MXPA97003425A - System and automated method of calling telephone - Google Patents

Abstract

The present invention relates to an automated system for classifying tasks that operates in a task objective of a user, the task objective that is expressed in the natural speech of the user, is characterized in that it comprises: means for determining a plurality of fragments speech from a set of speech pronunciations, each of the speech fragments that relate to an objective of a predetermined set of task objectives, a means to recognize at least one of the speech fragments in the speech of input of the user, the input recognizing means having as input the speech fragments determined by the means of determining speech fragments, and a means of interpretation sensitive to an input of the recognized speech fragments for making a classification decision based on speech fragments recognized for a goal of the set of task objectives, predetermined

Description

SYSTEM AND AUTOMATED METHOD OF TELEPHONE CALL ROUTING BACKGROUND OF THE INVENTION A. FIELD OF THE INVENTION This invention relates to speech processing, and more particularly to a system and method for automated call routing related to the performance of one or more desired tasks.

B. TECHNICAL BACKGROUND In telephone communications it is well known that the receiver of a call, particularly in a commercial establishment, uses an automated call routing system that initially presents the calling party with a menu of routing choices from which the caller that you select by pressing the particular numbers on the numeric keypad associated with the caller's phone, the routing system that recognizes a tone associated with the key pressed by the caller. REF: 24679 is also known to offer these callers a choice between pressing a number on the numeric keypad to select a menu choice or say the number, for example, "press or say one for customer service". In the particular context of telephone services, it is also known to use an automatic routing system to select among billing options for a call placed on this service. For example, in the case of a long-distance telephone call that is going to be billed to a different number from the origin, a menu can be presented to the calling party on the form, "please say" charge "," credit card ". calling "," third number "or" operator "". While these routing systems work reasonably well in cases where the number of routing choices is small, if the number of selections exceeds approximately 4 or 5, multi-level menus become necessary in general. These multi-scale menus are very unpopular with callers, from the perspective of a typical caller, the time and effort required to navigate through several layers of menus to achieve a desired goal may seem endless. Equally important, from the perspective of both the caller and the receiver, the percentage of successful routings through this multiscaled menu structure may be quite low, in some cases, less than 40 percent. Stated differently, in some circumstances, more than 60 percent of calls that enter this multiscaled menu structure could either be terminated without the caller reaching the desired goal or being abandoned to an attendant (or another station handled by default) . To address these limitations in the prior art, it would be desirable to provide a system that can understand and act on the spoken input of people. Traditionally, in these speech understanding systems, meaningful words, phrases and structures have been constructed manually, which involves a lot of work and leads to fragile systems that are not strong in real environments. Therefore, a main objective would be a speech understanding system that is trainable, adaptable and strong, that is, a system to automatically learn the language for its task.

BRIEF DESCRIPTION OF THE INVENTION An automated system and method for routing telephone calls that operates on a calling routing target of a calling party expressed in the natural speech of the calling party. The system incorporates a speech recognition function, for which a call routing target of natural speech of the calling party provides an input, and which is trained to recognize a plurality of meaningful phrases, each phrase relating to a specific goal of call routing. In recognition of one or more of these significant phrases in the incoming speech of the calling party, an interpretation function then acts on this request of the calling party's routing objective, depending on the trust function incorporated in the interpretation function, either to complement the requested routing target of the calling party or to enter a dialogue with the calling party to obtain additional information from which a sufficient level of trust can be achieved to implement that routing objective. Significant phrases are determined by a grammatical inference algorithm that operates on a predetermined body of speech pronunciations, each pronunciation associated with a specific call routing goal, and where each pronunciation is marked with its associated goal of routing calls. Each meaningful phrase developed by the grammatical inference algorithm can be characterized as having both a Mutual Information value and an Outstanding Feature value (relative to an associated routing target) above a predetermined threshold.

BRIEF DESCRIPTION OF THE DRAWINGS Figure 1 provides illustrative examples of false detection and not understood by a classifier for an automated call routing system based on the use of "meaningful phrases".

Figure 2 provides illustrative examples of the correct detection by a classifier for an automated call routing system based on the use of "meaningful phrases".

