US20060143015A1 - System and method for facilitating call routing using speech recognition - Google Patents

Abstract

A computer-implemented method is described for optimizing prompts for a speech-enabled application. The speech-enabled application is operable to receive communications from a number of users and communicate one or more prompts to each user to illicit a response from the user that indicates the purpose of the user's communication. The method includes determining a number of prompt alternatives (each including one or more prompts) to evaluate and determining an evaluation period for each prompt alternative. The method also includes automatically presenting each prompt alternative to users during the associated evaluation period and automatically recording the results of user responses to each prompt alternative. Furthermore, the method includes automatically analyzing the recorded results for each prompt alternative based on one or more performance criteria and automatically implementing one of the prompt alternatives based on the analysis of the recorded results.

Description

TECHNICAL FIELD OF THE INVENTION
• The present invention relates generally to speech-enabled applications and, more particularly, to a system and method for optimizing prompts for speech-enabled applications.
BACKGROUND OF THE INVENTION
• Developments in speech recognition technologies support more natural language interaction between services, systems and customers than previously supported. One of the most promising applications of speech recognition technology, Automatic Call Routing (ACR), seeks to determine why a customer has called a service center and to route the customer to an appropriate service agent for customer request servicing. Speech recognition technology generally allows an ACR application to recognize natural language statements from the customer, thus minimizing reliance on conventional menu systems. This permits a customer to state the purpose of their call “in their own words”.
• In order for an ACR application to properly route calls, the ACR generally must interpret the intent of the customer, identify the type or category of customer call, and identify the correct routing destination for the call type. An ACR application may attempt to match one or words in a statement by a customer to a particular pre-defined action to be taken by the ACR application.
• Although speech recognition technology has been improving over the years, speech recognition systems are limited by the quality and robustness of the statistical language models or other techniques used to recognize speech. Given these limits, developers of these systems strive to develop prompts, announcements, and other instructions to the users of such systems that guide these users to provide speech input that conforms with the capabilities of the particular speech recognition technology used by the system. Subtle differences in the way prompts or other instructions are worded may result in substantial differences in system performance.
SUMMARY OF THE INVENTION
• In accordance with a particular embodiment of the present invention, a computer-implemented method is provided for optimizing prompts for a speech-enabled application. The speech-enabled application is operable to receive communications from a number of users and communicate one or more prompts to each user to illicit a response from the user that indicates the purpose of the user's communication. The method includes determining a number of prompt alternatives (each including one or more prompts) to evaluate and determining an evaluation period for each prompt alternative. The method also includes automatically presenting each prompt alternative to users during the associated evaluation period and automatically recording the results of user responses to each prompt alternative. Furthermore, the method includes automatically analyzing the recorded results for each prompt alternative based on one or more performance criteria and automatically implementing one of the prompt alternatives based on the analysis of the recorded results.
• Technical advantages of particular embodiments of the present invention include a method and system for optimizing prompts for speech-enabled applications that improve the operation of such applications. For example, particular embodiments automate the evaluation of various prompts or other user instructions for speech-enabled applications and then automatically implement the most effective prompt(s). Such embodiments can automatically present numerous prompt variations to users, evaluate the impact of each prompt on some measure of system performance, and adopt the prompt(s) that lead to the best system performance. This automation of prompt evaluation and implementation can reduce development time and ensure high system performance.
• Other technical advantages will be readily apparent to one skilled in the art from the following figures, descriptions and claims. Moreover, while specific advantages have been enumerated above, various embodiments may include all, some or none of the enumerated advantages.
BRIEF DESCRIPTION OF THE DRAWINGS
• A more complete understanding of the present embodiments and advantages thereof may be acquired by referring to the following description taken in conjunction with the accompanying drawings, in which like reference numbers indicate like features, and wherein:
• FIG. 1 is a block diagram depicting an example embodiment of a service center system according to one embodiment of the present invention;
• FIG. 2 is a flow diagram depicting an example automatic call routing method according to one embodiment of the present invention;
• FIG. 3 is a diagram depicting an example embodiment of an automatic call router action-object matrix according to one embodiment of the present invention; and
• FIG. 4 is a flow diagram depicting an example method for optimizing prompts for speech-enabled applications according to one embodiment of the present invention.
DETAILED DESCRIPTION OF THE DRAWINGS
• FIG. 1 is a block diagram depicting an example embodiment of aservice center system100 according to one embodiment of the present invention.System100 enables users to conduct transactions via aservice center102. For example, as referred to herein,service center102 may be a customer service call center for a telephone services company. However, as described below, the present invention may be used in conjunction with any other types of call centers, as well as with any systems that use speech recognition to perform an action or otherwise facilitate an action in response to speech input of a user. As used herein, the term “transaction” or it variants refers to any action that a user desires to perform in conjunction with or have performed byservice center102.
