Robustness of a Spoken Dialogue Interface for a Personal Assistant
|
|
|
- Samuel Mills
- 10 years ago
- Views:
Transcription
1 Robustness of a Spoken Dialogue Interface for a Personal Assistant Anna Wong, Anh Nguyen and Wayne Wobcke School of Computer Science and Engineering University of New South Wales Sydney NSW 22, Australia {annawong anht wobcke}@cse.unsw.edu.au Abstract Although speech recognition systems have become more reliable in recent years, they are still highly error-prone. Other components of a spoken language dialogue system must then be robust enough to handle these errors effectively, to avoid recognition errors from adversely affecting the overall performance of the system. In this paper, we presenttheresultsofastudyfocusingontherobustnessof our agent-based dialogue management approach. We found that while the speech recognition software produced serious errors, the Dialogue Manager was generally able to respond reasonably to users utterances. 1. Introduction In recent years, considerable research effort has been put into the area of dialogue modelling for computer-based applications. The goal is to allow high-level user-system interaction, such as through spoken language interfaces, which become feasible in domains that are sufficiently constrained to overcome the limitations of speech recognition over large vocabularies. To ensure flexible, adaptive and continuous interaction, robust dialogue management for such interfaces is of particular importance. Central to the robustness of dialogue management is the capability of the system to recover from errors, particularly speech recognition errors. Note that although speech recognition systems have greatly improved in recent years and have become commercially viable, performance of such systems is still far from perfect, and speech recognition errors are inevitable. Speech recognition errors become more of a problem as the complexity of speech interaction increases, due to the increase in the variety of language constructions able to be used, and the potential for compound errors(further errors as a result of speech recognition errors). Our goal in designing a dialogue manager was to blend the flexibility of speech interaction with robust dialogue management, so that the user can continue a meaningful interaction with the system even when speech recognition produces errors. In this paper, we describe how the Dialogue Manager handles speech recognition errors from third-party speech recognition software in the context of a personal assistant application providing access to and calendar applications. The objectives of the current study are to quantify the accuracy of the speech recognition engine and then examine the robustness of the Dialogue Manager in processing individual utterances. We first give an overview of the Dialogue Manager and the strategies we have used to ensure robustness of the system, then describe the experimentalmethodforthestudy,andfinallypresenttheresultsof our evaluation. 2. The Dialogue Manager We have used several strategies in our Dialogue Manager to encourage robustness against speech recognition errors. Firstly, to extract the essence of an utterance with the aim of recognising the user s intention or purpose, rather than processing every word in a user s utterance, we look for specific words and task-specific phrases in order to identifykeywordsandconcepts.indoingso,weminimisethe effects of speech recognition errors in irrelevant parts of an utterance by not requiring the speech recognition to be accurate over the entire utterance. Secondly, to encourage robustness and to allow flexibility in speech interaction, we haveusedavarietyofrulestolettheuserexpressthesame request in multiple ways, e.g. Have I received any fromjohn?orarethereanynew sfromjohn?however, this still relies on the speech recognition software s output, which is error-prone. Thirdly, we have used an open vocabulary over a dictation-based language model, again to provide flexibility in speech interaction. Although constrained by the domain, the user s language can be arbitrary English sentences, so can include proper names, pronouns and a variety of grammatical constructions. This has required us to use the speech recognition software in dicta-
2 tion mode, as our observations indicated that an increase in thesizeofagrammarwouldleadtoanunacceptabledegradation of speech recognition performance. In the case of failures, an important function of the Dialogue Manager is to recover and present a response to continue the interaction.sucharesponsemaybeasimplerequesttotheuser to repeat their last statement, or a clarification request asking for specific information relevant to the current task. Designed to be used with any third-party speech recognition software, the Dialogue Manager is implemented as one agent in a multi-agent system. The dialogue model consists of a collection of modular plans, each handling a particular dialogue or problem-solving aspect of the system, Nguyen and Wobcke[1]. The Dialogue Manager is currently deployed as part of the Smart Personal Assistant (SPA),Wobckeetal.[4],thatenablesuserstoconducta natural language dialogue in the domains of and calendar management using a variety of devices. The system is implemented as a multi-agent system using JACK Intelligents TM,asshowninFigure1.Inourexperimental setup, users interacted with the system using a PDA interface via the internal microphone, with the speech recognition software running on a server. DesktopPC Laptop PDA SPA Client User Interface Partial Parser User Interaction Speech Processor SPA Server Dialogue Manager E mail Calendar Figure 1. SPA System Architecture E mail Server Calendar Server The typical flow of control is as follows. User utterances are first processed by the Speech Recogniser, which then passes the corresponding text output to the Partial Parser. The Partial Parser then performs a shallow syntactic analysis of this text, extracting key words and phrases(e.g. domain-specific keywords such as and phrases such as from John), discarding the remainder. The Partial Parser is based on pattern matching, and uses the ProBot scriptingengineofsammut[2].thepartialparserthensendsa simple structural representation of the utterance to the Dialogue Manager. The Dialogue Manager performs semantic analysis to determine the domain of the utterance( or calendar), pragmatic analysis to identify objects referenced by the user(e.g. messages, folders, appointments, days, etc.), task delegation to request the specialist task assistants to process the user s utterance, and response generation to convert the responses to spoken and GUI output. 