Facebook goes neural for translations: Machine translation has come a long way over the years, but the rise of neural networks as the favored method for processing translations stands to up the ante significantly. To that end, Facebook recently completed the shift of its translation services from phrase-based models to neural networks.
Underpinning it all is Caffe2, the open-source deep learning framework Facebook announced in April. Caffe2 tackles problems that phrase-based translation models have difficulty with, such as context, slang and abbreviations, the company said in a blog post:
Our previous phrase-based statistical techniques were useful, but they also had limitations. One of the main drawbacks of phrase-based systems is that they break down sentences into individual words or phrases, and thus when producing translations they can consider only several words at a time. This leads to difficulty translating between languages with markedly different word orderings. To remedy this and build our neural network systems, we started with a type of recurrent neural network known as sequence-to-sequence LSTM (long short-term memory) with attention. Such a network can take into account the entire context of the source sentence and everything generated so far, to create more accurate and fluent translations.
With the new system, we saw an average relative increase of 11 percent in BLEU — a widely used metric for judging the accuracy of machine translation — across all languages compared with the phrase-based systems.
POV: Facebook handles 4.5 billion translations per day. Given that massive corpus of data and ongoing training of the underlying models, you'd expect Caffe2's accuracy to get even better over time.
Accenture shells out for digital transformation talent: As enterprises look to hit the gas pedal on digital projects, the war is on among consultancies for in-demand talent. Accenture has made two more moves in this direction with the acquisitions of Search Technologies and Brand Learning for undisclosed sums.
Search Technologies is a consultancy focused on big data analytics and search projects. With the acquisition, Accenture gains about 200 experts in these areas spread across the U.S., Costa Rica, Phillipines and Europe. Search Technologies' staff will be rolled into Accenture Analytics and its proprietary Content Processing Framework is to be integrated with Accenture's Insights Platform.
Meanwhile, Accenture has also purchased Brand Learning, a consulting firm focused on marketing, sales and human resources. Brand Learning has about 120 employees and will be added to Accenture's Customer and Channels arms. Here's how Accenture describes its bona fides:
Brand Learning has served leading organizations across consumer goods, retail, life sciences, automotive, resources and financial services industries and has capability experts in 16 countries.
Brand Learning helped one retailer build its marketing capabilities and establish a common way of marketing in its Foods division, increasing its customer experience Net Promoter Score (NPS) by 14 percent in recent years. Work with a large consumer products company’s program for sales and commercial leaders focused on enhancing the capabilities of teams on the ground and building sales, commercial and leadership skills.
POV: Accenture has been on a bit of an acquisition tear of late. In June, it bought life science specialists LabAnswer and mobile design firm Intrepid. While none of the deals on their own appear to be massive in scope, Accenture is nonetheless moving aggressive to fill and shore up holes in its lineup of talent. The question, as is the case with all boutique consultancies swallowed up by the Big Five is whether their unique cultures will be retained or homogenized into the mothership.
Legacy watch: Not always sunny in Philadelphia: These are unhappy times in the City of Brotherly Love's IT department. No less than four major IT projects are running well over time and budget, and those woes are reflective of a deeply dysfunctional environment overall, as the Inquirer newspaper reports:
Three-and-a-half years ago, then-Mayor Nutter announced the city would spend $4 million to implement a computerized licensing and inspection system that would let people apply for permits online and enable inspectors to more easily track applications and violations.
“We’ll spend some money, we’ll save some money, and we’ll make a lot of money,” Nutter said at the 2014 news conference.
The program was supposed to be fully functioning by the end of 2015. Instead, it is only halfway done and the bill has topped $10 million.
Around the same time, the city hired a company for $15 million to replace its 30-year old payroll system and sync it with pensions, benefits, and time management under the same computerized system, all by mid-2015. Already $23 million has been spent on the project — and it’s far from complete.
City officials also wanted to move away from the thousands of Excel pages used to create annual budgets. A $3 million contract was signed to digitize the city’s budgeting system and make it more efficient and transparent. After shelling out $1.6 million, they now want to scrap it.
POV: There is much more detail in the Inquirer's full report, which is well worth a read. It includes comment from one city official who says that delays and overruns are simply to be expected. One wonders how that would go over if said by a CIO delivering a progress report to his or her board of directors. But it's also reflective of the built-in obstacles government IT projects face, such as a culture of bureaucracy and the whims of the political winds, which can blow new leadership in and out of power and with them, different priorities, awareness and levels of oversight.
