translated team is exploring Speech Translation and I expect that they will be quality leaders here too.

ModernMT will continue to expand its connectivity to other translation and content management infrastructure to make it easier to get translation-worthy data in and out of their environment. They also continue to explore ways to make the ModernMT continuous training infrastructure more computationally efficient so that it can be more easily deployed on smaller footprint hardware. 

I expect we will see more and more of ModernMT on the enterprise MT stage from now on, as buyers realize that this is a significantly improved next-generation MT solution that is more likely to produce successful outcomes in digital transformation-related enterprise use scenarios. The ModernMT approach reduces the uncertainty that is so common with most MT-related initiatives and does it so seamlessly that most would not realize how sophisticated the underlying technology is until they attempt to replicate the functionality.

On a completely different note, I participated some months ago in responding to a question posed by  Luca Di Biase, the Imminent Research Director. He posed this same question to many luminaries in the translation industry, and also to me. The question has already triggered several discussions on Twitter.

“Is language a technology or a culture?”  

My response was as follows, but I think you may find the many other responses more interesting and complete if you go to this link or look at some of the other Twitter comments.

It is neither. Language is a means of communication and an information-sharing protocol that employs sounds, symbols, and gestures. Language can sometimes use technology to enable amplification, extend the reach of messages, and accelerate information and knowledge sharing. Language can create a culture when shared with(in) a group and used with well-understood protocols and norms. Intercultural communication can also mean cross species, e.g., when communicating with dogs and horses.

Translated's Research Center has just released the Imminent publication which has a distinctive style coupled with interesting content, that I think most in the language industry would find compelling and worth a close look.  

The Quest for Human Parity Machine Translation

The Challenge of Defining Translation Quality 

The subject of  "translation quality" has always been a challenging communication issue for the translation industry. It is particularly difficult to explain this concept in a straightforward way to an industry outsider or a customer whose primary focus is building business momentum in international markets, and who is not familiar with localization industry translation-quality-speak.  Nowadays they tend to focus on creating and managing the dynamic and ever-changing content that enhances a global customer's digital journey, rather than the static content that is the more typical focus of localization managers. Thus, the conventional way in which translation quality is discussed by LSPs is not very useful to these customers. Since every LSP claims to deliver the "best quality " or "high quality" translations", it is difficult for these buyers to tell the difference in this service aspect from one service provider to another. The quality claim between vendors thus essentially cancels out. 

These customers also differ in other ways. They need larger volumes of content to be translated rapidly at the lowest cost possible, but yet at a quality level that is useful to the customer in digital interactions with the enterprise. For millions of digital interactions with enterprise content, the linguistic perfection of translations is not a meaningful and achievable goal given the volume, short shelf-life, and instant turnaround expectations a digital customer will have.  

As industry observer and critic Luigi Muzii describes it:

"Because of the intricacies related to the intrinsic nature of languages, objective measurement of translation quality has always been a much researched and debated topic that has borne very little fruit. The notion of understood quality level remains unsolved, together with any kind of generally accepted and clearly understood quality assessment and measurement."

The industry response to this need for a better definition of translation quality is deeply colored by the localization mindset and thus we see the emergence of approaches like the Dynamic Quality Framework (DQF). Many critics consider it too cumbersome and detailed to implement in translating modern fast-flowing content streams needed for superior digital experience. While DQF can be useful in some limited localization use-case scenarios, it will surely confound and frustrate enterprise managers who are more focused on digital transformation imperatives.  The ability to rapidly handle and translate large volumes of DX-relevant content cost-effectively is increasingly a higher priority and needs a new and different view on monitoring quality. The quality of the translation does matter in delivering superior DX but has a lower priority than speed, cost, and digital agility.

While machines do most of the translation on the planet today, this does not mean that there is no role for higher value-added human translation (increasingly supported by CAT tools). If the content is a critical and high-impact communication, most of us understand that human oversight is critical for success in the business mission. And if translation involves finesse, nuance, and high art, it is probably best to leave the "translating" computers completely out of the picture. 