• Theexample service center102 includes one ormore computing apparatuses104 that are operably coupled to one or more transactionprocessing service solutions106. Included incomputing apparatus104 is aprocessor108. Operably coupled toprocessor108 ofcomputing apparatus104 is amemory110.Computing apparatus104 employsprocessor108 andmemory110 to execute and store, respectively, one or more instructions of a program of instructions (i.e., software).
• Also included incomputing apparatus104 iscommunication interface112.Communication interface112 is preferably operable to couplecomputing apparatus104 and/orservice center102 to an internal and/orexternal communication network114.Communication network114 may be the public-switched telephone network (PSTN), a cable network, an internet protocol (IP) network, a wireless network, a hybrid cable/PSTN network, a hybrid IP/PSTN network, a hybrid wireless/PSTN network, the Internet, and/or any other suitable communication network or combination of communication networks.
• Communication interface112 preferably cooperates withcommunication network114 and one or moreuser communication devices116 to permit a user associated with eachuser communication device116 to conduct transactions viaservice center102.User communication device116 may be a wireless or wireline telephone, dial-up modem, cable modem, DSL modem, phone sets, fax equipment, answering machines, set-top boxes, televisions, POS (point-of-sale) equipment, PBX (private branch exchange) systems, personal computers, laptop computers, personal digital assistants (PDAs), other nascent technologies, or any other appropriate type or combination of communication equipment available to a user.Communication device116 may be equipped for connectivity tocommunication network114 via a PSTN, DSL, cable network, wireless network, or any other appropriate communications channel.
• In operation,service center102 permits a user to request, using speech, processing or performance of one or more transactions byservice solutions106. To enable such processing,computing apparatus104 may include or have access to one ormore storage devices118 including one or more programs of instructions operable to interpret user intent from the user's speech, identify a solution sought by the user, and route the user to anappropriate service solution106.
• To aid in the interpretation, identification and routing operations ofservice center102,storage118 includes an action-object matrix120, a look-up table122,utterance storage124, aprompt library126, one or more speech recognition modules (such as a statistical language modeling engine128), and one ormore dialog modules129. Furthermore, to analyze and optimize the performance of the prompts used byservice center102,storage118 also includes a prompttest control module144 and a prompttest analysis module146. Additional details regarding the operation and cooperation of the various components included instorage118 will be discussed in greater detail below.
• In the illustrated embodiment,computing apparatus104 is communicatively coupled to one or more connection switches or redirectdevices130. Connection switch or redirectdevice130 enablescomputing apparatus104, upon determining an appropriate destination for the processing of a user-selected transaction, to route the user viacommunication network132 and, optionally, one ormore switches134, to an appropriate service agent or module ofservice solutions106.
• Service solutions106 preferably include a plurality of service agents or modules operable to perform one or more operations in association with the processing of a selected user transaction. For example, ifservice center102 is a telephone services call center,service solutions106 may include one or more service agents or modules operable to performbilling service solutions136,repair service solutions138,options service solutions140, how-to-useservice solutions142, as well as any other appropriate service solutions. The service agents or modules implemented in or associated withservice solutions106 may include, but are not limited to, automated or self-service data processing apparatuses, live technician support (human support), or combinations thereof.
• FIG. 2 illustrates anexample method150 for a speech-enabled call routing using an action-object matrix according to one embodiment of the present invention. However, it should be emphasized that embodiments of the present invention may be used in association with any speech-enabled applications using prompts and is certainly not limited to service centers routing users using action-objects.Method150 ofFIG. 2 may be implemented in one ormore computing apparatuses104 of one ormore service centers102. As such the method will be described with reference to the operation ofservice center102 ofFIG. 1.
• Upon initialization ofservice center102 atstep152,method150 proceeds tostep154 whereservice center102 provides for and awaits an incoming communication from auser communication device116 viacommunication network114. However, a user may connect withservice center102 in any other suitable manner.
• Upon detection of an incoming contact atstep154,method150 preferably proceeds tostep156 where a communication connection with theuser communication device116 is established. As suggested above, establishing a communication connection with an incoming contact from a user atstep156 may include, but is not limited to, receiving a user phone call via a PSTN or other wireline network, a wireless network, or any of numerous other communication networks.
• Once a communication connection has been established atstep156,method150 proceeds tostep158 where one or more prompts, announcements, or other instructions to the user (collectively referred to herein as “prompts”) are communicated to the user ofuser communication device116. In particular embodiments, the communication of one or more prompts is aimed at eliciting a request from the user for the processing of one or more transactions or operations. For example, atstep158,dialog module129 may accessprompt library126 ofstorage118 to generate a user transaction selection prompt such as, “Thank you for calling our service center. Please tell me how we may help you today.” Furthermore, any other suitable prompts designed to elicit a response from the user regarding a transaction that the user desires to be performed may be used.
• Preferably, a prompt will serve to illicit an unambiguous statement of the user's intent. If the user utterance is ambiguous orincomplete service center102 will need to engage in additional dialog to clarify the user's intentions. For example, in response to initial prompt fromservice center102, the most desirable outcome is for the user's response to the prompt to result in a “direct route” to the appropriate destination. However, if the user's response requires additional clarification (further user responses) beforeservice center102 can determine the appropriate destination,service center102 will need to employdialog module129 to provide additional prompts to the user in an attempt to illicit an unambiguous statement from the user. This additional-prompting increases costs (for example, by occupying incoming communication channels) and reduces customer satisfaction.
