This Meetup is past

125 people went

Comcast Labs

1110 Vermont Ave, NW · Washington, DC

How to find us

ALL VISITORS TO 1110 VERMONT MUST BE PRE-REGISTERED W/SECURITY. PLEASE RSVP W/YOUR FIRST & LAST NAME AND BRING A PHOTO ID FOR CHECK-IN.

Location image of event venue

Details

Another month, another new venue! This time, we're meeting in a space generously provided by Comcast, near the McPherson Sq & Farragut North metro stations. They're also sponsoring food for the event - many thanks to Comcast!

We've got two great speakers on the lineup this month, both of whom will be giving presentations on speech-based systems.

Vikrant Sagar (https://www.linkedin.com/in/sagarvikrant) completed a Masters in Computer Science at the University ​o​f Florida in 2013, and is now an NLP Engineer at Comcast Labs (http://dclabs.comcast.com/research/) in downtown Washington, DC. Comcast Labs' NLP group is developing applications around the voice-enabled TV remote, which is part of the new X1 Entertainment System. Vikrant's presentation will discuss the kinds of NLP problems being worked on by the team at Comcast Labs.

Scott Novotney (https://www.linkedin.com/in/snovotney) has a PhD in Computer Science from Johns Hopkins and works as a Scientist at Raytheon BBN Technologies (http://www.bbn.com/) on multiple DARPA projects. His presentation is entitled Cheap, Fast and Good Enough: Transcribing Speech with Mechanical Turk:
Historical estimates of human transcription of conversational speech require at a minimum of 10 hours of effort per hour. These costs quickly multiply when the number of transcribers who have access to sensitive data, such as medical records, becomes limited. In this talk, we engage non-expert transcribers who are given minimal instructions to bootstrap an automatic speech recognition system. With transcription costs of only five dollars per hour of data, the transcribers have an error rate of 25%. Nonetheless, when this data is used to train a statistical model, the gap between on-experts and professionals disappears. We then show how non-experts can be used to rank other non-expert transcribers without requiring any gold standard data.

************************

We'll gather starting at 6:30pm at Comcast Labs (http://dclabs.comcast.com/) - someone will greet you outside and let you into the building. Introductions & announcements will start around 7:00pm, and presentations will begin at 7:30pm. Afterwards, there will be plenty of time for follow-up questions, networking, and more.

DC NLP meets each month to network, socialize, and learn about the interesting work folks are doing in natural language processing, computational linguistics, text analytics, and more.

Do you have something you'd like to share with the group? Let us know! We're always looking for speakers to give talks at future meetups, and don't forget to follow @DCNLP (https://twitter.com/DCNLP/) on Twitter!