Log on: Remember me
Powered by Elgg

timo :: Activity :: Friends & Community

People: Everyone | Friends & Community | Inbox | Just Me
Display: Full-text | Summary
Include: Blog Posts | Blog Comments | Files | Wiki Page | Wiki Comments

<< Back

Page 2 of 11

Forward >>
Literature Day
atterer | page | Thu May 29

LITERATURE LIST   

 

-- James Allen's group --

 Stoness, Tetreault, Allen (2004): Incremental Parsing with Reference Interaction

Aist (2006): Incrementally Segmenting Incoming Speech into Pragmatic Fragments

Aist, Allen, Campana, Stoness, Swift, Tanenhaus (2007): Incremental understanding in human-computer dialogue and experimental evicence for advantages over nonincremental methods

Aist, G.S., Allen, J., Campana, E., Galescu, L., Gomez Gallo, C.A., Stoness, S., Swift, M., and Tanenhaus, M. 2006. Software architectures for incremental understanding of human speech. Proceedings of the International Conference on Spoken Language Processing (ICSLP).Pittsburgh, September 17-21 

 

-- misc -- 

DeVault and Stone (2003): Domain Inference in Incremental Interpretation

Seginer (2007): Fast Unsupervised Incremental Parsing; ACL

Nivre (2004): Incrementality in Deterministic Dependency Parsing 

J.-C. Chappelier, M. Rajman, R. Aragues, A. RozenknopLattice Parsing for Speech Recognition, 6ème conférence sur le Traitement Automatique du Langage Naturel (TALN99), Cagèse, France, July 12-17, 1999 

-- Matsubara lab -- 

Murase, Matsubara, Kato, Inagaki (2001): Incremental CFG Parsing with Statistical Lexical Dependencies

... or maybe a more recent paper from that group?

Tomohiro Ohno, Shigeki Matsubara, Hideki Kashioka, Naoto Kato, Yasuyoshi Inagaki: Incremental Dependency Parsing of Japanese Spoken Monologue Based on Clause Boundaries, Proceedings of the 9th European Conference on Speech Communication and Technology (INTERSPEECH'2005-Eurospeech), 

 http://slp.el.itc.nagoya-u.ac.jp/en/publications 

 

Literature Day
atterer | page | Thu May 29

LITERATURE LIST   

Stoness, Tetreault, Allen: Incremental Parsing with Reference Interaction

Aist: Incrementally Segmenting Incoming Speech into Pragmatic Fragments

Aist, Allen, Campana, Stoness, Swift, Tanenhaus: Incremental understanding in human-computer dialogue and experimental evicence for advantages over nonincremental methods

DeVault and Stone: Domain Inference in Incremental Interpretation

Seginer (2007): Fast Unsupervised Incremental Parsing; ACL

Nivre: Incrementality in Deterministic Dependency Parsing 

Murase, Matsubara, Kato, Inagaki (2001): Incremental CFG Parsing with Statistical Lexical Dependencies

... or maybe a more recent paper from that group?

Tomohiro Ohno, Shigeki Matsubara, Hideki Kashioka, Naoto Kato, Yasuyoshi Inagaki: Incremental Dependency Parsing of Japanese Spoken Monologue Based on Clause Boundaries, Proceedings of the 9th European Conference on Speech Communication and Technology (INTERSPEECH'2005-Eurospeech), 

 http://slp.el.itc.nagoya-u.ac.jp/en/publications 

 

TEST 

minutes260508
das | page | Tue May 27
- InPro, meeting, minutes, 26/05/08
  - present: T, M, D
  - acoustic model: improving. coming along.
  - still to do:
    - from microphone to asr to parser
    - n-best, confidences
  - parser:
    - go parallel approach? Have both parser and pattern matcher that
      simply looks at the string so far and tries to grab from it
      whatever kinds of expressions it knows. Sort of named entity
      recognition.. 
      .. but how can this be done incrementally?
  - discussed abstr incr model:
    - problem w/ meaning of grounded_in vs same-level-link. Is not an
      S node grounded in its constituents, rather than the string?
      Design decision, ultimately. All we want is that the necessary
      information is transmitted. E.g., if a VP which is a constituent
      of an S is withdrawn (parser removes grounded-in, because it
      doesn't think anymore that the evidence (the words) justify
      hypothesis of VP), the S must be marked as ungrounded as
      well. This can be given to the processor as a housekeeping task
      (if it ungrounds stg, it must also unground everything it has
      built on it), it can be added to the purge function (if an LB-IU
      has become ungrounded, purge it and everything that links to it
      on same level), or it can be done automatically through
      interleaving of grounded-in and same-level-link (the grounded-in
      of larger constituents can be expressed through reference to the
      grounded-in of its same-level constituents, in which case every
      change in their status automatically percolates up). Last option
      is probably the most elegant.
    - parser as semi-circular module. Constituents it has built are
      put on LB, to be consumed together with LB-IUs coming from
      previous module.
    - question of boundaries of task. Is this also used to model
      dialogue history? The links btw utterances? Could be done, but
      probably practically not useful.. But if done, maybe useful to
      have commit as 3-way distinction. `freeze', but allow to go back
      later and revise it, e.g. after a misunderstanding. (?).
    - does the model have to say something about how to avoid timing
      issues? Race conditions, starving, etc. Probably not. If time,
      one could think about how problematic conditions could arrive
      (e.g., a purge coming to late, after a hypothesis has been
      extended, thus failing to percolate quickly enough through the
      network. -- Although this probably could be avoided if the
      processor does its functions in sequence each time its triggered
      into action, with purge coming first.)