However, in this age of digitally-driven business transformation and momentum, competent MT solutions are essential to the enterprise's mission. Increasingly, more and more content is translated and presented to target customers without EVER going through any post-editing modification. The business value of the translation is often defined by its utility to the consumer in a digital journey, basic understandability, availability-on-demand, and the overall CX impact, rather than linguistic perfection. Generally, useable accuracy and timely delivery matter more than perfect grammar and fluency. The phrase "good enough" is used both disparagingly, and as a positive attribute, for the translation output that is useful to a customer even in a less than “perfect” state.

So we have a situation today where the term translation quality is often meaningless even in "human translation" because it cannot be described to an inexperienced buyer of translation services (or regular human beings) in a clear, objective, and consistently measurable way. Comparing different human translation works of the same source material is often an exercise in frustration or subjective preference at best. Every sentence can have multiple correct, accurate translations, so how do we determine what is the best translation?  Since every LSP in the industry claims to provide the "best quality", such a claim is useless to a buyer who does not wish to wade through discussions on error counts, error categories, and error monitoring dashboards that are sometimes used to illustrate translation quality.

Defining Machine Translation Output Quality

The MT development community has also had difficulty establishing a meaningful and widely useful comparative measurement for translation quality. Fortunately, they had assistance from the National Institute of Standards & Technology (NIST) and developed a methodology to compare the translation quality of multiple competing MT systems under carefully controlled evaluation protocols. The NIST used a variant of BLEU scores and other measures of precision, recall, adequacy, and fluency to compare different MT systems rapidly in a standardized and transparent manner.  Their efforts probably helped to establish BLEU as a preferred scoring methodology to rate both evolving and different competing MT systems. 

The competitive evaluation approach works when multiple systems are compared under carefully monitored test protocols, but becomes less useful when an individual developer announces "huge improvements" in BLEU scores as it is easy to make extravagant claims of improvement that are not easily validated. Some independent evaluations used by many today, provide comparisons where several systems may have actually trained on the test sets - this is the equivalent of giving a student the exam with the answers before a formal test. This makes some publicly available comparisons done by independent parties somewhat questionable and misleading.  Other reference Test-set-based measurements like hLepor, Meteor, chrF, Rouge, and others are also plagued by similar problems. These automated measurements are all useful, but unreliable indicators of absolute quality.

Best practices today suggest that a combination of multiple automated measures needs to be used together with human assessments of MT output to really get a handle on the relative quality of different MT systems. Again, this quickly gets messy as soon as we start asking annoying questions like:

• What kind of content are we testing?
• Are we sure that these MT systems have not trained on the test data? 
• What kind of translators is evaluating the different sets of MT output?  
• How do these evaluators determine what is better and worse when comparing different correct translations?
• How many sentences are needed to make a meaningful assessment and draw accurate conclusions when comparing multiple MT systems' performance on the same source material?

So, we see that conducting an accurate evaluation is difficult, and messy, and it is easy to draw wrong conclusions stemming from easy-to-make errors in the evaluation process.

However, in the last few years, several MT developers have claimed to produce MT systems that have achieved human parity. This has been especially true with the advent of neural machine translation. These claims are useful for creating a publicity buzz among ignorant journalists and fear amongst some translators, but usually disappoint anybody who looks more closely.

I have been especially vocal in challenging the first of these broad human parity claims as seen here: The Google Neural Machine Translation Marketing Deception. The challenge is very specific and related to some specific choices in the research approach and how the supporting data was presented.  A few years later Microsoft claimed they reached human parity on a much narrower focus with their Chinese to English News system but also said: 

Achieving human parity for machine translation is an important milestone of machine translation research. However, the idea of computers achieving human quality level is generally considered unattainable and triggers negative reactions from the research community and end-users alike. This is understandable, as previous similar announcements have turned out to be overly optimistic. 

The goal of achieving human parity has become a way to say that MT systems have gotten significantly better as this Microsoft communication shows. I too was also involved with the SDL claim of having "cracked Russian", which is yet another broad claim stating that human parity has been reached😧. 