Figure 3 represents an illustrative example of the advantage provided by the classification parameter of "significant phrases" of the system of the invention.

Figure 4 presents, in the form of a block diagram, the structure of the system of the invention.

Figure 5 represents the methodology of the invention in the form of a flowchart.

Figure 6 provides illustrative examples of "significant phrases" determined in accordance with the invention.

DETAILED DESCRIPTION OF THE INVENTION The following discussion will be presented partially in terms of algorithms and symbolic representations of operations in data bits within a computer system. As will be understood, these algorithmic descriptions and representations are a means ordinarily used by those skilled in computer processing techniques to convey the essence of their work to other experts in the art. As used herein (and in general) an algorithm can be seen as a self-contained sequence of steps leading to a desired result. The steps generally comprise manipulations of physical quantities. Usually, although not necessarily, these quantities take the form of electrical or magnetic signals capable of being stored, transferred, combined, compared and manipulated in another way. For convenience of reference, as well as to agree with common usage, these signals will be described from time to time in terms of bits, values, elements, symbols, characters, terms, numbers, or the like. However, it should be emphasized that these similar terms and terms are going to be associated with appropriate physical quantities, these terms being only convenient marks applied to those quantities. It is also important that the distinction between the method of operations and to operate a computer, and the computation method itself, should be kept in mind. The present invention relates to methods for operating a computer in the processing of electrical or other physical signals (eg, mechanical, chemical) to generate other desired physical signals. For clarity of explanation, the illustrative embodiment of the present invention is presented as comprising individual functional blocks (including functional blocks labeled or marked "processors"). The functions that these blocks represent can be provided through the use of either shared or dedicated physical components of a computer, including, but not limited to, physical components of a computer capable of executing programming elements. For example, the functions of the processors presented in Figure 4 can be provided by a shared, individual processor. (The use of the term "processor" should not be considered as referring exclusively to the physical components of the computer capable of executing programming elements). Illustrative embodiments may comprise physical microprocessor and / or digital signal processor (DSP) components, such as the AT & T DSP16 or DSP32C, read only memory (ROM). ) to store the programming elements that perform the operations discussed later, and random access memory (RAM) to store the results. It is also possible to provide modalities with very large-scale integration physical components (VLSI, for its acronym in English), as well as a VLSI circuit system for use in combination with a general purpose DSP circuit. A fundamental objective of the invention is a call routing service that changes the weight of the understanding of a specialized vocabulary from the calling party to the receiver of the call. Thus, in a generalized embodiment, the invention is represented as a call routing system having the following characteristics: First, a caller entering the system will be presented with a greeting similar to "How can it help?" After the caller responds to that greeting with a natural speech statement from the caller's target (such as a routing target), the system is able to either classify the caller's request into one of a number of predefined target routings and implement that routing, or transfer the caller to an operator where the request does not conform to one of the predefined target routings or the system was unable to understand the caller's request. To minimize erroneous routing, as well as transfers to an operator position when a caller actually needs one of the predefined routing objectives, the system also incorporates a dialog function to obtain additional information from the caller to improve the confidence of the system in a Classification of a caller's objective. Subsequently, a preferred embodiment of the invention will be described based on access by a caller to a telephone service provider. In this mode, the routing goals for a caller can include call billing options (eg, charge, third person), dialing information to the desired number, billing issues, credit requests (such as for a wrong or wrong number) marked), area codes, etc.

I. Description of the preferred modality In traditional telephone switching systems, a user is often asked to know the separate numbers and / or dialing patterns to access the different services provided by a telephone company, as well as possibly having to navigate in a system operated by a telephone company. menus that then direct the user to the desired objective. In the system and method of the invention, the user is able to access a central number and the user's goal will be implemented by the telephone company based on its content. An example of this routing based on its content would be where a caller responds to a "can help" warning with the desire to change the charges, where the appropriate action is to connect the caller to an automated subsystem that processes the charge. of the calls. Another example would be an answer from the caller of I have a problem to tend my invoice, case in which the caller must be connected to the telephone company's trade office. In this way, the system needs to understand the spoken language to the extent of routing the call appropriately.