• Atstep160 ofmethod150,service center102 awaits a user response to the communicated prompt. Upon detection of a user response atstep160,method150 preferably proceeds to162 where a natural language response (a user “utterance”) from the user responsive to the communicated prompt is preferably received. Receipt of an utterance from a user may include storage of the user's utterance inutterance storage124 ofcomputing apparatus storage118. Permanent or temporary storage of a user utterance may enable and/or simplify the performance of speech recognition analysis thereon.
• Following receipt of a user utterance atstep162,method150 proceeds to step164 where the user utterance is evaluated to interpret or identify an intent of the user. and a requested transaction to be performed. In particular embodiments, evaluation of a user utterance atstep164 may include the use of one or more speech recognition technologies, such as that available from statisticallanguage modeling engine128 ofcomputing apparatus104. As suggested above, statisticallanguage modeling engine128 may cooperate withutterance storage124 in the evaluation of the user utterance.
• In certain embodiments, statisticallanguage modeling engine128 may evaluate the user utterance received atstep162 in cooperation with action-object matrix120, which defines a number of different action-objects (and which is described in greater detail below in conjunction withFIG. 3). In the evaluation of a user utterance atstep164, the speech recognition technology preferably employed by computingapparatus104 seeks to identify an action, an object or an action-object combination from the user utterance. By creating a finite number of transaction options (i.e., action-objects) via action-object matrix120, proper routing of a user to a service agent ormodule136,138,140 or142, may be accomplished with improved efficiency (for example, substantially eliminating user routing errors and, therefore, user re-routing).
• Each action-object in action-object matrix120 defines a particular action to be taken and an object that is the subject of the action (in other words, a transaction to be performed). For example, the action-object “pay/bill” defines an action “pay” to be carried out on an object “bill.” As described below, the assignment of an action-object to a user utterance enables efficient routing of the user to enable performance of a desired transaction.
• To assist in assigning a particular action-object to a user utterance in particular embodiments, statisticallanguage modeling engine128 may store and associate one or more salient action terms and one or more salient object terms with each action-object. The statisticallanguage modeling engine128 can then search for these salient terms in a user utterance to assign the user utterance to a particular action-object. The salient terms may be the actual action and object of the action-object and/or the salient terms may be different from the action and object. For example, the action-object “pay/bill” may be associated with the salient action term “pay” and the salient object term “bill.” In addition, the “pay/bill” action-object may be associated with the salient object terms “account” and “invoice.” Therefore, any user utterance including the term “pay” and at least one of the terms “bill,” “account” or “invoice” would preferably be associated with the “pay/bill” action-object. Multiple salient action terms could also or alternatively be associated with this action-object. At least a portion of the user utterance evaluation performed atstep164 may include determining whether the user utterance includes a salient action term, a salient object term, or both a salient action term and a salient object term.
• If it is determined that the user utterance contains only a salient action term(s)168,method150 proceeds to step176 where one or more additional prompts may be communicated to the user usingdialog module129, using adifferent dialog module129 than the module that communicated the initial prompt, or using any other suitable component. The prompts presented atstep176 are preferably designed to elicit the selection of an object (via a salient object term) in a subsequent user utterance. For example, referring to the action-object matrix depicted inFIG. 3, it may have been determined from the initial user utterance that the user desires to “inquire” about something. Having identified that the user wishes to make an “inquiry” (the action),computing apparatus104 may cooperate withdialog module129,prompt library126 and action-object matrix120 to prompt the user for selection of an object associated with the “inquire” action. As illustrated inFIG. 3, examples of objects associated with the “inquire” action include, in one embodiment, optional services, basic service, billing, cancellation, repair, payment, specials, and name and number. It should be understood that the action-object matrix depicted generally inFIG. 3 is included primarily for purposes of illustration. As such, alternate embodiments of an action-object matrix (or embodiments not using an action-object matrix) may be implemented without departing from the spirit and scope of teachings of the present invention.
• Similarly, if it is determined that the user utterance contains only asalient object term170,method150 preferably proceeds to step178 where one or more prompts designed to elicit the selection of an action (via a salient action term) in a subsequent user utterance. For example, referring again to the action-object matrix generally depicted inFIG. 3, if it is determined from the initial user utterance that the user desires some sort of action associated with a “bill”,computing apparatus104 may cooperate withdialog module129, action-object matrix120 andprompt library126 to generate one or more prompts directed to eliciting user selection of an “action” associated with the bill “object”. As shown inFIG. 3, examples of actions associated with a “bill” object may include, in one embodiment, inquiry, information, fixing or repairing, and paying.