3. Experimental Method Therewereusers,maleandfemale,involvedinthe study. All were native Australian English speakers, ranging from18to4yearsold.allhadnoorlimitedexperienceusing spoken dialogue systems. Thetestconsistedof12taskswithvaryinglevelsofdifficulty,rangingfromsmalltaskssuchasfindingasingle message to more complex tasks, requiring the user to move between the and calendar domains, as listed intable1.mindfulofthelowtaskcompletionratesreported in the SmartKom evaluations[3] and the difficulty in analysing system performance over long intervals, we wantedtasksthatwerenottoodifficultfortheuser,inthat we anticipated most users would complete the majority of tasks, though perhaps with some difficulty. On the other hand,wedidnotwanttasksthatweretooeasyinthatthey could be achieved simply by reading out the task descriptionasaquerytothesystem.thuswechosearangeof tasks,fromthosethatcouldbesolvedusingjustoneortwo utterances to more complex tasks that would require several utterances to complete. The tasks were specified in a way that required the user to reformulate the task statements into language that could be understood by the system, and, for the more complex tasks, to plan a sequence of utterances. The experiment ran over two sessions. During the first session,avoicemodeloftheuserwascreatedbyhaving themreadtwo1-minutepassages.theusertooka3- minutebreakasthevoicedatawasprocessed.thefirst training session then commenced. The user was given severalpointersonhowtousethesystemandtheywerethen asked to work through a series of simple training examples. Ifrequired,userswerecoachedonhowtospeaktothesystem by a facilitator when completing the training tasks. Althoughtheexampleswerestructuredinsuchawaythatthey couldsimplybereadout,userswerenottoldthis.instead, users were strongly encouraged to try different ways of performing the same task, e.g. phrasing requests as a question or using different wordings, particularly wordings that reflected how they themselves would phrase a question or statement. The speech output of the system was shown to theuserinatextboxatthebottomofthescreen,sousers could adjust their speech in the light of speech recognition errors, e.g. to better articulate misrecognised words. The second session involved further training and then testing. Firstly, the user was asked to work through the same setofsimpletrainingexamples.theuserwasthengiventhe moredifficulttesttasksandaskedtoworkthroughthemat hisorherownpace.duringtrainingandtesting,userswere notallowedtousethepdastylusexceptforscrollingand pressing the speech button, before starting and after stoppingspeech.userswereagainabletoseethespeechrecognitionoutputinthetextboxonthepdascreen.
3 Task Instruction Description 1 Find the from John Lloyd. search 2 Find your appointments for next week. Appointment search 3 Find all s which were sent to you today. search 4 View the list of s in the Seminars folder. Folder search Schedule a meeting with Kathy about the budget, for tomorrow at 11am. Appointment schedule 6 CheckthatyouhaveanappointmentonFridayat11am.RescheduleittoMondaynextweekat2pm. Appointment search, reschedule 7 CheckthatyouhaveanappointmentforTuesdaynextweekat3 Appointment search, pm. Then delete it. deletion 8 You are looking for an from Paul Compton about the project, whichyouknowhesenttoyoulastweeksometime.pleasefindand read his . Complex search 9 Find out what time your appointment with Jessica is today. Appointment search You have received s about the war with Israel. Please find andthendeleteallofthem. search, deletion 11 Find all s about seminars and move them to the Seminars search, folder. archive 12 Findyour sfortoday.ReadthemessagefromKateandcompleteanyrequeststhatthesenderhasaskedofyou. search and appointment search, reschedule Table 1. Evaluation Tasks 4. System Evaluation The design of the Dialogue Manager, and in particular, the strategies used for achieving robust natural language interaction, is based on the assumption that there will sometimes be serious speech recognition errors that will adversely affect the performance of the overall system, so that amajorfunctionofthedialoguemanagerwillbetorecover from such errors. Accordingly, the first aspect of our evaluationistodeterminethenatureandextentoftheseerrorsin practice, in order to validate this assumption. Togainanaccurateideaofhowwellthespeechrecognition system performs, we measure the percentage of concept-words correctly recognised by the speech engine. We consider that each utterance consists of several concepts that are required to be fully and correctly recognised for the system to determine the user s intention. These concepts correspond to objects in the system, such asthetypeofrequestedtask,thenameofafolder,aperson s name, etc. We define the concept-words as the words in the utterance that denote these concepts. Figure 2 shows the concept-word recognition rate per useroverthesession.wewereinterestedbothintherecognition error rate over the concept-words for each user and the variation between users. Based on the uniformity of the Australian accent for native speakers, we expected there to be relatively low variation in concept-word error rate between users. The concept-word recognition accuracy per userrangedfrom82%to91%,withanoverallaverageof Percentage of Correctly Recognized Concept-Words (%) Recognition Accuracy Per User Average Accuracy Figure 2.Concept-WordRecognition 87%, which validates our basic assumption that the speech recognition system produces errors that necessitate recovery by the Dialogue Manager. However, more tellingly, the number of utterances with no errors in concept-word recognitionrangedfrom6%foruser()to82%foruser(7), withanaverageof67%,indicatingthatahighproportionof utterances input to the Dialogue Manager either contained errors or were missing some important information. The wide variation in the number of utterances without conceptword errors shows that the accuracy of the speech recognition software across the group of users, in our experimental setup, was highly sensitive to the user s accent and/or the technical limitations of the hardware.