minutes260508
das | page | Tue May 27
- InPro, meeting, minutes, 26/05/08
  - present: T, M, D
  - acoustic model: improving. coming along.
  - still to do:
    - from microphone to asr to parser
    - n-best, confidences
  - parser:
    - go parallel approach? Have both parser and pattern matcher that
      simply looks at the string so far and tries to grab from it
      whatever kinds of expressions it knows. Sort of named entity
      recognition.. 
      .. but how can this be done incrementally?
  - discussed abstr incr model:
    - problem w/ meaning of grounded_in vs same-level-link. Is not an
      S node grounded in its constituents, rather than the string?
      Design decision, ultimately. All we want is that the necessary
      information is transmitted. E.g., if a VP which is a constituent
      of an S is withdrawn (parser removes grounded-in, because it
      doesn't think anymore that the evidence (the words) justify
      hypothesis of VP), the S must be marked as ungrounded as
      well. This can be given to the processor as a housekeeping task
      (if it ungrounds stg, it must also unground everything it has
      built on it), it can be added to the purge function (if an LB-IU
      has become ungrounded, purge it and everything that links to it
      on same level), or it can be done automatically through
      interleaving of grounded-in and same-level-link (the grounded-in
      of larger constituents can be expressed through reference to the
      grounded-in of its same-level constituents, in which case every
      change in their status automatically percolates up). Last option
      is probably the most elegant.
    - parser as semi-circular module. Constituents it has built are
      put on LB, to be consumed together with LB-IUs coming from
      previous module.
    - question of boundaries of task. Is this also used to model
      dialogue history? The links btw utterances? Could be done, but
      probably practically not useful.. But if done, maybe useful to
      have commit as 3-way distinction. `freeze', but allow to go back
      later and revise it, e.g. after a misunderstanding. (?).
    - does the model have to say something about how to avoid timing
      issues? Race conditions, starving, etc. Probably not. If time,
      one could think about how problematic conditions could arrive
      (e.g., a purge coming to late, after a hypothesis has been
      extended, thus failing to percolate quickly enough through the
      network. -- Although this probably could be avoided if the
      processor does its functions in sequence each time its triggered
      into action, with purge coming first.)

Home Page
das | page | Tue May 27

Besprechungsprotokolle / meeting minutes

(newest first)

26/05/08 minutes260508 

19/05/08 minutes190508 

05/05/08 minutes080508 

14/04/08 minutes140408

03/02/08 minutes030208b

04/12/07 minutes041207

26/11/07 @Timo

19/11/07 minutes191107

13/11/07 minutes131107

05/11/07 minutes051107

22/10/07 minutes221007

01/10/07 minutes2007_10_01

10/09/07 minutes100907

23/08/07 minutes230807

03/07/07 minutes030707

19/06/07 minutes190607

05/06/07 minutes050607_zeitwort2

21/05/07 minutes210507

 

Sonstiges

 

Conferences2008

Literature Day
atterer | page | Thu May 22

LITERATURE LIST   

Stoness, Tetreault, Allen: Incremental Parsing with Reference Interaction

Aist: Incrementally Segmenting Incoming Speech into Pragmatic Fragments

Aist, Allen, Campana, Stoness, Swift, Tanenhaus: Incremental understanding in human-computer dialogue and experimental evicence for advantages over nonincremental methods

DeVault and Stone: Domain Inference in Incremental Interpretation

Seginer: Fast Unsupervised Incremental Parsing

Nivre: Incrementality in Deterministic Dependency Parsing 

Murase, Matsubara, Kato, Inagaki: Incremental CFG Parsing with Statistical Lexical Dependencies

TEST 

Literature Day
atterer | page | Tue May 20

LITERATURE LIST   

Stoness, Tetreault, Allen: Incremental Parsing with Reference Interaction

Aist: Incrementally Segmenting Incoming Speech into Pragmatic Fragments

Aist, Allen, Campana, Stoness, Swift, Tanenhaus: Incremental understanding in human-computer dialogue and experimental evicence for advantages over nonincremental methods

DeVault and Stone: Domain Inference in Incremental Interpretation

Seginer: Fast Unsupervised Incremental Parsing

Nivre: Incrementality in Deterministic Dependency Parsing 

Murase, Matsubara, Kato, Inagaki: Incremental CFG Parsing with Statistical Lexical Dependencies

