Rule-based dialog state tracking

Patent metadata

Field	Value
Publication number	US-2017228366-A1
Application number	US-201615017305-A
Country	US
Kind code	A1
Filing date	Feb 5, 2016
Priority date	Feb 5, 2016
Publication date	Aug 10, 2017
Grant date	—

Abstract

The present disclosure relates dialog states, which computers use to internally represent what users have in mind in dialog. A dialog state tracker employs various rules that enhance the ability of computers to correctly identify the presence of slot-value pairs, which make up dialog states, in utterances or conversational input of dialog. Some rules provide for identifying synonyms of values of slot-values pairs in utterances. Other rules provide for identifying slot-value pairs based on coreferences between utterances and previous utterances of dialog sessions. Rules are also provided for carrying over slot-value pairs from dialog states of previous utterances to a dialog state of a current utterance. Yet other rules provide for removing slot-value pairs from candidate dialog states, which are later used as dialog states of utterances.

First claim

What is claimed is:

1 . A computer-implemented system comprising:
one or more processors and memory, coupled with the one or more processors, having executable instructions embodied thereon, which, when executed by the one or more processors cause the one or more processors to perform operations comprising: maintaining a history of predicted dialog states, each predicted dialog state in the history comprising a set of slot-value pairs assigned to an utterance of a plurality of utterances of a dialog session; string-matching a plurality of slot-value pairs from an ontology to a given utterance of the dialog session, said string-matching including, for each slot-value pair:
determining a failed match between a string that represents a value of the slot-value pair and a sub-string of the given utterance;
based on the determining the failed match, identifying a synonym of the value; and
determining a successful match between a string that represents the synonym of the value of the slot-value pair and the sub-string of the given utterance;
adding each slot-value pair of the plurality of slot-value pairs to a candidate dialog state of the given utterance based on the successful match for the slot-value pair; removing at least one of the plurality of slot-value pairs from the candidate dialog state; and designating the candidate dialog state as the predicted dialog state of the given utterance in the history.

2 . The computer-implemented system of claim 1 , wherein for each slot-slot value pair, the synonym is selected from a list of synonyms assigned to the value.

3 . The computer-implemented system of claim 1 , further comprising:
matching one or more sub-strings of the given utterance to a template that specifies a type of coreferenced slot-value pair in the ontology; and adding the coreferenced slot-value pair to the candidate dialog state based on determining that the coreferenced slot-value pair matches the type specified by the template.

4 . The computer-implemented method of claim 1 , further comprising determining that a historic slot-value pair is present in the set of slot-value pairs for at least one utterance that occurs prior to the given utterance in the dialog session; and
adding the historic slot-value pair to the predicted dialog state of the given utterance based on the determining.

5 . A computer-implemented method for determining dialog states of utterances comprising:
receiving a given utterance of a dialog session, the given utterance represented as computer readable text; identifying a sub-string of the computer readable text for string-matching to a slot-value pair of an ontology; determining, by the string-matching, a successful match between the sub-string and a string that represents a synonym of a value of the slot-value pair; adding the slot-value pair to a candidate dialog state of the given utterance based on the successful match; and assigning a predicted dialog state to the given utterance in a history of dialog states of the dialog session based on the candidate dialog state.

6 . The computer-implemented method of claim 5 , wherein the string-matching comprises:
identifying an initial synonym of the value; determining a failed match between the sub-string and an initial string that represents the initial synonym and; and based on the determining of the failed match, attempting the determining using the string that represents the synonym.

7 . The computer-implemented method of claim 5 , wherein the string-matching comprises:
selecting the synonym of the value from a list of synonyms assigned to the value; based on the selecting of the synonym, using the string in the determining the successful match.

8 . The computer-implemented method of claim 5 , wherein the synonym corresponds to a plurality of words and the string-matching comprises:
determining a failed match between the sub-string of the given utterance and an initial string that represents a first sequence of the words; and performing the determining of the successful match using the string, wherein the string represents a second sequence of the words of the synonym.

9 . The computer-implemented method of claim 5 , wherein the string-matching comprises:
selecting the synonym of the value from a list of synonyms assigned to the value, the list indicating a part-of-speech of the synonym; wherein the determining of the successful match includes based on the selecting of the synonym, determining that the part-of-speech of the synonym matches a part-of-speech corresponding to the sub-string.

10 . The computer-implemented method of claim 5 , further comprising, based on an analysis of the ontology:
enforcing part-of-speech matching in the string-matching for a first of a plurality of slot-value pairs; and refraining from enforcing part-of-speech matching in the string-matching for a second of the plurality of slot-value pairs.

11 . The computer-implemented method of claim 5 , further comprising:
determining a match between a first string that represents a value of a given slot-value pair and a sub-string of a second string that represents another slot-value pair that is in the candidate dialog state; refraining from including the given slot-value pair in the predicted dialog state of the given utterance based on the determining of the match.

12 . The computer-implemented method of claim 5 , further comprising:
matching one or more sub-strings to a template that specifies a type of coreferenced slot-value pair in the ontology; and adding the coreferenced slot-value pair to the candidate dialog state based on determining that the coreferenced slot-value pair matches the type specified by the template.

13 . The computer-implemented method of claim 5 , further comprising
determining that a historic slot-value pair is present in the history of dialog states for at least one utterance that occurs prior to the given utterance in the dialog session; and adding the historic slot-value pair to the predicted dialog state of the given utterance based on the determining.

14 . The computer-implemented method of claim 5 , further comprising:
identifying a historic slot-value pair in the history of dialog states for at least one utterance that occurs prior to the given utterance in the dialog session; determining the identified historic slot-value pair has a same slot as an existing slot-value pair in the candidate dialog state and a different value than the existing slot-value pair; based on the determining of the identified historic slot-value pair, refraining from including the historic slot-value pair in the predicted dialog state of the given utterance.

15 . The computer-implemented method of claim 5 , further comprising:
determining at least two slot-value pairs in the candidate dialog state correspond to a merged slot-value pair in the ontology; based on the determining of the at least two slot-value pairs, adding the merged slot-value pair to the candidate dialog state and removing the at least two of the slot-value pairs from the candidate dialog state.

16 . The computer-implemented method of claim 5 , comprising determining the predicted dialog state of the given utterance using a fuzzy matching algorithm that uses the candidate dialog state to generate a feature of the fuzzy matching algorithm to select slot-value pairs for the predicted dialog state.

17 . The computer-implemented method of claim 5 , further comprising outputting a response to the given utterance to a user device of a user associated with the

Assignees

• Adobe Systems Inc

Inventors

• Bui Trung Huu
• Bui Hung Hai
• Dernoncourt Franck

Classifications

• G06F40/284

  Lexical analysis, e.g. tokenisation or collocates · CPC title

• G06F40/35 Primary

  Discourse or dialogue representation · CPC title

• G06F40/247

  Thesauruses; Synonyms · CPC title

• G10L15/22 Primary

  Procedures used during a speech recognition process, e.g. man-machine dialogue · CPC title

• G06F17/277

  Physics · mapped topic