A. Basic approach The basic construction of this system has been described by one of the inventors in Gorin, A. "On automated language acquisition", J. Acoust, Soc. Am, 97 3441-3461, (June 1995) [later referred to as Gorin 95], which is incorporated herein and becomes a part of it. A number of considerations from this basic approach are the material for the system and the method of the invention. Subsequently, certain considerations will be briefly reviewed. As a preface to that review, it should be noted that the approach in Gorin 95 is directed to either a text or speech input medium and the classification parameter for the determination of an association between the input text or speech, and a The objective of a set of predefined routing objectives is implemented as outgoing words derived from a body of speech pronunciations that have routing goals, associated, marked. In the adaptation of that system described herein, the input means is only the speech and the classification parameter is implemented as the meaningful phrases derived from a body of speech pronunciations having associated, marked routing objectives. The main point of the approach here is a database of a large number of pronunciations, each of which is related to one of a predetermined set of routing objectives. This database forms an entry to a classification parameter algorithm. Preferably, these pronunciations will be extracted from the user's actual responses to a "how can it help?" (or similar words for the same effect). Then, each pronunciation is transcribed and labeled or marked with one of the predetermined set of routing objectives. The illustrative pronunciations of the database used by the inventors are as follows: In, I wish to change the charges I am simply disconnected from this number I was trying to hang up This is trying to call Mexico Charge this to my home phone In a related article of co-authorship by one of the inventors [Gorin, AL, Hanek, H., Rose, R. and Miller, L., "Spoken Language Acquisition for Automated Cali Routing ", in Proceedings of the International Conference on Sapoken Language Processing (ICSLP 94), Yokohama (Sept. 18-22, 1994)] [subsequently Gorin 94A], it is pointed out that the distribution of the routing objectives in this database can be substantially deviated. The implications of this deviation are well taken into account in determining the particular set of routing objectives that will be supported on an automated basis by the system of the invention. A principle of outstanding characteristic as it relates to the system of the invention has been defined in another article of co-authorship by one of the inventors [Gorin, A. L. Levinson, S. E. and Sankar, A. "An Experiment in Spoken Language Acquisition," IEEE Trans. On Speech and Audi or, Vol. 2, No. 1, Part II, pp. 224-240 (January 1994)] [subsequently Gorin 94]. Specifically, the salient feature of a word is defined as the content of that word's information for the task under consideration. It can be interpreted as a measure of how much is the meaning of that word for the task. The salient feature can be distinguished from and compared to Shannon's traditional information content, which measures the uncertainty of the occurrence of that word. As is known, this traditional information content can be estimated from examples of language, while an estimation of the salient feature requires both language and its extralinguistic associations.