Many, who are less skeptical than I am, will interpret that an MT engine that claims to have achieved human parity can ostensibly produce translations of equal quality to those produced by a human translator. This can indeed be true on a small subset of carefully selected test material, but alas we find that it is not usually true in general for much of what we submit with high expectations to these allegedly human parity MT engines. This is the unfortunate history of MT: over-promising and underdelivering. MT promises are so often empty promises 😏. 

While many in the translation and research communities feel a certain amount of outrage over these exaggerated claims (based on MT output they see in the results of their own independent tests) it is useful to understand what supporting documentation is used to make these claims. 

We should understand that at least among some MT experts there is no deliberate intent to deceive, and it is possible to do these evaluations with enough rigor and competence to make a reasonable claim of breakthrough progress, even if it falls short of the blessed state of human parity. 

There are basically two definitions of human parity generally used to make this claim.

Definition 1.If a bilingual human judges the quality of a candidate translation produced by a human to be equivalent to one produced by a machine, then the machine has achieved human parity.

Definition 2.If there is no statistically significant difference between human quality scores for a test set of candidate translations from a machine translation system and the scores for the corresponding human translations then the machine has achieved human parity

Again the devil is in the details, as the data and the people used in making the determination can vary quite dramatically. There are (50?) shades of grey rather than black-and-white facts in most cases.  The most challenging issue is that human judges and evaluators are at the heart of the assessment process. These evaluators can vary in competence and expertise and can range from bilingual subject matter experts and professionals to low-cost crowdsourced workers who earn pennies per evaluation. The other big problem is the messy, inconsistent, irrelevant, and biased data underlying the assessments.

Ensuring objective, consistent human evaluation is necessary but difficult to do consistently on the required continuous and ongoing basis. If the underlying data used in an evaluation are fuzzy and unclear we actually move to obfuscation and confusion rather than clarity. This can be the scientific equivalent of fake news. MT engines evolve over time and the better the feedback, the faster the evolution if developers know how to use this feedback to drive continuous improvements.  

Again, as Luigi Muzii states:

The problem with human evaluation is bias. The red-pen syndrome.

Indeed, human evaluation of machine translation is known for being expensive, time-consuming, and often biased, and yet it is supposed to overcome the drawbacks introduced by the limited accuracy and approximation of automatic evaluation. ... Indeed, translation quality metrics have become more and more complex and overly detailed, and always seem to be based on the error-catching [and categorization] approach that has proved costly and unreliable thus far.  

 

Useful Issues to Understand 

While the parity claims can be roughly true for a small sample of evaluated sentences, it is difficult to extrapolate parity to a broader range of content because it is simply not possible to do machine translation output evaluation on an MT scale (millions of sentences). Some of the same questions that obfuscate quality discussions with human translation services also apply to MT. If we cannot define what a "good translation" is for a human, how is it possible to do this for a mindless, common-sense-free machine, where instruction and direction need to be explicit and clear?  

Here are some validation and claim verification questions that can help an observer to understand the extent to which parity has been reached or also expose deceptive marketing spin that may motivate the claims.

What was the test data used in the assessments? 

MT systems are often tested and scored on news domain data which is most plentiful. This may not correlate well with system performance on the typical content in the global enterprise content domain. A broad range of different types of content needs to be included to make claims as extravagant as having reached human parity. 

What is the quality of the reference test set?

In some cases, researchers found that the test sets had been translated, and then back-translated with MTPE into the original source language. This could mean the content of the test sets would be simplified from a linguistic perspective, and thus easier to machine translate. Ideally, only expert human-created test sets should be used and should contain original source material, and should not be translated data from another language.

Who produced the reference human translations being used and compared?

The reference translations against which all judgments will be made should be "good" translations. Easily said but not so easily done. If competent humans are creating the source test set sentences, the test process will be expensive. Thus, it is often more financially expedient to use MT or cheap translators to produce the test material.  This can cause a positive bias for widely used MT systems like Google Translate. 

How much data was used in the test to make the claim? 