4 To analyse the performance of the Dialogue Manager at the utterance level, we measure the occurrence of two types of system responses, not mutually exclusive: unexpected and inappropriate responses. Unexpected responses arethoseresponsesthat,fromthepointofviewoftheuser, are not expected given their actual utterance and their interpretation of the current conversational context. Inappropriate responses are those system responses that are incorrect, assuming that the concept-word recognition is % correct. Thus an inappropriate response points to either an errorinthepartialparserordialoguemanager,oralimitation in the language handled by the system. Misrecognition of a concept-word typically is difficult to recover from, and usually gives rise to an unexpected response for the user. Our analysis attributes each unexpectedresponsetoafailureofone(ormore)systemcomponents, speech recognition, Partial Parser or Dialogue Manager. The statistical distribution of these sources of unexpectedresponsesisshowninfigure3,whereinthecaseof multiple sources, the failure is attributed to the first component producing an error. Of 69 utterances in total(over all usersandtasks),therewereahighnumber(22)ofunexpected responses(36%). However, it can also be seen that, asexpected,mostofthese(122)wereduetospeechrecognitionerrors.thisshowsthatamajorfunctionofthedialoguemanagerisindeedtorecoverfromsucherrorstoproduce an acceptable continuation of the dialogue. Number of Unexpected Responses Caused by Parser Caused by User Caused by Dialogue Caused by Speech Figure 3. Source of Unexpected Responses When the speech recognition output is highly erroneous, the Dialogue Manager can typically only request to the user to repeat their utterance. However, in many cases, the Dialogue Manager is able to produce a specific clarification request, a request for information that was missing or incor- Number of Clarification Requests Number of Clarification Requests Per User Average Number of Clarification Requests Figure 4. Number of Clarification Requests rect in the speech recognition output, e.g. Which appointment do you want to delete? To measure the effectiveness of this type of error recovery across users, we examined the number of clarification requests given by the system, shown in Figure 4. Further analysis of the clarification requests in relation to the unexpected responses(figure 3) shows that the Dialogue Manager was often able to identifyaspecificprobleminthespeechoutput,andsoproduce a relevant clarification request for the required information. The Dialogue Manager is designed to achieve an acceptable level of flexibility and robustness. Given the high likelihood of speech recognition errors, the range of language constructsacceptedbythesystemisrestrictedsoastoreduce the effects of compound errors(i.e. dialogue management errors produced as a result of previous speech recognition errors), although we emphasise that there is still a wide varietyofconstructionsthatcanbeused.asaresult,the system is likely to produce an unexpected response when the user s constructions fall outside the scope of the system s language. Moreover, users were not informed in advance of these language restrictions, but were expected to discover them during experimentation. The idea behind the definition of inappropriate responses is to quantify the correctness of the Partial Parser and Dialogue Manager, factoring out the negative effects of speech recognition errors, in order to measure the robustness of these components. Of the 69 utterances in total, there were 87 inappropriate responses, attributable to the Partial Parser, the Dialogue Manager and(in one rare case) the back-end assistants. of the inappropriate responses(63%) were attributed to the Dialogue Manager. The distribution of inappropriate responses over all users is shown in Figure. Thereareawidevarietyofsourcesoferror,butthereare two main causes: inability to identify the object being referenced in the user utterance, and, closely related, inability
5 Number of Inappropriate Responses help in solving the problem). Theoverallsystemisrobustinthesensethatuserswere largely able to complete their assigned tasks. Our evaluationshowedaveryhighaveragescore(.6outof12)over thesetofusers.therewere14taskfailuresinthe12tasks (12tasksforusers):4duetothesubjectgivingupand other cases in which the task results were incorrect(sometimes without the knowledge of the subject). The failures werefortasks(),(6),(),(11)and(12).thosefortask () were mainly due to speech recognition errors for proper names(e.g. Kathy misrecognised as the); if an appointment was created but the meeting attendee name was incorrect, this counted as a task failure. Caused by Back-end Caused by Parser Caused by Dialogue Figure. Source of Inappropriate Responses to extract the correct object attributes. For example, a commonprobleminidentifyinganobjectwasreferringtoanappointment by its title, e.g. conference paper in answer to the clarification question Which appointment? However, the Dialogue Manager expects users to provide a structured referring expression such as the appointment about conference paper. Problems extracting object attributes are similar, e.g. the result of using constructions such as the appointment regarding conference paper, where regarding is not understoodbythesystem.itmightbethoughtthatthelanguage coverage could be improved by simply extending the Partial Parser with an additional rule, however while this will sufficeinsomecases(e.g.theuseofregarding),theproblem is that the additional rules may increase the potential for othertypesoferror,asinthecaseofconstructionssuchas the health care appointment. Hence we believe it is preferable overall that users are restricted to more easily interpretable constructions. Figure shows the number of inappropriate responses; the converse is that for 8% of utterances, the Partial Parser and