Log on: Remember me
Powered by Elgg
  • Publish Comment:

  • David Schlangen's Pages:

    Pages
  • David Schlangen

  • Owned communities

David Schlangen : Home Page > minutes050607_zeitwort2

  present: Titus, Timo, David
  [ executive summary: need to make distinction between static tracks,
    which always show what happened (= what was annotated) at a given
    moment, and dynamic tracks, which represent one module's
    hypothesis about what happened at given moments. Dynamic tracks
    need to be updated completely (including the full past of the
    track and maybe even the future) every time new information
    becomes available, while static tracks can either be shown
    completely (if viewing from annotated files) or are only being added
    to towards the future (if used in live mode).
    That is, in effect dynamic tracks create little movies, while
    static tracks just move underneath the cursor.
  ]

  In our planning so far, we didn't consider the "show current
  hypothesis" use case, which requires showing the dynamics of
  changing hypotheses which may be about timed events in the past (or
  future). This is most important for ASR output, which includes 
  segmentations of previous material that can change, but may also
  be relevant for other modules. E.g., some modules may make
  predictions about the occurence of future events, where with more
  information these predictions can change.

  That is, the problem is that "current time" must be interpreted as
  "the system's (or rather: the module's) view of the past, the
  present, and the future" -- at least for those tracks that are
  directly linked to one module.

  Solution: have tracks that show current hypothesis state, together
  with control track that records when hypotheses changed. E.g.
 

  CT:    <>     <>   <>     <>    <>
  Hyp:  I-----I--------I-----|
                           |
 
  The <> are the events of a new hypothesis being posted (and hence it
  is a static track), and the track Hyp shows the then current
  hypothesis, either inline, for interval-type information like
  segments or in an external viewer for things like parse trees.

  (Inline may be more useful for modules that produce some form of
  timed event recognition or predictions (like segmentations or
  predictions of future events), because then the information can be
  placed on the same timeline as the main window. An external viewer
  may be more useful for modules that synthesise previous information,
  without reference to the temporal placement of the components in
  this; e.g., parse trees or views of current information state.)

  If you click on previous <>s, either the tool jumps there and shows
  the information state at that point in the inline view, or a viewer
  window opens so that this previous hypothesis state can be compared
  with the current hypothesis.

  Other tracks may be static as before, e.g. one for the gold-standard
  word recognition / segmentation.

  Two types of scrolling then:
  - moving "current time", which means that hypothesis tracks must
    show the state which is appropriate at the given current
    time. (This is just like "play", moving around the main cursor.)
  - scrolling around in the given hypothesis state. E.g., the viewer
    is stopped at 13'20, the left corner of the window is at 12'20,
    and you want to see the information at 10'10 (which for static
    tracks is what actually happened then and for dynamic tracks is
    what that module thinks what happened).
  One moves "red line / current time cursor" (or rather, moves scroll
  along the static line), the other unlocks the current time cursor
  from the middle of the window and moves whole scroll (as it
  keeps showing the state at that, now fixed, current time).

  Additional feature: would be nice to also represent links between
  hypotheses on different tracks. E.g., a <> event (posting of
  hypothesis) in one track may be linked to an earlier <> event in a
  different track, meaning that the former was constructed using the
  latter. (This information needs to be in the system anyways, so
  could just as well be visualised.)




das, 06/05/07 04:11 (GMT)

Keyword: minutes, viewer, zeitwort

Add a new page under this one