• Method150 may loop throughsteps176 or178 one or more times in an attempt to illicit an appropriate salient action term or an appropriate salient object term, respectively, for any desired number of loops. If evaluation of the user utterances does not lead to the utterance of asalient action term168 nor asalient object term170 after a predetermined number of loops, if neither asalient action term168 or asalient object term170 are identified (an “other” utterance174), or ifsalient action terms168 and/orsalient object terms170 associated with multiple action-objects are identified, thenmethod150 proceeds to step180 where a disambiguation dialogue may be initiated and performed bydialog module129. In such an event,method150 preferably provides for additional appropriate dialogue to be performed with the user in an effort to elicit a usable “action-object” combination from the user (for example, asking the user to be more specific in his or her request).
• Following prompting for an “object” atstep176, prompting for an “action” atstep178, or initiation and performance of disambiguation dialogue at180,method150 preferably returns to step160 where a response may be awaited as described above.Method150 then preferably proceeds through the operations atsteps162 and164 until an “action-object”combination172 has been elicited from the user in a user utterance. An escape sequence may also be included inmethod150 where it has been determined that a user requires human assistance, for example.
• After identification of an “action-object” combination172 (either from the initial utterance or from the repeated prompting described above),method150 preferably proceeds to step182. Atstep182,computing apparatus104 preferably cooperates with action-object matrix120 and look-up table122 to identify a preferred or proper routing destination for processing the user-selected transaction. As suggested above, the routing destinations identified atstep182 may include routing destinations associated with the service agents or modules available inservice solutions106. As mentioned above, service agents ormodules136,138,140 and142 may include automated transaction processing available viacomputing apparatus104 or a similar device, live support, or combinations thereof, as well as other suitable transaction processing options.
• Following identification of a preferred or proper routing destination atstep182,method150 preferably proceeds to step184 where the user connection is preferably routed to the appropriate destination indicated in look-up table122. Following the routing of the user connection,method150 preferably proceeds to step186 where one or more aspects of the user utterance or utterances are optionally forwarded to the service agent or module destination to which the caller and/or user connection is routed. For example, in particular embodiments,method150 provides for the identified action-object to be forwarded to the service agent associated with the selected routing destination. In yet other embodiments, no information is forwarded and the user is simply routed to the appropriate destination. Following the routing of the user connection (and any forwarding of information),method150 preferably returns to step154 where another user connection is awaited.
• It should be understood that some of the steps illustrated inFIG. 2 may be combined, modified or deleted where appropriate, and additional steps may also be added to the method. Additionally, as indicated above, the steps may be performed in any suitable order without departing from the scope of the present invention. Furthermore, it should be understood that although embodiments of the present invention are described in conjunction with a service center using action-objects and salient terms, the present invention may be used in conjunction with any speech-enabled applications that use prompts to illicit a response to determine the user's intent.
• Referring again toFIG. 3, an action-object matrix120 according to one embodiment of the present invention is shown. The example action-object matrix120 shown inFIG. 3 includes a number of columns ofactions202 and a number of rows ofobjects204. The intersection of an action column with an object row generally defines an action-object pair identifying a transaction available via service center102 (for example, using one or more service modules oragents136,138,140 and142).
• As described above, action-object matrix120 is used in association with other components ofservice center102 to interpret user intent and identify a desired transaction from a user utterance. For example, usingactions202 andobjects204 of action-object matrix120, in conjunction with themethod150 described above, a user utterance such as “How much do I owe on my bill?” may be evaluated to relate to the action-object “inquire/bill”206. In a further example, the user utterance, “I have a problem with a charge on my bill” may be associated with the action-object “fix-repair/bill”208. In still another example, the user utterance, “Where can I go to pay my phone bill?” may be associated with the action-object “where/payment”210. In yet another example, the user utterance, “How do I set up Call Forwarding?” may be associated with the action-object “how-to-use/option” services212. In a further example, the user utterance, “I'd like to get CallNotes” may be associated with the action-object “acquire/optional services”214.
• As mentioned above,service center102 uses one or more salient action terms and one or more salient object terms associated with each action-object to associate a user utterance with the action-object. The salient terms may be stored in association with action-object matrix120 or elsewhere in service center102 (or at a location remote to service center102). If stored in association with action-object matrix120, the salient terms may be linked to particular action-objects, to particular actions (for salient action terms), or to particular objects (for salient object terms).
• After an action-object has been identified through the user of action-object matrix120 and other components ofservice center102, look-up table122 is used to identify the routing destination associated with an identified action-object. For example, upon identifying action-object “inquire/bill”206 from a user utterance,computing apparatus104 may utilize action-object matrix120 and look-up table122 to determine that the appropriate routing destination for the “inquire/bill” action-object206 is “Bill” service agent ormodule136. In another example, upon identifying action-object “fix-repair/bill”208 from a user utterance,computing apparatus104 cooperating with action-object matrix storage120 and look-up table122 may determine that an appropriate routing destination for the user connection includes “Repair” service agent ormodule138. Additional implementations of associating a look-up table with an action-object matrix may be utilized without departing from the spirit and scope of teachings of the present invention.