Dialogue Manager were able to interpret their inputs correctly to perform their intended functions, even in the presence of speech recognition errors. In summary, the overall system achieves a high level of robustness across multiple users by using a variety of strategies. Finally, to measure the degree of task completion, we adopted a scoring scheme inspired by that used in the SmartKom evaluations, Schiel[3]. Users received 1 mark for a task successfully completed,. marks if they completed the task with help on wording from the experimenter (e.g. advice on pronunciation of words), and marks for afailuretocompletethetaskorifthesubjectcompleted the task but with excessive help from the experimenter(e.g.. Conclusion In this paper, we have described a dialogue management system to handle speech recognition errors from thirdparty speech recognition software. We then investigated the performance of our Dialogue Manager by measuring the number of unexpected and inappropriate responses. As expected, many of the unexpected responses were due to speech recognition errors; the inappropriate responses were duetoavarietyoffactors,butmainlytotheuseofvocabulary or language constructions outside the scope of the system, making it difficult for the system to determine which objects were referenced in the user s utterances. Although speech recognition was not perfect, the Dialogue Manager generally was able to respond to the user reasonably, even in the presence of speech recognition errors. Acknowledgements ThisworkwasfundedbytheCRCforSmartInternet Technology. We would like to thank Oriented SoftwarePtyLtdfortheuseoftheirJACKagentsplatform. References [1] Nguyen, A. and Wobcke, W. R. An -Based Approach to Dialogue Management in Personal Assistants. In Proceedings of the 2 International Conference on Intelligent User Interfaces, , 2. [2] Sammut, C. A. Managing Context in a Conversational. Electronic Transactions on Artificial Intelligence, (B):189 22, 21. [3] Schiel, F. Evaluation of Multimodal Dialogue Systems. In Wahlster, W., editor, SmartKom: Foundations of Multimodal Dialogue Systems. Springer-Verlag, Berlin, 26. [4] Wobcke,W.R.,Ho,V.H.,Nguyen,A.andKrzywicki,A.A BDI Architecture for Dialogue Modelling and Coordination in a Smart Personal Assistant. In Proceedings of the 2 IEEE/WIC/ACM International Conference on Intelligent Technology, , 2.
The Smart Personal Assistant: An Overview
The Smart Personal Assistant: An Overview Wayne Wobcke Anh Nguyen, Van Ho, Alfred Krzywicki, Anna Wong School of Computer Science and Engineering University of New South Wales Outline History: BT Intelligent
INF5820, Obligatory Assignment 3: Development of a Spoken Dialogue System
INF5820, Obligatory Assignment 3: Development of a Spoken Dialogue System Pierre Lison October 29, 2014 In this project, you will develop a full, end-to-end spoken dialogue system for an application domain
D2.4: Two trained semantic decoders for the Appointment Scheduling task
D2.4: Two trained semantic decoders for the Appointment Scheduling task James Henderson, François Mairesse, Lonneke van der Plas, Paola Merlo Distribution: Public CLASSiC Computational Learning in Adaptive
Specialty Answering Service. All rights reserved.
0 Contents 1 Introduction... 2 1.1 Types of Dialog Systems... 2 2 Dialog Systems in Contact Centers... 4 2.1 Automated Call Centers... 4 3 History... 3 4 Designing Interactive Dialogs with Structured Data...
Pasadena City College / ESL Program / Oral Skills Classes / Rubrics (1/10)
Pronunciation Classes Pasadena City College / ESL Program / Oral Skills Classes / Rubrics (1/10) ESL 246 SLO #1: Students will recognize and begin to produce correct American-English patterns in short
Download Check My Words from: http://mywords.ust.hk/cmw/
Grammar Checking Press the button on the Check My Words toolbar to see what common errors learners make with a word and to see all members of the word family. Press the Check button to check for common
Implementing an Electronic Document and Records Management System. Key Considerations
Implementing an Electronic Document and Records Management System Key Considerations Commonwealth of Australia 2011 This work is copyright. Apart from any use as permitted under the Copyright Act 1968,
Administrator s Guide
MAPILab Disclaimers for Exchange Administrator s Guide document version 1.8 MAPILab, December 2015 Table of contents Intro... 3 1. Product Overview... 4 2. Product Architecture and Basic Concepts... 4
Open-Source, Cross-Platform Java Tools Working Together on a Dialogue System
Open-Source, Cross-Platform Java Tools Working Together on a Dialogue System Oana NICOLAE Faculty of Mathematics and Computer Science, Department of Computer Science, University of Craiova, Romania [email protected]
International Journal of Scientific & Engineering Research, Volume 4, Issue 11, November-2013 5 ISSN 2229-5518
International Journal of Scientific & Engineering Research, Volume 4, Issue 11, November-2013 5 INTELLIGENT MULTIDIMENSIONAL DATABASE INTERFACE Mona Gharib Mohamed Reda Zahraa E. Mohamed Faculty of Science,
1. Introduction to Spoken Dialogue Systems
SoSe 2006 Projekt Sprachdialogsysteme 1. Introduction to Spoken Dialogue Systems Walther v. Hahn, Cristina Vertan {vhahn,vertan}@informatik.uni-hamburg.de Content What are Spoken dialogue systems? Types
IT Services Management Service Brief
IT Services Management Service Brief Service Continuity (Disaster Recovery Planning) Prepared by: Rick Leopoldi May 25, 2002 Copyright 2002. All rights reserved. Duplication of this document or extraction
FIPA agent based network distributed control system
FIPA agent based network distributed control system V.Gyurjyan, D. Abbott, G. Heyes, E. Jastrzembski, C. Timmer, E. Wolin TJNAF, Newport News, VA 23606, USA A control system with the capabilities to combine
How the Computer Translates. Svetlana Sokolova President and CEO of PROMT, PhD.