Often human assessments are done with as little as 50 sentences, and automated scoring is rarely done with more than 2,000 sentences. Thus, drawing conclusions on how any MT system will handle the next million sentences it will process is risky, and likely to be overly optimistic. For example, when an MT developer says that over 90% of the system’s output has been labeled as a human translation by professional translators, they may be looking at a sample of only 100 or so sentences. To then claim that human parity has been reached is perhaps overreaching.  

Who is making the judgments and what are their credentials?

It is usually cost-prohibitive to use expert professional translators to make the judgments and thus evaluators are often acquired on crowdsourcing platforms where evaluator and translator competence is not easily ascertained. 

It can be seen that doing an evaluation properly would be a significant and expensive task, and MT developers have to do this continuously while building the system. The process needs to be efficient, fast, and consistent. It is often only possible to do such careful tests on the most mission-critical projects and is not realistic to follow all these rigorous protocols for typical low ROI enterprise projects. This is why BLEU and other "imperfect" automated quality scores are so widely used. They provide the developers with continuous feedback in a fast and cost-efficient manner if they are done with care and rigor. Recently there has been much discussion about testing on documents to assess understanding of context rather than just sentences. This will add complexity, cost, and difficulty to an already difficult evaluation process, and IMO will yield very small incremental benefits in evaluative and predictive accuracy. There is a need to balance improved process recommendations with cost, and the benefit from improved predictability. 

The Academic Response

Recently, several academic researchers provided some feedback on their examination of these MT at human parity claims. The study is called “A Set of Recommendations for Assessing Human–Machine Parity in Language Translation” and is worth a look to see the many ways in which evaluations can go wrong. The study showed that human evaluation of MT quality depends on three factors: “the choice of raters, the availability of linguistic context, and the creation of reference translations.”

Some findings from this report in summary:

“Professional translators showed a significant preference for human translation, while non-expert [crowdsourced] raters did not”.

“Human evaluation methods which are currently considered best practice fail to reveal errors in the output of strong NMT systems”

The authors recommend the following design changes to MT developers in their evaluation process:

• Appoint professional translators as raters
• Evaluate documents, not sentences
• Evaluate fluency on top of adequacy
• Do not heavily edit reference translations for fluency
• Use original source texts

Most developers would say that implementing all these recommendations would make the evaluation process prohibitively expensive and slow. The researchers here do agree and welcome further studies into “alternative evaluation protocols that can demonstrate their validity at a lower cost.” Process changes need to be practical and reasonably possible, and we see that there is a need to balance improved process benefits with, cost, and improved predictability benefits.  

What Would Human Parity MT Look Like?

MT developers should refrain from making claims of achieving human parity until there is clear evidence that this is happening at scale. Most current claims on achieving parity are based on laughably small samples of 100 or 200 sentences. I think it would be useful to the user community at large that MT developers restrain from making these claims until they can show all of the following:

• 90% or more of a large sample (>100,000 or even 1M sentences) that are accurate and fluent and truly look like they were translated by a competent human
• Catch obvious errors in the source and possibly even correct these before attempting to translate 
• Handle variations in the source with consistency and dexterity
• Have at least some nominal amount of contextual referential capability

Note that these are things we would expect without question from an average translator. So why not from the super-duper AI machine? 

Until we reach the point where all of the above is true, it would be useful to CLEARLY state the boundary limits of the claim with key parameters underlying the claim. Such as:

• How large the test set was (e.g. 90% of 50 sentences where parity was achieved) 
• Descriptions of what kind of source material was tested
• How varied the test material was: sentences, paragraphs, phrases, etc...
• Who judged, scored, and compared the translations

 For example, if we saw an MT developer state a parity claim as follows perhaps:

We found that a sample of 45/50 original human sourced sentences translated by the new MT system were judged by a team of three crowdsourced translator/raters as indistinguishable from the translations produced by two professional human translators.  Based on this data, we claim the system has achieved "limited human parity".

Until the minimum set of capabilities is shown at the MT scale (>100,000 or even 1M sentences) we should tell MT developers to STFU and give us the claim parameters in a simple, clear, summarized way, so that we can weigh the reality of the data versus the claim for ourselves.  