Literature Day
atterer | page | Tue May 20

   

Stoness, Tetreault, Allen: Incremental Parsing with Reference Interaction

Aist: Incrementally Segmenting Incoming Speech into Pragmatic Fragments

Aist, Allen, Campana, Stoness, Swift, Tanenhaus: Incremental understanding in human-computer dialogue and experimental evicence for advantages over nonincremental methods

DeVault and Stone: Domain Inference in Incremental Interpretation

Seginer: Fast Unsupervised Incremental Parsing

Nivre: Incrementality in Deterministic Dependency Parsing 

 Murase, Matsubara, Kato, Inagaki: Incremental CFG Parsing with Statistical Lexical Dependencies

Literature Day
atterer | page | Tue May 20

   

Stoness, Tetreault, Allen: Incremental Parsing with Reference Interaction

Aist: Incrementally Segmenting Incoming Speech into Pragmatic Fragments

Aist, Allen, Campana, Stoness, Swift, Tanenhaus: Incremental understanding in human-computer dialogue and experimental evicence for advantages over nonincremental methods

DeVault and Stone: Domain Inference in Incremental Interpretation

Seginer: Fast Unsupervised Incremental Parsing

Nivre: Incrementality in Deterministic Dependency Parsing 

minutes190508
das | page | Tue May 20
  - present: M, T, D
  1. akustisches Modell
     - momentan nicht besonders gut. Trainiert mit Daten von vox forge
       (?).
     - to be redone with Pento Naming Corpus data
     - also add Verbmobil data? Kiel Korpus?
     - when that is done, tackle other points:
       - LexTree instead of SimpleLM, so that tri-gram LMs can be used
       - n-best lists, or lattices
       - confidence scores
     --> by end of May, we will have model that is good enough to at
     least get an idea of what we will be working with.
Germany's next top (acoustic) model!
  2. literature day(s)
     - topics:
       - parsing / semantics for SDS
       - incremental parsing (in general)
       - parsing and prosody
       - incremental systems
       - EOT prediction / recognition
     ----> collect paper suggestion on Wiki. Michaela will organise
       the (first) day.
  X. brief interlude: do we need new collaboration management system?
     What do we use at the moment:
     - email .. anouncements, rarely for content, moving of files
       - pro:
       - archivable / searchable
- attachments
     - IM    .. quick questions (T and D; M doesn't use at all)
       - pro:
       - fast, instant (d'uh)
       - cons:
       - not easy to search. breaks unity of transmission (where did
           I read that?)
     - Elgg  .. mostly for meeting minutes (and there the wiki part is
           used only). T uses it for status reports. M doesn't
   use it at all. Not used for literature notes etc.
       - pro:
       - archival.. ?
       - cons:
       - active effort needed to put things there & to check for new
     entries (since integration of RSS in our workflows doesn't
     seem to work yet)
     What we'd need:
     - ideal would be a system that has more than one interface,
       including email. That is, new content can be contributed by
       email, web, whatever, is spread via email, can be searched in
       one central place.
       Bonus: has interface to svn, e.g. one can link to documents in
       the svn. also has IM client, and archives IMs. 
     .. is that trac? probably not. Does that exist? Probably not.
  3. WOz
     - controls mouse and prompts
     - data can be used for acoustic model and language model, also
       for learning about dialogue dynamics that can be expected. Main
       goal is to see how people behave if they assume that the
       capabilities of the instruction follower are limited.
  4. Parser, requirements
     - robustness. Can't assume that it always will be able to parse
       into sentences. Should be happy with intermediate
       constituents.
       Doesn't this requirement fall out of incrementality in any
       case? If partial results are passed on, they always will be
       sub-S constiuents.
       Yes, but the problem has a slightly different aspect as
       well. The question is what to do if what the parser can
       possibly recognise (because of ASR problems) are sequences of
       NPs. If there is no syntactic rule that could potentially
       integrate those, if not specially prepared, the parser would
       not even attempt to build the later NPs.
       So what it boils down to again is the question of whether the
       parser needs a notion of being "restarted" or not.
     - incrementality. d'uh.
     - non-commital, capable of making revisions
       related to topic above. What happens if the parser decides to
       give up integrating new material into the current structure?
     - mid-term: probabilistic, integrate prosody (as information on
       words or as pseudo tokens), parse lattices
     Discussed:
     - is top-down parsing a good idea for an incremental parser? can
       this work? Think through, what are the problems that can arise
       in either case?
       --> Michaela? Think up a few example cases (including ambiguous
         and garbage-full sentences) and see what either parsing
     strategy will do.
     - how is commital (when parser decides that what it currently is
       consuming is unrelated to structure it has previously built)
       realised technically, how is it triggered? 
       Cleaning datastructures. Empty chart, or un-link datastructure?

<< Back

Page 2 of 11

Forward >>