Svetlana Sokolova President and CEO of PROMT, PhD. How the Computer Translates Machine translation is a special field of computer application where almost everyone believes that he/she is a specialist.
Touchstone Level 2. Common European Framework of Reference for Languages (CEFR)
Touchstone Level 2 Common European Framework of Reference for Languages (CEFR) Contents Introduction to CEFR 2 CEFR level 3 CEFR goals realized in this level of Touchstone 4 How each unit relates to the
Text-To-Speech Technologies for Mobile Telephony Services
Text-To-Speech Technologies for Mobile Telephony Services Paulseph-John Farrugia Department of Computer Science and AI, University of Malta Abstract. Text-To-Speech (TTS) systems aim to transform arbitrary
Flattening Enterprise Knowledge
Flattening Enterprise Knowledge Do you Control Your Content or Does Your Content Control You? 1 Executive Summary: Enterprise Content Management (ECM) is a common buzz term and every IT manager knows it
Transcription FAQ. Can Dragon be used to transcribe meetings or interviews?
Transcription FAQ Can Dragon be used to transcribe meetings or interviews? No. Given its amazing recognition accuracy, many assume that Dragon speech recognition would be an ideal solution for meeting
Exchange Granular Restore. User Guide
User Guide Contents 1. overview... 2 2. Backup considerations... 3 Backup user identity... 3 Exchange VM Detection... 3 Restore vs. Recovery... 3 3. Creating an Exchange backup... 4 4.... 7 Step 1 - Locate
Exchange Granular Restore User Guide
User Guide Contents 1. overview... 2 2. Backup considerations... 3 Exchange VM Detection... 3 VSS Application backups... 3 Restore vs. Recovery... 3 Backup user identity... 3 3. Creating an Exchange backup...
White Paper. Lepide Software Pvt. Ltd.
Lepide Software Pvt. Ltd. White Paper Purpose of this White Paper is to present a business case for use of Lepide Exchange Manager as a complete Exchange Backup and Recovery solution in conjunction with
Australian Standard. Interactive voice response systems user interface Speech recognition AS 5061 2008 AS 5061 2008
AS 5061 2008 AS 5061 2008 Australian Standard Interactive voice response systems user interface Speech recognition This Australian Standard was prepared by Committee IT-022, Interactive Voice Response
Natural Language to Relational Query by Using Parsing Compiler
Available Online at www.ijcsmc.com International Journal of Computer Science and Mobile Computing A Monthly Journal of Computer Science and Information Technology IJCSMC, Vol. 4, Issue. 3, March 2015,
Dragon Solutions. Using A Digital Voice Recorder
Dragon Solutions Using A Digital Voice Recorder COMPLETE REPORTS ON THE GO USING A DIGITAL VOICE RECORDER Professionals across a wide range of industries spend their days in the field traveling from location
Dragon Solutions Enterprise Profile Management
Dragon Solutions Enterprise Profile Management summary Simplifying System Administration and Profile Management for Enterprise Dragon Deployments In a distributed enterprise, IT professionals are responsible
SPeach: Automatic Classroom Captioning System for Hearing Impaired
SPeach: Automatic Classroom Captioning System for Hearing Impaired Andres Cedeño, Riya Fukui, Zihe Huang, Aaron Roe, Chase Stewart, Peter Washington Problem Definition Over one in seven Americans have
The Big Data methodology in computer vision systems
The Big Data methodology in computer vision systems Popov S.B. Samara State Aerospace University, Image Processing Systems Institute, Russian Academy of Sciences Abstract. I consider the advantages of
CONATION: English Command Input/Output System for Computers
CONATION: English Command Input/Output System for Computers Kamlesh Sharma* and Dr. T. V. Prasad** * Research Scholar, ** Professor & Head Dept. of Comp. Sc. & Engg., Lingaya s University, Faridabad, India
A Multi-Agent Approach to a Distributed Schedule Management System
UDC 001.81: 681.3 A Multi-Agent Approach to a Distributed Schedule Management System VYuji Wada VMasatoshi Shiouchi VYuji Takada (Manuscript received June 11,1997) More and more people are engaging in
Administrator s Guide
Attachment Save for Exchange Administrator s Guide document version 1.8 MAPILab, December 2015 Table of contents Intro... 3 1. Product Overview... 4 2. Product Architecture and Basic Concepts... 4 3. System
Towards a Visually Enhanced Medical Search Engine
Towards a Visually Enhanced Medical Search Engine Lavish Lalwani 1,2, Guido Zuccon 1, Mohamed Sharaf 2, Anthony Nguyen 1 1 The Australian e-health Research Centre, Brisbane, Queensland, Australia; 2 The
Keywords academic writing phraseology dissertations online support international students
Phrasebank: a University-wide Online Writing Resource John Morley, Director of Academic Support Programmes, School of Languages, Linguistics and Cultures, The University of Manchester Summary A salient
Dragon speech recognition Nuance Dragon NaturallySpeaking 13 comparison by product. Feature matrix. Professional Premium Home.