I am also skeptical that we will achieve human parity by 2029 as some "singularity" enthusiasts have been saying for over a decade. 

 

"There is not the slightest reason to believe in a coming singularity. Sheer processing power [and big data] is not pixie dust that magically solves all your problems." 

Steven Pinker 

Elsewhere, Pinker also says:

"… I’m skeptical, though, about science-fiction scenarios played out in the virtual reality of our imaginations. The imagined futures of the past have all been confounded by boring details: exponential costs, unforeseen technical complications, and insuperable moral and political roadblocks. It remains to be seen how far artificial intelligence and robotics will penetrate into the workforce. (Driving a car is technologically far easier than unloading a dishwasher, running an errand, or changing a baby.) Given the tradeoffs and impediments in every other area of technological development, the best guess is: much farther than it has so far, but not nearly so far as to render humans obsolete."

Recently some in the Singularity community have admitted that "language is hard" as you can see in this attempt to explain why AI has not mastered translation yet.

Michael Housman, a faculty member of Singularity University, explained that the ideal scenario for machine learning and artificial intelligence is something with fixed rules and a clear-cut measure of success or failure. He named chess as an obvious example and noted machines were able to beat the best human Go player. This happened faster than anyone anticipated because of the game’s very clear rules and limited set of moves.

Housman elaborated, “Language is almost the opposite of that. There aren’t as clearly-cut and defined rules. The conversation can go in an infinite number of different directions. And then of course, you need labeled data. You need to tell the machine to do it right or wrong.”

Housman noted that it’s inherently difficult to assign these informative labels. “Two translators won’t even agree on whether it was translated properly or not,” he said. “Language is kind of the wild west, in terms of data.”

Perhaps, we need to admit that human parity MT at scale is not a meaningful or achievable goal. If it is not possible to have a super-competent human translator capable of translating anything and everything with equal ease, why do we presume a machine could?

Perhaps what we really need is an MT platform that can rapidly evolve in quality with specialized human feedback. Post-editing (MTPE) today is generally NOT a positive experience for most translators. But human interaction with the machine can be a significantly better and positive experience.  Developing interactive and highly responsive MT systems that can assist, learn, and improve the humdrum elements of translation tasks instantaneously might be a better research focus. This may be a more worthwhile goal than having a God-like machine that can translate anything and everything at human parity. 

Even in the AI-will-solve-all community, we know that "language is hard" so maybe we need more focus on improving the man-machine interface, and the quality of the interaction and finding more sophisticated collaborative models. Rapid evolution, intuitive and collaborative interaction, and instant learning seem like a more promising vision to me than crawling all the data on the web and throwing machine learning pixie dust at your ten trillion word TM training data. 

Getting to a point where the large majority of translators ALWAYS WANT TO USE MT because it simply makes the work easier, more pleasant, and more efficient is perhaps a better focus for the future.  I would bet also that this different vision will be a more likely path to better MT systems that consistently produce better output over millions of sentences.   

The Impact of MT on the Freelance Translator

The ProZ.com Podcast

This is a conversation or interview that I did with Paul Urwin of Proz where the links will take you to the podcast.

The conversation covers possible strategies that freelance translators can adopt to deal with PEMT and provides some guidance (hopefully) on potential new skills that professionals can develop. 

It also provides context on how valuable the translator is even with continuously improving MT and points to a growing awareness that translators are a resource whose value can only grow in importance given the never-ending momentum on content that needs to be translated.

 

This is Part 1.

Paul talks with machine translation expert Kirti Vashee about interactive-adaptive MT, linguistic assets, freelance positioning, how to add value in explosive content situations, e-commerce translation and the Starship Enterprise.

This is Part 2.

Paul continues the fascinating discussion with Kirti Vashee on machine translation. In this episode, they talk about how much better MT can get, which languages it works well for, data, content, pivot languages and machine interpreting.

The Future is NOT just MT

A relatively easy way to understand the power of an adaptive MT solution (that learns from corrective feedback dynamically) is to test ModernMT with the free open-source Matecat CAT tool. A more detailed overview of the capability is given here.