B. Adaptation of the basic approach As noted previously, the basic Gorin 95 approach (which has been incorporated herein by reference) uses as a classification parameter words from test speech pronunciations that have an outgoing association with target routings, particular . A significant point of departure for the methodology of the invention from the base approach lies in the use of the significant phrases as the classification parameter, a clear improvement on that basic approach. Before describing the nature of that improvement, or the methodology to determine those significant phrases, it is useful to define two types of errors experienced in this automated call routing system and a related "successful" concept: False detection of the routing objective occurs when a (significant) outgoing phrase related to a routing target is detected in the incoming caller's speech when the caller's actual request is directed to another routing target. The probability that this false detection will occur later will be referred to by the designation: PFD. Missing detection of the routing objective occurs when the incoming caller's speech is directed to that routing target and none of the meaningful phrases associated with that routing target are detected in the incoming speech. The probability of this detection not being understood will be referred to later by the designation: PMD. Coverage for a routing objective refers to the number of successful translations by the system of a request for a routing target to that routing target relative to the total number of incoming requests for that routing target. As an illustrative example, a routing target for which 60 successful translations of the 100 input requests for that routing target occur will experience a 60% coverage. It is noted that the Coverage = 1-PMD. Of the two types of errors defined above, one is significantly more "expensive" than the other in an automated call routing device. The consequence of a false detection error is the routing of a caller to a different routing target than that requested by the caller. This result is at least very annoying for the caller. The possibility also exists such that an error could result in a direct cost to the system provider, an annoying customer or potential customer that is classified here as an indirect cost, although some non-system errors result from the caller connecting to a target incorrect routing. The consequence of a misunderstood detection error, on the other hand, is simply the routing of the caller to a default operator position and the cost is only the lost cost of the opportunity not to handle that particular call on an automated basis. In this way, while ideally the probabilities of both the missed detection and the false detection would be close to zero, it is more important that this objective be performed for the false detection errors. As will be seen later, there are circumstances where transactions must be made between minimizing one or other of these error probabilities, and this principle will apply in these circumstances. Figure 1 provides several illustrative examples of False Detections and Detections Not Understanding from the database of speech pronunciations used by the inventors. While the basis for the error in each of these examples is believed to be largely self-explanatory, the error in the first example in each set will be briefly described. In the first example under false detection, the significant phrase is I-NEED-CREDIT. EOS, and in this way this phrase would have been classified as a request for credit. However, for the reading of the full pronunciation, it is apparent that the caller actually needs to be transferred to another company (the company that receives this request which is AT &; T). In this first example under detections not understood, there are no significant phrases identified in the pronunciation (and therefore there are no bases to classify the caller's objective), although it is apparent from the reading of the pronunciation that the caller seeks a credit billing. As a comparative illustration, Figure 2 shows several examples of correct detections of a billing credit objective from meaningful phrases in the input speech. There are two significant advantages of the methodology of the invention in using meaningful phrases as the classification parameter on the use of outgoing words in the basic approach described in Gorin 95. First, with the use of words as the classification parameter, the choices of words to detect a given routing objective can be highly constrained in order to achieve an extremely low probability of false detection, that is, the use of only words that have a probability close to 100% of predicting the proposed routing target, and therefore the coverage for this routing objective is probably very low, leading to a high probability of detection errors not understood. With the significant phrases as a classification parameter, on the other hand, both low probability of false detection and low detection probability not included are achieved.

Figure 3 provides an illustrative example of this advantage. This Figure shows the classification ratio and coverage for an example routing goal, billing credit, as the phrase used for the ranking parameter to grow in length and / or complexity. The classification ratio is defined as the probability of the routing objective (CREDIT) that has been requested, given the occurrence of the selected phrase in the incoming speech (ie, P (CREDIT phrase). Coverage is defined as the probability of the selected phrase that appears in the incoming speech, given that the desired routing objective has been requested (CREDIT) In the phrase column, the parentheses surrounding a series of terms separated by "" indicate one of those terms that appear in the position indicated by other terms in that row The nomenclature "F (Wrong)" indicates a grammar fragment that surrounds the word "wrong", the phrase in the fourth row of that column that is representative of that fragment of grammar that surrounds a salient word The designation "previous" indicates a transfer to everything in the preceding line, and finally, the abbreviation "eos" indicates "end of l statement. " The second area of improvement relates to the speech recognition function that operates in the incoming speech of the caller. Essentially, in the present state of the art of speech recognition systems, the greater the fragment of speech presented to this speech recognizing device, the greater the probability of a correct recognition of that speech fragment. In this way, the speech recognizing device programmed to recognize one of a set of outgoing words can be expected to fail in its task in a significant way more frequently than a device programmed to recognize meaningful phrases, comprising two or more words.