matrix Recognition accuracy Recognition speed System configuration Turns your voice into text with up to 99% accuracy New - Up to a 15% improvement to out-of-the-box accuracy compared to Dragon version
A HAND-HELD SPEECH-TO-SPEECH TRANSLATION SYSTEM. Bowen Zhou, Yuqing Gao, Jeffrey Sorensen, Daniel Déchelotte and Michael Picheny
A HAND-HELD SPEECH-TO-SPEECH TRANSLATION SYSTEM Bowen Zhou, Yuqing Gao, Jeffrey Sorensen, Daniel Déchelotte and Michael Picheny IBM T. J. Watson Research Center Yorktown Heights, New York 10598 zhou, yuqing,
Building Semantic Content Management Framework
Building Semantic Content Management Framework Eric Yen Computing Centre, Academia Sinica Outline What is CMS Related Work CMS Evaluation, Selection, and Metrics CMS Applications in Academia Sinica Concluding
The Clinical Evaluation of Language Fundamentals, fourth edition (CELF-4;
The Clinical Evaluation of Language Fundamentals, Fourth Edition (CELF-4) A Review Teresa Paslawski University of Saskatchewan Canadian Journal of School Psychology Volume 20 Number 1/2 December 2005 129-134
Modern foreign languages
Modern foreign languages Programme of study for key stage 3 and attainment targets (This is an extract from The National Curriculum 2007) Crown copyright 2007 Qualifications and Curriculum Authority 2007
Dragon Solutions Using A Digital Voice Recorder
Dragon Solutions Using A Digital Voice Recorder COMPLETE REPORTS ON THE GO USING A DIGITAL VOICE RECORDER Professionals across a wide range of industries spend their days in the field traveling from location
Data management plan
FACILITATE OPEN SCIENCE TRAINING FOR EUROPEAN RESEARCH 612425 Data management plan Course for Doctoral Students at ECPR Summer School 2015 Faculty of Social Sciences, University of Ljubljana, Slovenia
Agency Pre Migration Tasks
Agency Pre Migration Tasks This document is to be provided to the agency and will be reviewed during the Migration Technical Kickoff meeting between the ICS Technical Team and the agency. Network: Required
Exchange Granular Restore Instructional User Guide
Exchange Granular Restore Instructional User Guide www.backup-assist.ca Contents 1. Exchange Granular Restore overview... 2 2. Creating an Exchange backup... 3 3. Exchange Granular Restore... 6 Step 1
AS-LEVEL German. Unit 2 Speaking Test Mark scheme. 1661 June 2015. Version 1.0 Final Mark Scheme
AS-LEVEL German Unit 2 Speaking Test scheme 1661 June 2015 Version 1.0 Final Scheme schemes are prepared by the Lead Assessment Writer and considered, together with the relevant questions, by a panel of
ANGLAIS LANGUE SECONDE
ANGLAIS LANGUE SECONDE ANG-5054-6 DEFINITION OF THE DOMAIN SEPTEMBRE 1995 ANGLAIS LANGUE SECONDE ANG-5054-6 DEFINITION OF THE DOMAIN SEPTEMBER 1995 Direction de la formation générale des adultes Service
Capacity Plan. Template. Version X.x October 11, 2012
Template Version X.x October 11, 2012 This is an integral part of infrastructure and deployment planning. It supports the goal of optimum provisioning of resources and services by aligning them to business
A Guide to Cambridge English: Preliminary
Cambridge English: Preliminary, also known as the Preliminary English Test (PET), is part of a comprehensive range of exams developed by Cambridge English Language Assessment. Cambridge English exams have
Design with Reuse. Building software from reusable components. Ian Sommerville 2000 Software Engineering, 6th edition. Chapter 14 Slide 1
Design with Reuse Building software from reusable components. Ian Sommerville 2000 Software Engineering, 6th edition. Chapter 14 Slide 1 Objectives To explain the benefits of software reuse and some reuse
Portfolio: Transformation, Modernisation and Regulation
Portfolio: Transformation, Modernisation and Regulation Procurement Committee 19 October 2006 Procurement of E-mail, Calendar and Archiving System Report by: Ward Implications: Head of City Service and
Endowing a virtual assistant with intelligence: a multi-paradigm approach
Endowing a virtual assistant with intelligence: a multi-paradigm approach Josefa Z. Hernández, Ana García Serrano Department of Artificial Intelligence Technical University of Madrid (UPM), Spain {phernan,agarcia}@dia.fi.upm.es
