All news

Last year, Microsoft’s speech and dialog research group announced a milestone in reaching human parity on the Switchboard conversational speech recognition task, meaning we had created technology that recognized words in a conversation as well as professional human transcribers.

After our transcription system reached the 5.9 percent word error rate that we had measured for humans, other researchers conducted their own study, employing a more involved multi-transcriber process, which yielded a 5.1 human parity word error rate. This was consistent with prior research that showed that humans achieve higher levels of agreement on the precise words spoken as they expend more care and effort. Today, I’m excited to announce that our research team reached that 5.1 percent error rate with our speech recognition system, a new industry milestone, substantially surpassing the accuracy we achieved last year. A technical reportpublished this weekend documents the details of our system.

Switchboard is a corpus of recorded telephone conversations that the speech research community has used for more than 20 years to benchmark speech recognition systems. The task involves transcribing conversations between strangers discussing topics such as sports and politics.

We reduced our error rate by about 12 percent compared to last year’s accuracy level, using a series of improvements to our neural net-based acoustic and language models. We introduced an additional CNN-BLSTM (convolutional neural network combined with bidirectional long-short-term memory) model for improved acoustic modeling. Additionally, our approach to combine predictions from multiple acoustic models now does so at both the frame/senone and word levels.

Moreover, we strengthened the recognizer’s language model by using the entire history of a dialog session to predict what is likely to come next, effectively allowing the model to adapt to the topic and local context of a conversation.

Our team also has benefited greatly from using the most scalable deep learning software available, Microsoft Cognitive Toolkit 2.1 (CNTK), for exploring model architectures and optimizing the hyper-parameters of our models. Additionally, Microsoft’s investment in cloud compute infrastructure, specifically Azure GPUs, helped to improve the effectiveness and speed by which we could train our models and test new ideas.

Reaching human parity with an accuracy on par with humans has been a research goal for the last 25 years. Microsoft’s willingness to invest in long-term research is now paying dividends for our customers in products and services such as Cortana, Presentation Translator, and Microsoft Cognitive Services. It’s deeply gratifying to our research teams to see our work used by millions of people each day.

Many research groups in industry and academia are doing great work in speech recognition, and our own work has greatly benefited from the community’s overall progress. While achieving a 5.1 percent word error rate on the Switchboard speech recognition task is a significant achievement, the speech research community still has many challenges to address, such as achieving human levels of recognition in noisy environments with distant microphones, in recognizing accented speech, or speaking styles and languages for which only limited training data is available. Moreover, we have much work to do in teaching computers not just to transcribe the words spoken, but also to understand their meaning and intent. Moving from recognizing to understanding speech is the next major frontier for speech technology.

As language-industry professionals, we hear a lot about endangered languages and how the number of spoken languages keeps dwindling worldwide. But what about language writing systems? With roughly 6,000 languages throughout the world, there are surprisingly only about 120 to 140 written language scripts and alphabets. Many of these are disappearing as well.

What does it mean to the people who speak languages with dying writing systems? What happens when a new generation can no longer read its traditional script? And why do writing systems matter when language is essentially an oral process?

These are just some of the questions Renato Beninatto and Michael Stevens discuss with Tim Brooks on this week’s episode of Globally Speaking.

Tim is the founder of the Endangered Alphabets Project, an organization whose mission is to help preserve endangered cultures by using their writing systems to create artwork and educational materials.

His story is a fascinating one, and so are the many different ways writing can impact and preserve cultures. Topics include:

• Why writing can be viewed as a beautiful form of art.
• What are some of the languages whose writing systems are disappearing?
• Why is there a growing effort to revive traditional scripts?
• How can we help protect more writing systems from disappearing?

Listen to podcast >>

Issues related to gender, the workplace, and family are among the most important social and political concerns today. Language services – translation, localization, interpreting, and related tasks – are frequently seen as a female-dominated profession, but little data has been available to compare with other industries.

“Gender and Family in the Language Services Industry” is the first in a series of CSA Research reports dealing with gender and family issues among those who are employed in the language industry or who work with language services. It covers high level issues concerning men’s and women’s experience with the workplace and family/work-life balance. Based on 2,200 global responses, the report shines light on topics ranging from pay to personality to promotions.

The data from this research separates perception from reality.