C. Configuration of the System of the Invention Figure 4 shows in block diagram form, the essential structure of the invention. As can be seen from the Figure, this structure comprises two related subsystems: subsystem 1 for generation of significant sentences and subsystem 2 for classification of incoming speech. As already described, the significant phrase generation subsystem 1 operates in a database of a large number of pronunciations each of which is related to one of a predetermined set of routing objectives, where each pronunciation is tagged or tagged. with its associated goal of routing. The operation of this subsystem is essentially carried out by the processor of significant phrases which produces as an output a set of significant phrases having a probabilistic relationship with one or more of the set of predetermined routing objectives with which the incoming speech pronunciations. The operation of the processor 10 of meaningful phrases is generally determined according to a grammatical inference algorithm, described below. The operation of the incoming speech classification subsystem 2 begins with the introduction of a caller routing target request, in the natural speech of the caller, to the recognizing device 15 of the incoming speech. This input speech recognizing device can be of any known design and performs the function of detecting, or recognizing, the existence of one or more significant phrases in the input speech. As can be seen in the figure, the significant phrases developed by the generation subsystem 1 of significant phrases are provided as an input to the speech recognizing device 15. The output of speech recognition device 15, which will comprise the significant phrase (s) recognized (s) appearing in the request of the caller's routing target, the interpretation module 20 is provided. This interpretation module applies a confidence function, based on the probabilistic relationship between the significant phrase (s) recognized and selects the routing objectives, and makes a decision on whether it implements the objective of chosen routing (in the case of high confidence) with an announcement to the caller that this objective is being implemented, or looking for additional information and / or caller confirmation via a module 25 of the dialogue with the user, in the case of low levels of trust. Based on the caller's feedback in response to this question generated by the interpretation module 20, the interpretation module again applies the confidence function to determine whether a decision to implement a particular routing objective is appropriate. This feedback process continues until either a decision is made to implement a particular routing objective or a determination is made that is probably not that decision, in which case the caller is left to an operator position. As will be apparent in this manner, the significant phrases developed by the significant phrase generating subsystem 1 are used by the recognizing device 15 of the input speech, to define the phrases that a recognizing device is programmed to recognize, and by the module 20 of interpretation, both to define the routing objectives related to the input of significant phrases of the speech recognizing device 15 and to establish the level of confidence for a relationship of this significant phrase (s) of input to a particular purpose of routing. The methodology of the invention is illustrated graphically in the flowchart of Figure 3. It will be appreciated from the above discussion that the process depicted in Figure 3 is widely separated between the generation functions 100 significant phrases and the functions 200 of automated call routing. Considering first the functions of generation of significant phrases, a database of speech pronunciations marked with a routing objective is provided in step 105, and that database is entered in step 110 to extract a set of n- grams. The mutual information of these n-grams is determined in step 115 and the determined MI values are thus compared to a predetermined threshold in step 120, the n-grams having an MI value below the threshold are discarded and those which are above the threshold are operated by the measurement step 125 of the protruding feature. In step 130 those values of outstanding characteristics are compared with an outstanding, predetermined characteristic threshold and those n-grams having an outstanding characteristic value above the threshold are discarded, the n-grams passing the characteristic threshold test Outstanding is stored in step 135 as the set of meaningful phrases. The automated call routing functions begin with step 205 where a caller entering the routing device is provided with a greeting similar to "How can it help you?". The caller's response to the greeting is obtained in step 210 and that response is examined for the presence of one or more significant phrases.

As can be seen in the figure, the set of meaningful phrases will have been provided as an input to step 215 as well as to step 220. In step 220, an attempt is made to classify the caller's goal based on the (s) significant phrase (s) found in the incoming caller's speech. It is called a confidence factor in step 225, and a decision is made as to whether there is a sufficient level of confidence to implement the classified objective. If so, the objective is implemented in step 245. If the trust function dictates either that a user claim or more information is needed, a dialogue is conducted with the user in step 230. This dialogue will typically begin with a question to the user of the form "Do you need?". If the user responds in negative to this question, in step 235, the process returns to step 210 with a new declaration by the user of his request. A "yes" or silent response from the user moves the process along step 240 where consideration is given whether he needs other information to carry out the objective that was not in the input speech, for example, the number for give credit in the case of a credit request. Where additional information is needed, the process returns to step 230 for further dialogue with the user. If no additional information is required, the goal of step 245 is implemented.