• FIG. 4 is a flow diagram depicting anexample method300 for optimizing prompts for speech-enabled applications according to one embodiment of the present invention. As described above, although theexample method300 is described with respect toservice center102, the method may be applied to any speech-enabled application that use prompts (again, this term refers to prompts, announcements, or any other instructions provided to a user) to illicit a response to determine a user's intent (regardless of how the user's intent is determined).
• Upon initialization atstep302,method300 proceeds to step304 where aparticular dialog module129 to evaluate is selected. A speech-enabled application, such asservice center102, may includemultiple dialog modules129 and each dialog module may be tested separately. For example, an application may have onedialog module129 that provides the initial prompt (and any associated announcements) and may include other dialog modules that provide additional prompts to obtain more detailed or unambiguous responses from a user. Usingservice center102 as an example, control of the prompt testing process may be performed bytest control module144 and thusmodule144 may select a dialog module to evaluate. Ifservice center102 includesmultiple dialog modules129,test control module144 may serially select eachdialog module129 for testing at particular intervals. This selection may be performed automatically based on a pre-determined configuration or may be based on input from a person configuring the test procedure.
• Method300 continues atstep306 where the prompt alternatives to evaluate for the selected dialog module are determined. For example,prompt library126 may include multiple alternative prompts for the initial prompt provided byservice center102.Test control module144 may accessprompt library126 to retrieve these alternative prompts and may determine which prompts are to be evaluated. For example,prompt library126 may initially include several alternative initial prompts andtest control module144 may initially select all the alternative prompts for testing. As an example, after testing all of the alternative initial prompts,test control module144 may eliminate certain prompts that performed poorly relative to the other prompts and may repeat the testing process on the remaining prompts if necessary. Alternatively, the particular prompts to be evaluated may be determined based on input from a person configuring the test procedure.
• In some cases, one or more of the prompt alternatives may include a combination of prompts (including announcements, etc.) to be evaluated. For example,method300 could be used to test an announcement followed by a prompt or a series of prompts. Therefore, atstep308 it is determined whether prompt combinations are being evaluated. If so,method300 continues to step310 where the order and particular combinations of the prompts to be tested are determined. For example, the same initial prompt could be tested with three different preceding announcements, or an announcement could be tested with three different prompts following the announcement. As another example, an initial prompt could be followed by pauses of different lengths (or no pause) before a series of example responses are provided to the user. As yet another example, the order of two or more prompts may be tested, with each different order being a different alternative. Any particular combination and/or order of prompts may be tested. Furthermore, other variations of the way in which multiple prompts are played may be evaluated. The information regarding the different combinations to be tested may be provided to and stored bytest control module144 for use in executing the testing.
• Once the order the prompts in each of the prompt alternatives has been determined (or if it is determined atstep308 that the prompt alternatives do not include any prompt combinations),method300 proceeds to step312 where the number of prompt cycles, length of time, and/or other characteristics of the evaluation period during which each prompt alternative is to be evaluated is determined. For example, a first prompt alternative “How may I help you?” may be used as the initial prompt for a one week period and then a second prompt alternative “What is the purpose of your call?” may be used for the following one week period. Alternatively, as an example, the first prompt alternative may be tested for the first one thousand prompt cycles (for example, calls from users) and the second prompt alternative may be tested for the next thousand prompt cycles. Furthermore, prompt alternatives may be testing in an alternating fashion (for example, a first prompt may tested for a single user, a second prompt may be tested on the next user, and the process may be repeated any desired number of times). Moreover, one or more of the alternative prompts may be tested for differing lengths of times, number of prompt cycles, or other evaluation periods as desired. Information about the testing period(s) to be used for each prompt alternative may be provided to and stored bytest control module144 for use in executing the testing.
• Method300 continues atstep314, where a first prompt alternative is selected from the multiple prompt alternatives determined to be part of the evaluation atstep306. This step may be performed automatically bytest control module144. For example, if the two prompt alternatives to be evaluated are “How may I help you?” and “What is the purpose of your call?”, then testcontrol module144 may select one of these alternatives to begin testing.
• Atstep316, the selected prompt alternative is retrieved from prompt library126 (or any other suitable location). For example, the prompt alternative may be an audio file (such as a .wav file) that is retrieved from prompt library126). Atstep318, the retrieved prompt alternative is presented to a user. For example, a .wav file may be played for the user. The playing or other presentation of the prompt alternative (which, again, may include a combination of prompts, announcements, pauses, etc.) is repeated for each user (such as callers calling into service center102) during the evaluation period determined atstep312. The selected prompt may be retrieved and presented by the associateddialog module129,test control module144, or any combination thereof (and reference todialog module129 performing this task is meant to include any of these options).
• Atstep320,test analysis module146 or any other suitable component (such as test control module144) records the results of each user's actions taken in response to the prompt alternative. Any suitable results of the user's interaction withservice center102 or other application presenting a prompt can be recorded for later evaluation. For example, the user's actual response(s) to the prompt alternatives may be recorded. In addition or alternatively, whether the user's response included a salient action term and/or a salient object term (for example, based on an evaluation of the response by statistical language modeling engine128) may be recorded. As another example,test analysis module146 or another suitable component may record whether the user's response resulted in a direct route or whether it required additional dialog from adialog module129 to clarify the user's intent. The component, such astest analysis module146, recording this information for evaluation can cooperate with any other suitable components of the system being analyzed to obtain this information. For example,test analysis module146 may communicate with statisticallanguage modeling engine128, action-object matrix120, dialog module(s)129, or any other appropriate components ofservice center102 to record suitable information for evaluating the performance of a particular prompt alternative.
• Furthermore, it is again emphasized that the present invention may apply to any type of speech-enabled applications, and is not limited to the service center example provided herein. Therefore, suitable performance criteria may be unrelated to the routing of users/callers to a destination. For a given prompt, or set of prompts, a variety of performance measures may be evaluated, such as the percentage of timeouts (for example, when no response is received from a user), the percentage of “too much speech” (for example, the user says more than the speech engine can process), the percentage of utterances that are “in grammar” (for example, the utterance matches a defined grammar item in an engine that recognizes types of grammar), or the number of times a caller asks for “help.” These are but a few examples and there are a wide range of different measures for prompt performance other than action-objects and direct routes.
• At the end of the evaluation period for the first prompt alternative,method300 proceeds to step322 where it is determined whether there are additional prompt alternatives to evaluate. For example,test control module144 can determine whether all the prompt alternatives selected atstep306 have been evaluated. If there are additional prompt alternatives to evaluate,method300 returns to step314, where the next prompt alternatives to evaluate is selected.Steps316 through320 are then performed for that prompt alternatives, as described above.
• Once it is determined atstep322 that no further prompt alternatives remain to be evaluated (for the selected dialog module),method300 proceeds to step324 wheretest analysis module146 or any other suitable component analyzes the information recorded atstep320 for each of the prompt alternatives. For example,test analysis module146 may determine which prompt alternative resulted in the most direct routes (for example, the highest percentage of direct routes) or which alternative had the most initial responses that resulted in a match with an action-object (for example, responses that included both a salient action term and a salient object term). Any number of additional or alternative performance criteria that assist in identifying the most effective prompt alternative may be analyzed byanalysis module146, as desired. This analysis may be performed in “real-time” whileservice center102 or other system being evaluated is continuing to interact with users. Furthermore, some or all ofanalysis step324 may occur for each prompt alternative while that prompt alternative or other prompt alternatives are being tested.
• Based on the analysis performed atstep324,test analysis module146 compares the results of the analysis and determines the “best” prompt alternative atstep326. Which prompt alternative is the best depends on the criteria being evaluated. As examples only, the prompt alternative resulting in the most direct routes or resulting in the most action-object matches (either initially or after further prompting) may be selected as the best prompt alternative atstep326. Once the best prompt alternative is selected,test analysis module146 or any other suitable component automatically adjusts the dialog module atstep328 to implement the chosen prompt alternative. This automatic evaluation and adjustment of prompts is advantageous since manually testing and adjusting various prompts is time consuming and may interrupt or impede the performance of the system being tested.
• After thedialog module129 being tested is appropriately adjusted,method300 continues to step330 where it is determine whether there areadditional dialog modules129 to evaluate. If so,method300 returns to step304 where anew dialog module129 is selected for evaluation. If not,method300 ends.Method300 may be repeated as often as desired during the operation of a service center or other speech-enabled application to which the method might apply. For example,method300 may be performed when a service center or other speech-enabled application is initially put into operation. Thereafter, the method could be performed periodically as desired (for example, every six months). Such periodic testing may be helpful as users become experienced with the prompts used in a system. The users' behavior can change with such experience and periodic updating of the prompts can be beneficial to “tune” the prompts to take into account the users' changed behavior.
• It should be understood that some of the steps illustrated inFIG. 4 may be combined, modified or deleted where appropriate, and additional steps may also be added to the method. Additionally, as indicated above, the steps may be performed in any suitable order without departing from the scope of the present invention.
• Although the present invention has been described in detail with reference to particular embodiments, it should be understood that various other changes, substitutions, and alterations may be made hereto without departing from the spirit and scope of the present invention. For example, although the present invention has been described with reference to a number of components included withinservice center102, other and different components may be utilized to accommodate particular needs. The present invention contemplates great flexibility in the arrangement of these elements as well as their internal components. Moreover, speech-enabled applications or systems other than service centers may also be used in conjunction with embodiments of the present invention.
• Furthermore, numerous other changes, substitutions, variations, alterations and modifications may be ascertained by those skilled in the art and it is intended that the present invention encompass all such changes, substitutions, variations, alterations and modifications as falling within the spirit and scope of the appended claims. Moreover, the present invention is not intended to be limited in any way by any statement in the specification that is not otherwise reflected in the claims.

Claims (24)

1-28. (canceled)

29. A computer-implemented method for facilitating call routing, the method comprising:

receiving one or more user utterances from a user during a communication session;

automatically identifying an action and an object from the one or more user utterances using one or more speech recognition techniques;

automatically accessing an action-object data set defining, for each of a plurality of action-object combinations, a transaction corresponding to that action-object combination;

automatically identifying from the action-object data set the transaction corresponding to the action and object identified from the user utterance; and

routing the communication session to a routing destination based on the identified transaction.

30. A method according toclaim 29, wherein routing the user communication based on the identified transaction comprises:

automatically accessing a routing table corresponding a plurality of different routing destinations with a plurality of different transactions; and

automatically identifying from the routing table the routing destination corresponding with the identified transaction.

31. A method according toclaim 29, wherein the action-object data set comprises an action-object matrix.

32. A method according toclaim 29, wherein receiving one or more user utterances and automatically analyzing the one or more user utterances to identify an action and an object comprises:

automatically analyzing one or more first user utterances to attempt to identify an action and an object from the one or more first user utterances; and

if at least one of the action and the object is not identified from the analysis of the one or more first user utterances, automatically prompting the user for one or more second user utterances for determining the unidentified action or object.

33. A method according toclaim 29, wherein receiving one or more user utterances and automatically analyzing the one or more user utterances to identify an action and an object comprises:

automatically analyzing one or more first user utterances to identify at least one of an action and an object from the one or more first user utterances; and

if an action, but not an object, is identified from the analysis of the one or more first user utterances:

automatically prompting the user for an object;

receiving one or more second user utterances from the user; and

automatically analyzing the one or more second user utterances to identify an object from the one or more second user utterances.

34. A method according toclaim 29, wherein receiving one or more user utterances and automatically analyzing the one or more user utterances to identify an action and an object comprises:

automatically analyzing one or more first user utterances to identify at least one of an action and an object from the one or more first user utterances; and

if an object, but not an action, is identified from the analysis of the one or more first user utterances:

automatically prompting the user for an action;

receiving one or more second user utterances from the user; and

automatically analyzing the one or more second user utterances to identify an action from the one or more second user utterances.

35. A method according toclaim 29, wherein the routing destination comprises an agent associated with the identified transaction.

36. A method according toclaim 29, wherein the routing destination comprises an automated transaction module for facilitating the identified transaction for the user.

37. A method according toclaim 29, further comprising forwarding at least one of the identified action and object to the routing destination.

38. A computer-implemented system for facilitating call routing, the system comprising:

data storage including an action-object data set defining, for each of a plurality of different action-object combinations, a transaction corresponding to that action-object combination;

a dialog module that receives one or more user utterances from a user during a communication session;

a language modeling engine coupled to the dialog module and configured to automatically identify an action and an object from the one or more user utterances using one or more speech recognition techniques, and identify from the action-object data set the transaction corresponding to the action and object identified from the user utterance; and

a routing module coupled to the language modeling engine and configured to route the user communication to a routing destination based on the identified transaction.

39. A system according toclaim 38, further comprising a routing table corresponding a plurality of different routing destinations with a plurality of different transactions; and

wherein the routing module is configured to identify from the routing table the routing destination corresponding with the identified transaction.

40. A system according toclaim 38, wherein the language modeling engine is configured to automatically analyze one or more first user utterances to attempt to identify an action and an object from the one or more first user utterances, and if at least one of the action and the object is not identified from the analysis of the one or more first user utterances, automatically prompt the user for one or more second user utterances for determining the unidentified action or object.

41. A system according toclaim 38, wherein the language modeling engine is configured to:

automatically analyze one or more first user utterances to identify at least one of an action and an object from the one or more first user utterances; and

if an action, but not an object, is identified from the analysis of the one or more first user utterances:

automatically prompt the user for an object; and

automatically analyzing one or more second user utterances received from the user to identify an object from the one or more second user utterances.

42. A system according toclaim 38, wherein the language modeling engine is configured to:

automatically analyze one or more first user utterances to identify at least one of an action and an object from the one or more first user utterances; and

if an object, but not an action, is identified from the analysis of the one or more first user utterances:

automatically prompt the user for an action; and

automatically analyzing one or more second user utterances received from the user to identify an action from the one or more second user utterances.

43. A system according toclaim 38, wherein the routing destination comprises an agent associated with the identified transaction.

44. A system according toclaim 38, wherein the routing destination comprises an automated transaction module for facilitating the identified transaction for the user.

45. A system according toclaim 38, wherein the routing module is configured to forward at least one of the identified action and object to the routing destination.

46. A computer-readable medium including computer-executable instructions for facilitating call routing, comprising:

instructions for receiving one or more user utterances from a user during a communication session;

instructions for identifying an action and an object from the one or more user utterances using one or more speech recognition techniques;

instructions for accessing an action-object data set defining, for each of a plurality of action-object combinations, a transaction corresponding to that action-object combination;

instructions for identifying from the action-object data set the transaction corresponding to the action and object identified from the user utterance; and

instructions for routing the communication session to a routing destination based on the identified transaction.

47. A computer-readable medium according toclaim 46, wherein the instructions for routing the user communication based on the identified transaction comprise:

instructions for accessing a routing table corresponding a plurality of different routing destinations with a plurality of different transactions; and

instructions for identifying from the routing table the routing destination corresponding with the identified transaction.

48. A computer-readable medium according toclaim 46, comprising:

instructions for analyzing one or more first user utterances to attempt to identify an action and an object from the one or more first user utterances; and

instructions for, if at least one of the action and the object is not identified from the analysis of the one or more first user utterances, prompting the user for one or more second user utterances for determining the unidentified action or object.

49. A computer-readable medium according toclaim 46, comprising:

instructions for analyzing one or more first user utterances to identify at least one of an action and an object from the one or more first user utterances; and

instructions for, if an action, but not an object, is identified from the analysis of the one or more first user utterances:

prompting the user for an object;

receiving one or more second user utterances from the user; and

analyzing the one or more second user utterances to identify an object from the one or more second user utterances.

50. A computer-readable medium according toclaim 46, comprising:

instructions for analyzing one or more first user utterances to identify at least one of an action and an object from the one or more first user utterances; and

instructions for, if an object, but not an action, is identified from the analysis of the one or more first user utterances:

prompting the user for an action;

receiving one or more second user utterances from the user; and

analyzing the one or more second user utterances to identify an action from the one or more second user utterances.

51. A computer-readable medium according toclaim 46, further comprising instructions for forwarding at least one of the identified action and object to the routing destination.

Images