Weekly news round-up prepared by the Editorial Staff of LangTechNews.
Top of the Week: AT&T
On an industry-wide
scale, LTN notes the announcement that venerable US telecoms research lab AT&T
will open up its speech-tech nest egg to developers. AT&T has a wealth of research results in every area of speech communication, so there will be a huge
box of toys (and speech resources) for access via the API. The usual app suspects
- speech translation and virtual assistants - got immediate mention. However, doubts
about whether ASR systems can handle ‘dialect’ and ambient noise may put the
brakes on global solutions. Solving this will play an important role in producing
enterprise-strength ASR that can search all kinds of spoken streams (not just
videos but phone calls) and can use speech content in all languages and dialects as input to
analytics dashboards. When real data matters – as in training translation
engines or speech recognizers - opening up existing technologies to the
developer crowd will, if nothing else, help quickly eliminate methods that don’t work in advancing the speech
agenda.
Most Promising New Start-up
The Irish start-up Xcelerator has just gone BETA with its new KantanMT translation platform and is
ready to serve customers. Unlike its natural rival ALS’s SmartMATE solution, which targets the enterprise market
and includes post-editing services, KantanMT is a technology pureplay focusing
on individual translators working in a “traditional” by-the-word service
industry who can afford a modest monthly fee for the technology clout of a
major statistical MT system on a secure cloud, delivered via a simple web
interface. With millions of people – including professional translators – already using MT, the
market for this professional-level offering is far from trivial. The technology
behind both these companies was spawned by Irish MT researchers. Ireland, and
Dublin in particular, largely pioneered the European software localisation industry
back in the late 1980s and has since developed an academic and real-world
knowledge hub to address next-generation translation, much of it centred at CNGL (Centre for Next Generation Localisation, at Dublin City University). KantanMT looks like a promising, well-targeted spinoff.
Digital Agenda: The End of Paper Patents in the EU
Patents provide a vital form of protection for intellectual property, and a source of income in a highly competitive innovation culture. In Europe, the red-tape involved filing a patent has made it an expensive business, and seen by some as a brake on innovation – not least because of the need to make multiple filings in many countries using different languages. To improve access to European patent information, the European Patent Office (EPO) last year worked out a deal with Google to provide automated translation for patents in multiple languages – much to the dismay of European translation service suppliers. At the same time another Irish-led effort called PLuTO (EC-funded this time) is being developed to provide a patent translation/search service; it was launched last October and is now live online in BETA. Everyone it seems is trying to level the playing field for patent filing, while hoping to make some income in the process. This week, in an effort to upgrade the entire patent process, the EPO announced that patent filing had gone digital. This will, among other things, simplify the translation process for Google Translate and PluTO, and also make searching, analytics and compliance-checking operations far less costly. Expect patent filing to morph into a mobile app sometime soon…
Virtual Business Agents
Virtual Assistants (VAs) – usually voice-based interfaces to computer systems such as retail sites, laptops and now smartphones – have been around for a while in the form of chatbots. They took on a whole new meaning for consumers with the introduction
of Apple’s Nuance-driven Siri for iPhone late last year. In many ways VAs pose an AI-complete challenge: delivering language understanding for free-range human speech input under messy real-world conditions. Which is why it is imperative for early developers (such as Apple and Vlingo) to capture rich data about the user experience across many languages, so they can debug and
improve these complex products over time. The next – and possibly easier – step, which is already underway, is to switch the target from generic chatbots to business-specific Enterprise VAs in customer-facing environments where cost,
time, focus and quality are critical sales factors. The analyst firm Gartner predicts that by next year some 15% of Fortune 1000 companies will use a virtual assistant to serve up Web self-service content for CRM service delivery. And the trend is not limited to customer service: last week EasyAsk launched a mobile service provocatively called Quiri to allow business users to ask questions and get answers from any number of existing corporate data sources, including CRM and business intelligence systems. Meanwhile Forrester reported that Synthetix is one of 10 leading vendors in its analysis of Five Essential eBusiness Criteria For a Successful Virtual Agent Vendor Selection. Now that tablets are expected to become the preferred computing device by 2016, VA-friendliness (including speech interfaces) will likely evolve into a crucial selling point for enterprise users. LTN expects that equipping these next-generation interfaces with voice
biometrics for security will be a milestone on that path.