D. Determination of Significant Phrases As will be understood in this point, a fundamental focus of this invention is that of providing a device that learns to understand and act on the spoken input. It will be apparent that the ultimate goal of this speech understanding system will be to extract the meaning of the speech signal. Furthermore, as shown in Gorin 95, for systems that understand spoken language, the semantic aspects of communications are highly significant. Not surprisingly, these semantic considerations are fundamental to the development of the meaningful phrases used as the classification parameters in the system of the invention. The determination of the meaningful phrases used by the invention is found in the concept of combining a measure of the common state of the words and / or the structure within the language, that is, as co-occur frequently groupings of things, with a measure of the meaning for a task defined for this grouping. In a preferred embodiment of the invention, that measurement of the common state within the language is manifested as the mutual information in n-grams derived from a training speech pronunciation database and the utility measurement for a task is manifested as a measurement of the outstanding feature. Other manifestations of these general concepts will be apparent to those skilled in the art. As it is known, mutual information ("MI"), which measures the probability of co-occurrence for two or more words, comprises only the language itself. For example, given War and Peace of the original Russian, one could compute the mutual information for all possible pairs of words in that text without yet understanding a word of the language in which it is written. In contrast, the computation of the salient feature comprises both the language and its extralinguistic associations for a device environment. Through the use of this MI combination and an outstanding feature factor, phrases can be selected that have both a positive MI (indicating a relatively strong relationship between the words comprising the phrase) and a high outstanding feature value. The determination of significant sentences is implemented in a grammatical inference algorithm ("Gl"). This Gl algorithm searches for the set of phrases that occur in the training database using as a selection criteria both mutual information (within the language) of a sentence and a measurement of the outstanding characteristic of the sentence for a low task consideration. It is your general form, the algorithm of Gl carry out the following steps: 1. n = number of words in a sentence to be evaluated 2. Set n = 1 3. Generate a list of sentences in the length n in the database of training 4. Evaluate the common state of those sentences (as a function of frequency / probability) and the salient characteristic of the sentences. 5. Select the subset according to a predetermined threshold. 6. Generate the list of sentences of length n + 1 by extending the set of sentences in length n. 7. Evaluate the mutual information and the salient characteristic for the phrases generated in step 6. 8. Select the subset according to the predetermined threshold. 9. Set n = n + 1 10. Go to step 6.

The algorithm can be carried out at any desired n level and the selection threshold can be varied upwards or downwards with successive interactions to vary the output of the algorithm. That output will be a set of meaningful phrases. An example output of this algorithm is shown in Figure 7. In that figure, the example significant sentences are shown together with the value of MI, the outstanding feature value, and the action to which the meaningful phrase is addressed in the probability value shown. It is noted that the maximum probability (sometimes designated PMAX is used as a substitute for the salient feature value.) This maximum probability is defined as the maximum of the probability distribution of each action, given the phrase under consideration, probabilities that are determined by a sub-step of the Gl algorithm.

E. Illustrative Example of System Operation As an end to further understand the invention, two illustrative dialogs are presented after the operation of the system of the invention. In section one, the system immediately understands the entry, based on the associations learned between the content of the statement and the selected goal of routing the invoicing of the third number. Hi, how can I help you? "I need to charge this call to my home phone" Do I need to bill this call to a third party? . Please enter the invoice number: As can be seen, the absence of a caller response during the pause between the second and third questions of the system was interpreted as approval to the second question. In session two, the system finds an ambiguous entry, which is resolved via dialogue. The hesitant response is indicated, based on the system's confidence model. It is also noted that the second input comprises both the negative reinforcement (no) and additional clarification information Hello, how can you help? "Can I load this call please?" Do you need to bill this call to a third party? "No, I want to put this on my universal card" Do you need to invoice this loan to a credit card? Please enter the number of the card: CONCLUSION A new automated call routing system has been described that performs the search function for a classification parameter in natural speech, a classification parameter that manifests itself as a set of significant phrases, which define themselves in terms of mutual information and an outstanding feature factor. Then, depending on whether one or more particular significant phrases are found in the input speech, a decision rule is applied to classify the appropriate routing target. An important advantage of the system and method of the invention is that the vocabulary and grammar for the system are unimpeded, being acquired by the system during the course of carrying out its task. In contrast, in the prior art systems, the outgoing vocabulary words and their meanings are explicitly provided to the system. While the language recognition systems of the prior art have been constantly made by hand and lack strength, the system and method of the invention brings with it an automated training procedure for the language recognition function, automated procedures for determining the Significant phrases in test speech procedures marked with objective, associated routings, then find these phrases significant in the incoming caller's speech (possibly including a confidence improvement dialogue with the caller) increase the goal routing requested by the caller. caller Although the present embodiment of the invention has been described in detail, it should be understood that various changes, alterations and substitutions may be made herein without departing from the spirit and scope of the invention as defined by the appended claims.

