1 / 7

Sausage Comprehension and Listening Experiments

Sausage Comprehension and Listening Experiments. Lidia Mangu, Geoffrey Zweig & Bhuvana Ramabhadran. Human Experiments: Motivation. Where should we focus research? More sophisticated linguistic models? Morphology, syntax, semantics, pragmatics More sophisticated acoustic models?

dorit
Télécharger la présentation

Sausage Comprehension and Listening Experiments

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Sausage Comprehension and Listening Experiments Lidia Mangu, Geoffrey Zweig & Bhuvana Ramabhadran

  2. Human Experiments: Motivation • Where should we focus research? • More sophisticated linguistic models? • Morphology, syntax, semantics, pragmatics • More sophisticated acoustic models? • Phone, syllable, word levels •  Sausage Comprehension and Listening • How good does a recognizer have to be? • Question Answering with Faulty Transcripts

  3. Sausage Comprehension & Listening: Comprehension “… spending and an indebted economy. But I have a really good friend who keeps trying to convince me that being in debt is a healthy economy. And I just do not see that. I I wish that uh…” too dependent heavily petroleum who on we’re depend not we to depended own controlling were have depending so

  4. Context Context stem could that on it cuts down and I comes stay I’m they cut them Sausage Comprehension & Listening: Listening Confusion-bin

  5. Experimental Setup • History of 5 segments • Average of 4 choices per bin • Acoustic context of 2 words left/right • Average duration of 1 second • RT03 • 29.2% Consensus error • 8% Oracle WER • 180 sausages; 950 words • MALACH • 28.9% Consensus error • 9.5% Orcale WER • 190 sausages; 1000 words • Results aggregated over 20 people

  6. Preliminary Results

  7. Preliminary Conclusions • Substantial gain possible from better acoustic modeling of isolated 3 to 5 word segments • Sophisticated non-statistical LMs unlikely to help conversational speech (WER) • Need to investigate linked AM/LMs • Do people do it all together?

More Related