Learning of dialogue states and language model of spoken information system

• US 20030091163A1
• Filed: 05/20/2002
• Published: 05/15/2003
• Est. Priority Date: 12/20/1999
• Status: Active Grant

• Current Assignee: British Telecommunications PLC (BT Group PLC)
• Original Assignee: British Telecommunications PLC (BT Group PLC)
• Inventors: Durston, Peter J, Edgington, Michael D, Attwater, David J
• Granted Patent: US 6,839,671 B2
• Time in Patent Office: Days
• US Class Current: 379/88.01
• CPC Class Codes: G10L 15/1822 Parsing for meaning understanding
• CPC Class Codes: G10L 15/183 using context dependencies, e.g. language models

Abstract

In this invention dialogue states for a dialogue model are created using a training corpus of example human-human dialogues. Dialogue states are modelled at the turn level rather than at the move level, and the dialogue states are derived from the training corpus.

The range of operator dialogue utterances is actually quite small in many services and therefore may be categorized into a set of predetermined meanings. This is an important assumption which is not true of general conversation, but is often true of conversations between telephone operators and people.

Phrases are specified which have specific substitution and deletion penalties, for example the two phrases ‘I would like to’ and ‘can I’ may be specified as a possible substitution with low or zero penalty. Thus allows common equivalent phrases are given low substitution penalties. Insignificant phrases such as ‘erm’ are given low or zero deletion penalties.

Claims

1. A method of classifying a plurality of sequences of symbols to form a plurality of sets of sequences of symbols comprising the steps of

a) determining a distance between each sequence and each other sequence in said plurality of sequences in dependence upon a set of insignificant symbol sequences and a set of equivalent symbol sequence pairs;

and

b) grouping the plurality of sequences into a plurality of sets in dependence upon said distances.

2. A method according to claim 1 in which the symbols are words transcribed from operator speech signals generated during an enquiry to a call centre.

3. A method according to claim 2, in which the words are transcribed from operator speech signals using a speaker dependent speech recogniser.

4. A method of generating a set of insignificant symbol sequences for use in the method of any one of the preceding claims, comprising the steps of

c) classifying a plurality of sequences of symbols into a plurality of sets;

d) for each of the sets, determining an optimal alignment between each sequence thereof and each other sequence in that set; and

e) allocating a symbol or sequence of symbols to the set of insignificant symbol sequences, the symbol or sequence of symbols having been deleted to obtain an optimal alignment between two sequences of a set.

5. A method of generating a set of equivalent symbol sequence pairs for use in the method of any one of claims 1 to 3, comprising the steps of

f) classifying a plurality of sequences of symbols into a plurality of sets;

g) determining an optimal alignment between each sequence in a set and each other sequence in that set; and

h) allocating a pair of symbols or sequences of symbols to the set of equivalent symbol sequences, the symbols or sequences of symbols having been substituted for each other to obtain an optimal alignment between two sequences of a set.

6. A method of generating a grammar for enquiries made to a call centre, using the plurality of sets of sequences of words generated according to claim 2 or claim 3, comprising the steps of

i) transcribing a plurality of enquiries according to which of the sets the sequences of words in the enquiry occur;

and

j) generating a grammar in dependence upon the resulting transcription.

7. A method of measuring the occurrence of particular types of telephone enquiry received in a call centre using the plurality of subsets of sequences of words generated according to claim 2.

13. A data carrier loadable into a computer and carrying instructions for causing the computer to carry out the method according to any one of claims 1 to 7.

8. An apparatus for classifying a plurality of sequences of symbols to form a plurality of sets of sequences of symbols comprising

a store for storing a set of insignificant symbol sequences;

a store for storing a set of equivalent symbol sequence pairs;

means for determining a distance between each sequence and each other sequence in said plurality of sequences in dependence upon the set of insignificant symbol sequences and the set of equivalent symbol sequence pairs; and

means for grouping the plurality of sequences into a plurality of sets in dependence upon said distances.

9. An apparatus according to claim 8, further comprising a speaker dependent recogniser for transcribing operator speech signals generated during an enquiry to a call centre and in which the determining means is connected to receive transcribed operator speech signals.

10. An apparatus for generating a set of insignificant symbol sequences for use by the apparatus of claim 8 or claim 9, comprising

a classifier for classifying a plurality of sequences of symbols into a plurality of sets;

alignment means for determining an optimal alignment for each of the sets between each sequence thereof and each other sequence in that set; and

means for allocating a symbol or sequence of symbols to the set of insignificant symbol sequences, the symbol or sequence of symbols having been deleted to obtain an optimal alignment between two sequences of a set.

11. An apparatus for generating a set of equivalent symbol sequence pairs for use by the apparatus of claim 8 od claim 9, comprising

a classifier for classifying a plurality of sequences of symbols into a plurality of subsets;

means for determining an optimal alignment between each sequence in a set and each other sequence in the that set; and

means for allocating a pair of symbols or sequences of symbols to the set of equivalent symbol sequences, the symbols or sequences of symbols having been substituted for each other to obtain an optimal alignment between two sequences of a set.

14. A data carrier loadable into a computer and carrying instructions for enabling the computer to provide the apparatus according to any one of claims 8 to 12.

12. An apparatus for generating a grammar for enquiries made to a call centre comprising

a store for storing a plurality of sets of sequences of words;

means for transcribing a plurality of enquiries according to which of the sets the sequences of words in the enquiry occur; and

means for generating a grammar in dependence upon the resulting transcription.

15. A method of classifying a plurality of sequences of words to form a plurality of sets of sequences of words, the method comprising the steps of:

transcribing the plurality of sequences of words from operator speech signals generated during an enquiry to a call centre;

determining a distance between each sequence of words and each other sequence of words in said plurality of sequences; and

grouping the plurality of sequences of words into a plurality of sets in dependence upon said distances.

16. A method according to claim 15, in which the words are transcribed from operator speech signals using a speaker dependent speech recogniser.

17. A method of generating a grammar for enquiries made to a call centre, using the plurality of sets of sequences of words generated according to claim 15 or 16, comprising the steps of

transcribing a plurality of enquiries according to which of the sets the sequences of words in the enquiry occur;

and

generating a grammar in dependence upon the resulting transcription.

18. A method of measuring the occurrence of particular types of telephone enquiry received in a call centre using the plurality of subsets of sequences of words generated according to claim 15.

19. An apparatus for classifying a plurality of sequences of words to form a plurality of sets of sequences of words comprising:

transcribing means for transcribing the plurality of sequences of words from operator speech signals generated during an enquiry to a call centre;

means for determining a distance between each sequence and each other sequence in said plurality of sequences in dependence upon the set of insignificant symbol sequences and the set of equivalent symbol sequence pairs; and

means for grouping the plurality of sequences into a plurality of sets in dependence upon said distances.

20. An apparatus according to claim 19, wherein the transcribing means further comprise a speaker dependent recogniser for transcribing the operator speech signals generated during an enquiry to a call centre and in which the determining means is connected to receive transcribed operator speech signals.