It is noted that in relation to this date, the best method known by the applicant to carry out the present invention is that which is clear from the present description of the invention. Having described the invention as above, the content of the following is claimed as property:

Claims (45)

1. An automated task classification system that operates on a task target of a user, the task goal that is expressed in the user's natural speech, is characterized in that it comprises: a means for determining a plurality of speech fragments from a set of speech pronunciations, each of the speech fragments that relate to an objective of a predetermined set of task objectives; means for recognizing at least one of the speech fragments in the input speech of the user, the input recognizing means having as input the speech fragments determined by the means of determining speech fragments; and a means of interpretation sensitive to an input of the recognized speech fragments to make a classification decision based on speech fragments recognized for a purpose of the set of predetermined task objectives.

2. The automated task classification system according to claim 1, characterized in that the interpretation means further includes a means of dialogue to allow the system to obtain a feedback response from the user.

3. The automated task classification system according to claim 2, characterized in that the feedback response includes additional information with respect to the user's initial input speech.

4. The automated task classification system according to claim 2, characterized in that the feedback response includes confirmation with respect to at least one of the set of task objectives determined in the classification decision.

5. The automated task classification system according to claim 1, characterized in that the relationship between the determined speech fragments and the predetermined set of task objectives includes a measure of the usefulness of one of the speech fragments to one of the objectives of task, default, specified.

6. The automated task classification system according to claim 5, characterized in that the utility measure is manifested as a measure of the outstanding feature.

7. The automated task classification system according to claim 6, characterized in that the measure of outstanding characteristic is represented as a conditional probability of the task target that is requested given an appearance of the speech fragment in the input speech, the conditional probability which is a higher value in a distribution of conditional probabilities over the set of predetermined task objectives.

8. The automated task classification system according to claim 6, characterized in that each of the plurality of speech fragments determined by the means for determining has a measure of outstanding characteristic that exceeds a predetermined threshold.

9. The automated task classification system according to claim 1, characterized in that the relationship between the determined speech fragments and the predetermined set of task objectives includes a measure of the common state within a language of the speech fragments.

10. The automated task classification system according to claim 9, characterized in that the measurement of the common state is manifested as a measure of mutual information.

11. The automated task classification system according to claim 10, characterized in that each of the plurality of speech fragments determined by the determining means has a measure of mutual information that exceeds a predetermined threshold.

12. The automated task classification system according to claim 1, characterized in that the means of interpretation includes a trust function with respect to the classification decision.

13. The automated task classification system according to claim 1, characterized in that the user input speech represents a request for a specified objective of the set of predetermined task objectives.

14. The automated system for classifying tasks according to claim 1, characterized in that the incoming speech is sensitive to a question of the system in a way "How can it help?".

15. The automated task classification system according to claim 1, characterized in that each of the speech pronunciations are directed to a target of the set of task objectives, predetermined, and each pronunciation is tagged or tagged with a task goal the which one is directed

16. An automated call routing system that operates on a routing target of a calling party, the routing target that is expressed in the natural speech of the calling party, characterized in that it comprises: a means for determining a plurality of fragments of speaks from a set of speech pronunciations, each of the speech fragments that relates to an objective of a predetermined set of routing objectives; means for recognizing at least one of the speech fragments in the input speech of the calling party, the input recognition means having as input the speech fragments determined by the means of determining the speech fragments; and a means of interpretation sensitive to an input of the recognized speech fragments to make a classification decision based on the speech fragments recognized for a purpose of the set of predetermined routing objectives.

17. The automated call routing system according to claim 16, characterized in that the means of interpretation further includes a means of dialogue to allow the system to obtain a feedback response from the calling party.

18. The automated call routing system according to claim 17, characterized in that the feedback response includes additional information with respect to the initial input speech of the calling party.

19. The automated call routing system according to claim 17, characterized in that the feedback response includes confirmation with respect to at least one objective of the set of routing objectives determined in the classification decision.

20. The automated call routing system according to claim 16, characterized in that the relationship between the determined speech fragments and the predetermined set of routing objectives includes a measure of the usefulness of one of the speech fragments to a specified target of the routing objectives, predetermined.

