andthe buildingwitheehbrickeehrooforIlet

1
PICKERING Robust interpretation in the HIGGINS
spoken dialogue system Gabriel Skantze and Jens
Edlund
Evaluation 8 subjects were given the task to walk
around in a HIGGINS virtual 3D-city and describe
their positions relative to objects in their
surrounding, at locations of their choice. A
grammar was build based on this corpus. 16 new
subjects were recorded using the same procedure
and the resulting corpus was processed to
evaluate the interpreter. Example
utterance(s) Ok now I am standing on
körsbärsvägen behind me Ive got a red house and
up ahead to the right there is a red brick
house Robustness against errors ASR errors were
induced by using increasing pruning
levels. Effects of parameters
Introduction In this paper, the effects of
prosodic features on the interpretation of
elliptical clarification requests in dialogue are
studied. An experiment is presented where
subjects were asked to listen to short
human-computer dialogue fragments in Swedish,
where a synthetic voice was making an elliptical
clarification after a user turn. The prosodic
features of the synthetic voice were
systematically varied, and the subjects were
asked to judge what was actually intended by the
computer. The results show that an early low F0
peak signals acceptance, that a late high peak is
perceived as a request for clarification of what
was said, and that a mid high peak is perceived
as a request for clarification of the meaning of
what was said. The study can be seen as the
beginnings of a tentative model for intonation of
clarification ellipses in Swedish, which can be
implemented and tested in spoken dialogue systems.

• Interpreter features
• Grammar independent robustness. Pickering
  automatically
• Allows insertions.
• Allows non-agreement.
• Finds the best combination of phrases.
• Flexible grammar writing.
• Extended CFG (or feature grammar)
• Lexicon and morphology are optional.
• From keyword spotting to full wall-to-wall
  parsing.
• Flexible semantic structures.
• From simple key-value structures to nested
  features structures (trees).
• Unification is based on the domain model, which
  allows for more general rules.
• Grammar features may be copied into the
  semantics.
• Incremental interpretation.
• The semantic content may determine the end of an
  utterance.

• Robustness challenges
• Generation of many similar (correct) solutions.
• Use semantics for weeding out solutions.
• Generation of sub-optimal solutions.
• A chart parser is used to find all possible
  phrases with insertions allowed everywhere.
• The chart is searched for the best combination of
  phrases that (criteria in order of precedence)
• 1. Covers as many words as possible.
• 2. Uses as few top rules as possible.
• 3. Contains as few insertions as possible.
• 4. Contains as little non-agreement as possible.
• Efficiency.
• Incremental interpretation saves time.

and the building with eeh brick
eeh roof or I lets see am at it