2014/02/13 Sphinx Lunch
2014/02/13 Sphinx Lunch Best Student Paper Award @ 2013 IEEE Workshop on Automatic Speech Recognition and Understanding Dec. 9-12, 2013 Unsupervised Induction and Filling of Semantic Slot for Spoken Dialogue
Interactive Recovery of Requirements Traceability Links Using User Feedback and Configuration Management Logs
Interactive Recovery of Requirements Traceability Links Using User Feedback and Configuration Management Logs Ryosuke Tsuchiya 1, Hironori Washizaki 1, Yoshiaki Fukazawa 1, Keishi Oshima 2, and Ryota Mibe
Dell AppAssure Local Mount Utility
Technology spotlight Dell AppAssure Local Mount Utility A light-weight tool for file and folder restoration The Local Mount Utility (LMU) provides an alternate method to quickly recover files and folders
Evaluation Case Study
Australian Government Department of Education More Support for Students with Disabilities 2012-2014 Evaluation Case Study Team teaching by speech pathologists and teachers in the classroom MSSD Output
Any Town Public Schools Specific School Address, City State ZIP
Any Town Public Schools Specific School Address, City State ZIP XXXXXXXX Supertindent XXXXXXXX Principal Speech and Language Evaluation Name: School: Evaluator: D.O.B. Age: D.O.E. Reason for Referral:
Industry Guidelines on Captioning Television Programs 1 Introduction
Industry Guidelines on Captioning Television Programs 1 Introduction These guidelines address the quality of closed captions on television programs by setting a benchmark for best practice. The guideline
Janison Terms and Conditions. Updated Jan 2013
Janison Terms and Conditions Updated Jan 2013 Terms and Conditions 1. Interpretation 1.1. In this Agreement, unless otherwise indicated by the context (a) (b) (c) (d) (e) (f) (g) (h) (i) words importing
Voice Driven Animation System
Voice Driven Animation System Zhijin Wang Department of Computer Science University of British Columbia Abstract The goal of this term project is to develop a voice driven animation system that could take
System Center Configuration Manager 2007
System Center Configuration Manager 2007 Software Distribution Guide Friday, 26 February 2010 Version 1.0.0.0 Baseline Prepared by Microsoft Copyright This document and/or software ( this Content ) has
ALBUQUERQUE PUBLIC SCHOOLS
ALBUQUERQUE PUBLIC SCHOOLS Speech and Language Initial Evaluation Name: Larry Language School: ABC Elementary Date of Birth: 8-15-1999 Student #: 123456 Age: 8-8 Grade:6 Gender: male Referral Date: 4-18-2008
Robot Task-Level Programming Language and Simulation
Robot Task-Level Programming Language and Simulation M. Samaka Abstract This paper presents the development of a software application for Off-line robot task programming and simulation. Such application
CASSANDRA: Version: 1.1.0 / 1. November 2001
CASSANDRA: An Automated Software Engineering Coach Markus Schacher KnowGravity Inc. Badenerstrasse 808 8048 Zürich Switzerland Phone: ++41-(0)1/434'20'00 Fax: ++41-(0)1/434'20'09 Email: [email protected]
Course Syllabus My TOEFL ibt Preparation Course Online sessions: M, W, F 15:00-16:30 PST
Course Syllabus My TOEFL ibt Preparation Course Online sessions: M, W, F Instructor Contact Information Office Location Virtual Office Hours Course Announcements Email Technical support Anastasiia V. Mixcoatl-Martinez
Adaptive Sampling and the Autonomous Ocean Sampling Network: Bringing Data Together With Skill
Adaptive Sampling and the Autonomous Ocean Sampling Network: Bringing Data Together With Skill Lev Shulman, University of New Orleans Mentors: Paul Chandler, Jim Bellingham, Hans Thomas Summer 2003 Keywords:
GAME: A Generic Automated Marking Environment for Programming Assessment
GAME: A Generic Automated Marking Environment for Programming Assessment Michael Blumenstein, Steve Green, Ann Nguyen and Vallipuram Muthukkumarasamy School of Information Technology, Griffith University
INCREASE YOUR PRODUCTIVITY WITH CELF 4 SOFTWARE! SAMPLE REPORTS. To order, call 1-800-211-8378, or visit our Web site at www.pearsonassess.