21. The automated call routing system according to claim 20, characterized in that the utility measure is manifested as a measure of the remarkable characteristic.

22. The automated call routing system according to claim 21, characterized in that the measure of the remarkable characteristic is represented as a conditional probability of the routing target that is requested given an appearance of the speech fragment in the input speech, the probability conditional which is a higher value in a distribution of the conditional probabilities over the set of routing objectives, predetermined.

23. The automated call routing system according to claim 21, characterized in that each of the plurality of speech fragments determined by the determining means has a measure of the outstanding feature exceeding a predetermined threshold.

24. The automated call routing system according to claim 16, characterized in that the relationship between the determined speech fragments and the predetermined set of routing objectives includes a measure of the common state within a language of the speech fragments.

25. The automated call routing system according to claim 24, characterized in that the measurement of the common state is manifested as a measure of the mutual information.

26. The automated call routing system according to claim 25, characterized in that each of the plurality of speech fragments determined by the determining means has a measure of mutual information that exceeds a predetermined threshold.

27. The automated call routing system according to claim 16, characterized in that the means of interpretation includes a trust function with respect to the classification decision.

28. The automated call routing system according to claim 16, characterized in that the input speech of the user represents a request for a specified target of the set of predetermined routing objectives.

29. The automated call routing system according to claim 16, characterized in that the incoming speech is sensitive to a system question of the form "How can it help?" .

30. The automated call routing system according to claim 16, characterized in that the speech pronunciations are directed to an objective of the set of predetermined routing objectives and each pronunciation is marked with the purpose of routing to which it is addressed.

31. A method for the automated classification of tasks that operates in a task objective of a user, the task objective that is expressed in the natural speech of the user, characterized in that it comprises the steps of: determining a plurality of speech fragments of a set of speech pronunciations, each of the speech fragments that relates to an objective of a predetermined set of task objectives; recognize at least one of the speech fragments in the user's input speech; and making a classification decision in response to an entry of the speech fragments recognized as being for a purpose of the set of predetermined task objectives.

32. The automated task classification method according to claim 31, characterized in that it includes the additional step of entering a dialogue with the user to obtain a feedback response from the user.

33. The automated task classification method according to claim 32, characterized in that the feedback response includes additional information with respect to the user's initial input speech.

34. The automated task classification method according to claim 32, characterized in that the feedback response includes confirmation with respect to at least one objective of the set of task objectives determined in the classification decision.

35. The automated task classification method according to claim 31, characterized in that the relationship between the determined speech fragments and the predetermined set of task objectives includes a measure of the usefulness of one of the speech fragments to a specified goal of the predetermined task objectives.

36. The automated task classification method according to claim 35, characterized in that the utility measure is manifested as a measure of the outstanding feature.

37. The automated task classification method according to claim 36, characterized in that the measurement of the outstanding characteristic is represented as a conditional probability of the task target that is requested given an appearance of the speech fragment in the input speech, the probability conditional that is a higher value in a distribution of conditional probabilities over the set of predetermined task objectives.

38. The automated task classification method according to claim 36, characterized in that each of the plurality of speech fragments determined by the determination step has a measure of the outstanding feature exceeding a predetermined threshold.

39. The automated task classification method according to claim 31, characterized in that the relationship between the determined speech fragments and the predetermined set of task objectives includes a measure of the common state within a language of the speech fragments.

40. The automated task classification method according to claim 39, characterized in that the measurement of the common state is manifested as a measure of mutual information.

41. The automated task classification method according to claim 40, characterized in that each of the plurality of speech fragments determined by the determination step has a measure of mutual information that exceeds the predetermined threshold.

42. The automated task classification method according to claim 31, characterized in that the step of making a classification decision includes a confidence function.

43. The automated task classification method according to claim 31, characterized in that the user's input speech represents a request for a specified objective from the set of predetermined task objectives.

44. The automated task classification method according to claim 31, characterized in that the incoming speech is sensitive to a question in a way "How can I help you?".

45. The automated task classification method according to claim 31, characterized in that each of the speech pronunciations is directed to an objective of the set of predetermined task objectives and each pronunciation is marked or labeled with a task objective to which it is assigned. direct