1 / 18

Language Processing Technology

Machines and other artefacts that use language. Language Processing Technology. HAL. The computer in 2001: A Space Odyssey could: Speak and understand English. Recognize speech. Read lips… HAL doesn’t exist yet, but there is hope (maybe this decade). HAL’s capabilities.

brie
Télécharger la présentation

Language Processing Technology

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. Machines and other artefacts that use language Language Processing Technology

  2. HAL The computer in 2001: A Space Odyssey could: Speak and understand English. Recognize speech. Read lips… HAL doesn’t exist yet, but there is hope (maybe this decade).

  3. HAL’s capabilities “Open the pod bay doors please, HAL” “I’m sorry Dave, I’m afraid I can’t do that”

  4. Speech recognition HAL must be able to analyze the speech signal and work out which words were said. Needs to know about phonetics and phonology.

  5. Speech synthesis HAL needs to be able to take a series of words and pronounce it naturally and fluently. Use contractions “I’m sorry”, not “I am sorry” Needs to get rhythm and intonation right.

  6. Methods for speech synthesis By imitation of articulator movements (very flexible) By stringing together units (easy to achieve naturalness) Phones (simplest) Diphones (captures co-articulation) Larger units (can be very accurate)

  7. Part of speech tagging Current computational linguistics can handle the task of assigning parts of speech to words. HAL does this, presumably, but this is really just a step along the way to syntax.

  8. Syntax HAL understood that the request was a request. It was “open the pod bay doors HAL” It wasn’t “the pod bay doors are open” It wasn’t “are the pod bay doors open?” It wasn’t “could you open the pod bay doors?” To do that, we need syntax (which forms groups of words)

  9. Semantics HAL has to know that the request is about the pod bay doors (not Dave’s breakfast) It has to know lexical semantics – what kind of thing the word “doors” can refer to. It also has to know compositional semantics – how to put the meanings of “pod”,”bay” and “doors” together to get the right composite meaning.

  10. Pragmatics HAL could have said “No” Instead it said “I’m sorry”, “I’m afraid…” and “I can’t” Would have been more truthful to say “I won’t”. But the correct use of polite and indirect language is usual for humans. Human listeners realize that HAL refused.

  11. Discourse conventions HAL could have just failed to reply. Human conversation requires and expects that we engage in structured interactions. HAL correctly uses the word “that” to refer to the topic raised by Dave. Correctly structuring such conversations requires HAL to know when to speak and when to listen.

  12. HAL overview Speech synthesis Speech recognition Syntax Lexical semantics Compositional semantics Discourse and pragmatics

  13. Real world applications Can’t do HAL (yet), but can do various useful things with more limited abilities. Most of these rely on good choice of a task. Suitable tasks involve something that computers do better (or much faster) than people do. Language is in service of the task, not the other way round.

  14. Information retrieval Given a (perhaps huge) collection of documents and a user need expressed as a query, present the most relevant documents. Query “pod bay door malfunction” Response – a ranked list of documents Technology is essentially glorified word counting.

  15. Text categorization Given a text and a set of categories, assign the text to a categories Here’s an email message, is it Junk mail About sailing boats A rant from alt.atheism or comp.sys.ibm.pc Technology relies on sophisticated weighting of evidence.

  16. Information extraction Given hundreds of articles, find information (not just documents) of interest. “John Smith joined the board of IBM as chairman” {IN=“John Smith”; POST=“chairman”; COMPANY=“IBM”} Technology is syntactic parsing plus lots of task specific tricks.

  17. Machine translation Take a sentence in English and put it into (for example) Japanese. Very demanding: requires all HAL’s levels plus a sophisticated understanding of literary style. But commercial and research systems exist which can “Gloss” a passage producing something that a human translator can adapt. Technology is usually word-to-word models with good bilingual dictionaries but limited syntax and some smarts for choosing between alternatives.

  18. Dialog systems You phone up a system and talk to it. It listens, talks back and does something for you. Tourist information, airline bookings. Your assignment for next time is to imagine a travel plan, phone 1-877-CMU-PLAN (268-7526), which is toll-free, and pretend to book a flight.

More Related