• What region of the world has the lowest language services gender pay gap?
• Should gender balance be mandatory in the hiring process?
• Which gender has experienced discrimination – either positive or negative – based on specific personal characteristics?
• Who are more likely to reduce work hours, take time off, quit jobs, or turn down promotions to care for family members?

“Gender and Family in the Language Services Industry,” and subsequent reports, is available for free with registration.

Dowload the report at Common Sense Advisory >>

During the Cold War, Soviet leader Nikita Khrushchev made a statement to a group of Western ambassadors that was translated as “We will bury you.” Naturally, what was seen as a rude and bellicose remark by a top Soviet official speaking to foreign diplomats made headlines, and it exacerbated tensions between the rival Eastern and Western blocs. But what Khrushchev actually said was slightly different.

http://thebulletin.org/shoddy-translation-western-media-increasing-nuclear-tensions-again11046?platform=hootsuite

ELSPEET, THE NETHERLANDS, AUGUST 10th 2017 – Pieter Beens, freelance translator and owner of Dutch translation company Vertaalt.nu, introduces xl8 review. This new review project focuses on products that will bring health and productivity improvements for translators. The project initially starts with a monthly review, but inventors and manufacturers are already eager to participate.

“xl8 review is a great new way for translators to look at innovations that can improve their lives”, says Pieter Beens. He started the project out of curiosity, initially reviewing books on his business blog. Combining his interest in product innovations and review experience for various newspapers and magazines, he decided to bring out xl8 review to specifically focus on products that can be of use for translators. “Every year many tools and products are introduced to improve our lifes, but it is up to xl8 review to prove what they are worth.” The success of the new series of product reviews is already indicated by a huge list of inventors and manufacturers wanting to have their products reviewed, says Beens. “I have a book scanner, innovative flower pot for offices and hydration bottle among others. Manufacturers are really interested in having their products tested for the specific translation industry.”

The first review is to be published in a couple of weeks, and a new review will be added each month afterwards. Beens: “My initial plan is to publish monthly, but the long list makes it almost essential to increase the frequency.” xl8 reviews will be posted on The Open Mic as well. ProZ.com has also shown an interest in useful reviews for translators. The xl8 review project will therefore have a reach of tens of thousands of translators.

Website: https://www.vertaalt.nu/xl8-review

Facebook announced this morning that it had completed its move to neural machine translation — a complicated way of saying that Facebook is now using convolutional neural networks (CNNs) and recurrent neural networks (RNNs) to automatically translate content across Facebook.

Google, Microsoft and Facebook have been making the move to neural machine translation for some time now, rapidly leaving old-school phrase-based statistical machine translation behind. There are a lot of reasons why neural approaches show more promise than phrase-based approaches, but the bottom line is that they produce more accurate translations.

Traditional machine translation is a fairly explicit process. Relying on key phrases, phrase-based systems translate sentences then probabilistically determine a final translation. You can think of this in a similar light as using the Rosetta Stone (identical phrases in multiple languages) to translate text.

In contrast, neural models deal in a higher level of abstraction. The interpretation of a sentence becomes part of a multi-dimensional vector representation, which really just means we’re trying to translate based on some semblance of “context” rather than phrases.

It’s not a perfect process, and researchers are still tinkering with how to deal with long-term dependencies (i.e. retaining understanding and accuracy throughout a long text), but the approach is incredibly promising and has produced great results, thus far, for those implementing it.

Google announced the first stage of its move to neural machine translation in September 2016 and Microsoft made a similar announcement two months later. Facebook has been working on its conversion efforts for about a year and it’s now at full deployment. Facebook AI Research (FAIR) published its own research on the topic back in May and open sourced its CNN models on GitHub.

“Our problem is different than that of most of the standard places, mostly because of the type of language we see at Facebook,” Necip Fazil Ayan, engineering manager in Facebook’s language technologies group, explained to me in an interview. “We see a lot of informal language and slang acronyms. The style of language is very different.”

Facebook has seen about a 10 percent jump in translation quality. You can read more into the improvement in FAIR’s research. The results are particularly striking for languages that lack a lot of data in the form of comparative translation pairs.

Source: TechCrunch, article by John Mannes posted on 3 August 2017 – Read the original at:
https://techcrunch.com/2017/08/03/facebook-finishes-its-move-to-neural-machine-translation/