INCREASE YOUR PRODUCTIVITY WITH CELF 4 SOFTWARE! Report Assistant SAMPLE REPORTS To order, call 1-800-211-8378, or visit our Web site at www.pearsonassess.com In Canada, call 1-800-387-7278 In United Kingdom,
Natural Language Database Interface for the Community Based Monitoring System *
Natural Language Database Interface for the Community Based Monitoring System * Krissanne Kaye Garcia, Ma. Angelica Lumain, Jose Antonio Wong, Jhovee Gerard Yap, Charibeth Cheng De La Salle University
Project Management Challenges in Software Development
Abstract Research Journal of Management Sciences ISSN 2319 1171 Project Management Challenges in Software Development Uma Sankar S.S. 1 and R. Jubi 2 1 Research and Development Centre, Bharathiar University,
Wilkes University Information & Instructions for Students using Outlook 2003
Outlook 2003 Wilkes University These tips are for how use some of the features of Outlook 2003. You will first have to install Outlook 2003 on your personal computer. After installing Outlook follow the
Chapter 1. Dr. Chris Irwin Davis Email: [email protected] Phone: (972) 883-3574 Office: ECSS 4.705. CS-4337 Organization of Programming Languages
Chapter 1 CS-4337 Organization of Programming Languages Dr. Chris Irwin Davis Email: [email protected] Phone: (972) 883-3574 Office: ECSS 4.705 Chapter 1 Topics Reasons for Studying Concepts of Programming
Linguistics 19 Syllabus List ARABIC
ARABIC List Arabic is a Semitic language belonging to the Afro-Asiatic family of languages which includes Berber; Chadic (including Hausa); Cushitic (including Somali); and Ancient Egyptian. It is the
IC2 Class: Conference Calls / Video Conference Calls - 2016
IC2 Class: Conference Calls / Video Conference Calls - 2016 Technology today is wonderful. That doesn t mean, however, that conferencing calling in a foreign language is easy. In fact, the experience can
ONTOLOGY-BASED APPROACH TO DEVELOPMENT OF ADJUSTABLE KNOWLEDGE INTERNET PORTAL FOR SUPPORT OF RESEARCH ACTIVITIY
ONTOLOGY-BASED APPROACH TO DEVELOPMENT OF ADJUSTABLE KNOWLEDGE INTERNET PORTAL FOR SUPPORT OF RESEARCH ACTIVITIY Yu. A. Zagorulko, O. I. Borovikova, S. V. Bulgakov, E. A. Sidorova 1 A.P.Ershov s Institute
French Language and Culture. Curriculum Framework 2011 2012
AP French Language and Culture Curriculum Framework 2011 2012 Contents (click on a topic to jump to that page) Introduction... 3 Structure of the Curriculum Framework...4 Learning Objectives and Achievement
FAQs Frequently Asked Questions
FAQs Frequently Asked Questions BURLINGTON ENGLISH Table of Contents Page installation Q1 What are the minimum system requirements for installing BurlingtonEnglish? 4 Q2 What are the installation instructions
Turkish Radiology Dictation System
Turkish Radiology Dictation System Ebru Arısoy, Levent M. Arslan Boaziçi University, Electrical and Electronic Engineering Department, 34342, Bebek, stanbul, Turkey [email protected], [email protected]
Common European Framework of Reference for Languages: learning, teaching, assessment. Table 1. Common Reference Levels: global scale
Common European Framework of Reference for Languages: learning, teaching, assessment Table 1. Common Reference Levels: global scale C2 Can understand with ease virtually everything heard or read. Can summarise
Virtual Infrastructure Security
Virtual Infrastructure Security 2 The virtual server is a perfect alternative to using multiple physical servers: several virtual servers are hosted on one physical server and each of them functions both
Big Data with Rough Set Using Map- Reduce
Big Data with Rough Set Using Map- Reduce Mr.G.Lenin 1, Mr. A. Raj Ganesh 2, Mr. S. Vanarasan 3 Assistant Professor, Department of CSE, Podhigai College of Engineering & Technology, Tirupattur, Tamilnadu,
SMART Considerations for Active Directory Migration. A Strategic View and Best Practices for Migrating the Corporate Directory
SMART Considerations for Active Directory Migration A Strategic View and Best Practices for Migrating the Corporate Directory Table of Contents Introduction: The Strategic View of Active Directory Migrations...
1 (a) Audit strategy document Section of document Purpose Example from B-Star
Answers Fundamentals Level Skills Module, Paper F8 (IRL) Audit and Assurance (Irish) June 2009 Answers 1 (a) Audit strategy document Section of document Purpose Example from B-Star Understanding the entity
How to Practice Pronunciation Without a Microphone
dp corporate language training user manual Important information Instruction manual Trademarks corporate language training, clt and digital publishing are either registered trademarks or trademarks of
For each requirement, the Bidder should indicate which level of support pertains to the requirement by entering 1, 2, or 3 in the appropriate box.
Annex Functional Requirements for: The integrated reconciliation system of Back-Office and Cash Accounts operations: Instructions: The Required or Desired column represents whether a feature is a business
Reading Listening and speaking Writing. Reading Listening and speaking Writing. Grammar in context: present Identifying the relevance of
Acknowledgements Page 3 Introduction Page 8 Academic orientation Page 10 Setting study goals in academic English Focusing on academic study Reading and writing in academic English Attending lectures Studying
How to send meeting invitations using Office365 Calendar
How to send meeting invitations using Office365 Calendar This guide tells you how to send meeting requests using the Calendar in your Office365 webmail. There are separate sections for setting this up
Criticality of Schedule Constraints Classification and Identification Qui T. Nguyen 1 and David K. H. Chua 2
Criticality of Schedule Constraints Classification and Identification Qui T. Nguyen 1 and David K. H. Chua 2 Abstract In construction scheduling, constraints among activities are vital as they govern the
A Framework-based Online Question Answering System. Oliver Scheuer, Dan Shen, Dietrich Klakow
A Framework-based Online Question Answering System Oliver Scheuer, Dan Shen, Dietrich Klakow Outline General Structure for Online QA System Problems in General Structure Framework-based Online QA system
