You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
When utterance does not correspond to an actionable command, possibly due to stop words, verbal fillers or extraneous text which cannot be directly parsed.
In short, if a given phrase, e.g., ``open uh foo java'' is received, we attempt to repair the utterance.
At the user's discretion, or in case of ambiguity when the phrase has multiple plausible alternatives, we can either visually or verbally prompt the user to choose from a set of actionable phrases, e.g:
Did you mean (a) open file foo.java, (b) open folder foo/java, or (c) something else?'
To correct recognition errors, we apply Considine et al.'s~\cite{considine2022tidyparse} work on Tidyparse, which supports recognition and parsing of context-free and mildly context-sensitive grammars, and computing language edit distance. Tidyparse implements a novel approach to error correction based on the theory of context-free language reachability, conjunctive grammars and Levenshtein automata.
We use a SAT solver to find the smallest edit transforming a string outside the language to a string inside the language.
Only when the utterance is one or two tokens away from a known command do we attempt a repair.
The text was updated successfully, but these errors were encountered:
When utterance does not correspond to an actionable command, possibly due to stop words, verbal fillers or extraneous text which cannot be directly parsed.
In short, if a given phrase, e.g., ``open uh foo java'' is received, we attempt to repair the utterance.
At the user's discretion, or in case of ambiguity when the phrase has multiple plausible alternatives, we can either visually or verbally prompt the user to choose from a set of actionable phrases, e.g:
To correct recognition errors, we apply Considine et al.'s~\cite{considine2022tidyparse} work on Tidyparse, which supports recognition and parsing of context-free and mildly context-sensitive grammars, and computing language edit distance. Tidyparse implements a novel approach to error correction based on the theory of context-free language reachability, conjunctive grammars and Levenshtein automata.
We use a SAT solver to find the smallest edit transforming a string outside the language to a string inside the language.
Only when the utterance is one or two tokens away from a known command do we attempt a repair.
The text was updated successfully, but these errors were encountered: