24

Improved call completion with Natural Language Processing Peter Trompetter

  • Upload
    shima

  • View
    18

  • Download
    0

Embed Size (px)

DESCRIPTION

“Just simply say it”. Improved call completion with Natural Language Processing Peter Trompetter Vice President Global Development [email protected] www.GyrusLogic.com (+1) 602 - 432 1995. DTMF / IVR WWW Speech IVR Speech NLU. TUI / Spoken Data Out GUI / Graphical Data Out - PowerPoint PPT Presentation

Citation preview

Page 1: Improved call completion with  Natural Language Processing Peter Trompetter
Page 2: Improved call completion with  Natural Language Processing Peter Trompetter

© GyrusLogic, Inc.

Improved call completion with Natural Language Processing

Peter TrompetterVice President Global [email protected](+1) 602 - 432 1995

“Just simply say it”

Page 3: Improved call completion with  Natural Language Processing Peter Trompetter

© GyrusLogic, Inc.

Evolution of Self-Service Solutions

Late 1970’s

Early 1990’s

Late 1990’s

Early 2000’s

DTMF / IVR

WWW

Speech IVR

Speech NLU

TUI / Spoken Data Out

GUI / Graphical Data Out

VUI / Spoken Data

VUI / Natural Language

Page 4: Improved call completion with  Natural Language Processing Peter Trompetter

© GyrusLogic, Inc.

The Economics of Service Levels

• On average, every 1 percent increase in first contact resolution results in a 0.64 % increase in customer satisfaction.

• On average, with every 1 second decrease in the average speed of answer, customer satisfaction improves only 0.03 %

• Focusing on things that drive first contact resolution generally creates greater customer satisfaction and is often less costly than focusing on things that improve the average speed of answers

Page 5: Improved call completion with  Natural Language Processing Peter Trompetter

© GyrusLogic, Inc.

Typical Natural Language DialogSystem: How may I help you?

User: I would like to make a reservation for a flight to Chicago on July 17th

System: At what time and from what city do you want to travel?User: In the morning from Phoenix, Arizona System: OK, you could travel at 10 am. On Wednesday July 17th from Phoenix to Chicago. Do you want to make this reservation?User: What are my options?System: On July 17th you could travel at 10am., noon and 2pm. Do you want to make the reservation for the 17th at 10am?User: No, I would like to go on the 7th. In the afternoon.System: OK, you like to travel at 2pm on Sunday July 7th from Phoenix to

Chicago. Do you want to make this reservation?User: Yes, can I make a seat selection? System: Based on your customer profile we assigned a window seat 9A. Would that work for you?User: Yes, thank you.System: Let’s confirm your reservation, you are traveling on Sunday July 7th. at 2pm from Phoenix to Chicago and we assigned you seat 9A. Do you want to purchase this ticket and confirm the reservation?User: Yes, perfect!System: OK, reservation completed. How else may I help you?

Note how the user can be spontaneous and say as much as he

wants

Note how the user can be spontaneous and say as much as he

wants

Interruption of dialog with any question, even

if it was unrelated

Interruption of dialog with any question, even

if it was unrelated

User can make a correction on the fly at

any point in time

User can make a correction on the fly at

any point in time

The system still handles the question effectively

without coding of additional rules

The system still handles the question effectively

without coding of additional rules

The system resumes the previous dialog and

requests the missing information

The system resumes the previous dialog and

requests the missing information

Page 6: Improved call completion with  Natural Language Processing Peter Trompetter

© GyrusLogic, Inc.

Voice Self Service Savings Opportunity

$0.00

$10.00

$20.00

$30.00

$40.00

$50.00

$60.00

High

Low

Average

High $5.00 $42.00 $28.00 $57.00 $27.00 $12.00 $28.00 $40.00

Low $0.25 $2.00 $1.50 $3.00 $4.50 $0.85 $2.50 $4.00

Average $0.50 $4.50 $2.50 $7.50 $5.25 $1.85 $4.50 $3.25

Web Self Service

Basic e-mail

Auto e-mail

Web Chat

Auto Chat

IVR Phone CIH Intervoice self service

Page 7: Improved call completion with  Natural Language Processing Peter Trompetter

© GyrusLogic, Inc.

Typical Call Center Voice Metrics

• Speech recognition applicationo Overall automation rateo Speech play rate / don’t play rateo Zero-out and call abandon rateo Transaction / query completion rateo Prompt / conversational usability rateo Call durationo Call completion rateo Customer satisfaction metrics

Page 8: Improved call completion with  Natural Language Processing Peter Trompetter

© GyrusLogic, Inc.

Selecting a Speech Solution

• Speech Technologyo IBM, Microsoft, Nuance,

Scansoft, Telisma

• Text to Speecho Several suppliers

• Platform providerso Industry standards

• Hardware• Languages, tools• VoiceXML / SALT

o IVR / CTI functionso Hosting vs. on premise

• Application developmento Industry standards

• languages, grammars, tools

o Existing available skillso Development environment

• Development standards• Development time• Development budget• Maintainability

o In house vs. outsourcingo Packaged applicationso Business and Customer

benefits

Page 9: Improved call completion with  Natural Language Processing Peter Trompetter

© GyrusLogic, Inc.

Speech Deployment Metrics Professional Services > 60%

Item 2002 ($M) % of total 2007 ($M) % of total CAGR %

System Integration 102.1 22.2 433.3 17.4 33.5

Application Development 128.4 27.9 671.2 26.9 39.2

Training/Installation/Support 52.7 11.5 389.5 15.6 49.2

Hardware 54.1 11.0 281.1 11.3 39.1

Base Software Licence 121.0 26.4 721.9 28.9 42.9

Total 458.3 100.0 2497.1 100.0 40.4

Source: Voice Information Associates, ASR in Telephony Applications, the World Market

Page 10: Improved call completion with  Natural Language Processing Peter Trompetter

© GyrusLogic, Inc.

Selecting your “Voice” Application

• DTMFo Dual Tone Multi-

Frequency

• Speech enabled IVRo “Press or say one”

• Directed dialogo Voice prompted menu

• Open dialogo “How can I help

you?”

» Significant limitationso The menu challenge

» Not really an enhancement

» Typical start with speech» still the menu, correction and

confirmation frustration

» Natural language» an answer at first contact

Page 11: Improved call completion with  Natural Language Processing Peter Trompetter

© GyrusLogic, Inc.

Natural Language Understanding

• Typical ASR approaches for Natural Language Understanding (NLU)o Statistical Language Modeling (SLM)o Statistical Semantic Modeling (SSM)

• “Say Anything” - Nuance• “Speak Freely” - ScanSoft

o Transcription of 20,000 - 30,000+ utteranceso Complex long scripting procedureso Keyword or Word spotting, interpretation

difficulties of output phraseso Procedural developments

Page 12: Improved call completion with  Natural Language Processing Peter Trompetter

© GyrusLogic, Inc.

Another NLU Approach• Artificial Intelligence technology build with computational

linguistic models for a fully automated conversational dialog.

No dialog design is necessary, since the system knows how to adapt dynamically to the requirements of the natural conversation.

Declarative development process.

Selectable automated text and / or audio responses.

At any time requests can be spontaneous, no static menus.

User requests can be specific or vague, the system will answer the question any way.

Automatic explicit or implicit corrections.

Rating for "n-best" ASR results based upon meaning and context.

Page 13: Improved call completion with  Natural Language Processing Peter Trompetter

© GyrusLogic, Inc.

Another NLU Approach - Tools

Easy to use Java API Standard integration templates for a quick interface to an

IVR or VoiceXML browser. Language development tools and templates for adding new

languages or unique words to existing language dictionaries. XML interface for building a quick interactive application

knowledge base. Standard universal English grammar. Standard templates for easy interaction with enterprise

databases. Automatic static and dynamic speech grammar generation. Batch test tool for quick dialog testing.

Page 14: Improved call completion with  Natural Language Processing Peter Trompetter

© GyrusLogic, Inc.

Typical Two Types of NLU Requests

• Fuzzy (informational)o They can be spontaneous, arbitrary, ad-hoc, and/or vague o The majority of user queries are fuzzy questionso If not all the information is available, users prefer some

information rather than no information• Exact (transactional)

o Specific transactions don’t allow for fuzzy approach, but they can still be spontaneous

• Transfer 100 bucks from checking to savings• Travel at 3 this afternoon from Phoenix to Burbank• Did check number 123 clear on July 1st.?

Page 15: Improved call completion with  Natural Language Processing Peter Trompetter

© GyrusLogic, Inc.

Application Implementation• Defining the application, Natural Language Understanding or

Directed Dialog or any Combination of both.o Initial functional application requirementso Document possible caller requests / answers

• Separate transactional requests • Generate grammar for transactional requests (BNF)• Develop a XML based state machine for transaction requests • Universal Grammar for informational requests provided• Use template answers for informational “fuzzy” requests

o Document sample answers o Parse corpus of questions

• Identify the Enterprise backend dB interfaces (XML)

Page 16: Improved call completion with  Natural Language Processing Peter Trompetter

© GyrusLogic, Inc.

GyrusLogic Language Support• Today, users can build any grammar for their specific application in

any language in a matter of hours. • With GyrusLogic, languages are supported by composition of

dictionaries and grammars. • Language specific dictionaries and grammars are kept outside the

GyrusLogic engine and are standard text files, allowing users to easily provide a new or updated content to a language.

• For Fuzzy requests we do provide a universal grammar of English, a user doesn't have to compose anything.

• New language Universal Grammar development can be completed in several days, utilizing specific language experts / natives.

• GyrusLogic language limitations are based upon the languages

supported by the ASR used.

Page 17: Improved call completion with  Natural Language Processing Peter Trompetter

© GyrusLogic, Inc.

Speech Grammar Generation Process

Speech GrammarGenerator

(static grammars)

Corpusfor Fuzzyqueries

Corpusfor Fuzzyqueries

staticgrammar

&dictionaries

User’s Gyrus ABNF

grammar

User’s Gyrus ABNF

grammar

User XMLwith embedded

grammars

User XMLwith embedded

grammars

GyrusLogicServer

GyrusLogicClient/API

ASRdynamicgrammar

Page 18: Improved call completion with  Natural Language Processing Peter Trompetter

© GyrusLogic, Inc.

Typical Natural Language Dialog

System: How can we help you? User: I would like to transfer funds from my checking to my

savings.System: How much money do you want to transfer?User: How much money do I have in my savings account?System: The balance in your savings account is 4000 dollars.

I apologize for asking again, how much did you want to transfer?

User: Five hundred.System: Ok, let's review this transaction, you want to transfer 500

dollars from your checking to your savings, is this correct?User: No, actually transfer a thousand bucks from savings!System: Ok, let's review this transaction, you want to transfer 1000

dollars from your savings to your checking, is this correct?User: YesSystem: How else can we help you?User: What are the last five transactions in my checking?System: The last three transactions in your checking account are….

Note how the user can be spontaneous and say as much as he

wants

Note how the user can be spontaneous and say as much as he

wants

Interruption of dialog with any question, even

if it was unrelated

Interruption of dialog with any question, even

if it was unrelated

User can make a correction on the fly at

any point in time

User can make a correction on the fly at

any point in time

The system still handles the question effectively

without coding of business rules

The system still handles the question effectively

without coding of business rules

The system resumes the previous dialog and

requests the missing information

The system resumes the previous dialog and

requests the missing information

Page 19: Improved call completion with  Natural Language Processing Peter Trompetter

© GyrusLogic, Inc.

VoiceXML Mixed Initiative Example• A Natural Language solution is more spontaneous!• Exponential savings for more complex applicationsA benchmark with a web published VoiceXML example:

Agent: Thanks for calling Acme Travel Company. How can I help you today?

Caller: I'd like to book a flight.

Agent: Okay. What is your point of origin, and where are you going?

Caller: I wanna fly to Boston, Massachusetts.

Agent: You want to fly to Boston, Massachusetts. Where are you flying from?

Caller: From San Francisco, California.

Agent: Okay, you'll be traveling from San Francisco, California to Boston,

Massachusetts. Is that correct?

Caller: Yes.

10 times

less effort

+

Page 20: Improved call completion with  Natural Language Processing Peter Trompetter

© GyrusLogic, Inc.

• How does a GyrusLogic NLU solution help?o Declarative vs. Procedural or Icon based developments.o Open industry standards system and development

environment.o Implicit and Explicit corrections without additional coding.o Spontaneous user interruptions in any call flow without

additional development effort.o Context and semantics recognition at any point in the dialog.o Significant savings in time and money with the

deployment and maintenance of GyrusLogic based NLU applications.

o The same application can be used for Voice, chat, web and SMS. (text messages)

Page 21: Improved call completion with  Natural Language Processing Peter Trompetter

© GyrusLogic, Inc.

Add the Advantages of NLU• Give the right answer at the first request!

• No dialog design, flat “menu” with “How can I help you”, human like full conversational.

• Managing distinctions between fuzzy informational queries and exact transactional queries.

• Implicit and explicit transaction correction functionality.

• With a declarative development paradigm, developments will be significantly less error prone and less time consuming than with other procedural developments.

• Virtually no language dependency.

• ASR and TTS independence, interfaces with VoiceXML or your existing IVR.

Page 22: Improved call completion with  Natural Language Processing Peter Trompetter

© GyrusLogic, Inc.

Improved Call Completion with

Natural Language Processing• First contact customer resolution achieved• Increase in customer satisfaction accomplished• Reduced relative call time accounted for• More transactions / requests completed automatically• Combine Directed Dialog with NLU and create

additional application opportunities• NLU has transitioned from “emerging” to “applied”

solution • Improved call completion with a full

conversational natural language processing approach will result in significant cost reduction.

Page 23: Improved call completion with  Natural Language Processing Peter Trompetter

© GyrusLogic, Inc.

Average Speech Deployment Costs

Source: Voice Information Associates, ASR in Telephony Applications, the World Market

A minimum of 27% project costs savings

Average Total Project Costs Today

Application Development

27.5%

System Integration

19.9%

Base Software Licence27.8%

Hardware11.2%

TrainingInstallation

Support13.6%

Average Project Costs with GyrusLogic(including NLU capabilities)

Base Software Licence27.7%

Hardware11.2%

Training Installation

Support13.6%

Application Development

2.7%

System Integration

17.8%

Page 24: Improved call completion with  Natural Language Processing Peter Trompetter

© GyrusLogic, Inc.

“Just Simply Say It”Questions or you want to know more about

our Natural Language Understanding solutions?

please contact us at:

http://GyrusLogic.com - (+1) 602 - 432 1995 - [email protected]

JustSimplySayIt.com