Successful proposal for NeCTAR Virtual Lab funding: Above and Beyond Speech, Language and Music: A Virtual Lab for Human Communication Science (HCSvLab) by The University of Western Sydney is licensed under a Creative Commons Attribution-NoDerivs 3.0 Unported License.
A.1 Program and Proposal Title
Program: Virtual Laboratory
Title: Above and Beyond Speech, Language and Music: A Virtual Lab for Human Communication Science (HCS vLab)
|Organisation Name||University of Western Sydney|
|Contact Name||Professor Denis Burnham|
|Position||Director, Marcs Institute|
|Business Address||University of Western Sydney|
|Postal Address||Locked Bag 1797, Penrith South DC NSW 2751|
|Telephone||(02) 9772 6681|
|Facsimile||(02) 9772 6040|
A.2.2 Participating Organisations
|Organisation / Group Name||Role|
New South Wales University of
A.2.3 Project Funding Summary
|EIF Funds Requested||Total Co-investment Offered|
for public version]
for public version]
for public version]
Section B. Proposal Summary
B.1 Executive Summary
Australia plays a strong and prominent international role in Human
(HCS), encompassing research in speech science, speech technology, computer science, language technology, behavioural science, linguistics, music science,
phonetics, phonology, and sonics and acoustics. However, this position is in jeopardy.
Local Research Effort and Expertise: In Australia we have excellent researchers analysing their own corpora of data using their own analysis tools in relative isolation. Yes,
these researchers meet and share their knowledge, at national/international conferences but (a) relatively infrequently and (b) in discipline-centred contexts.
and Inefficiency: Research conducted in isolation entails local unshared mark-up or augmentation of local data sets, and inefficient repetition of search, information retrieval, annotation, and analysis using tools that are usually home-grown, inaccessible (e.g., idiosyncratic command line execution)
In order to keep abreast of the modern pace of research, to leverage the available yet unrealised interdisciplinary
challenges and opportunities, to go beyond the isolated desk-PC-lab-university-bound model of research – a quantum leap in the way in which we do our Human Communication Science is absolutely essential. We must move into a research environment that will eradicate the waste involved in repeated unshared analyses; ignite the research spark that affords the serendipity of new tool-corpus combinations; and dramatically improve scientific replicability by moving local and idiosyncratic desktop-based tools and data to an easy access, in-the-cloud, public, replicable environment that standardises, defines, and captures procedures and data output (see, e.g., www.myexperiment.org/). We must connect HCS researchers, and their desks, computers, labs, and universities in order to build upon the achievements so far, produce emergent research knowledge, and instil this approach in our new interdisciplinary PhD graduates.
Fortunately, but not without a good deal of forward planning, the common ground for such a Virtual Lab has been carefully prepared. The 2004-2009
ARC-funded UWS-administered Human Communication Science Network (HCSNet) identified a community of over 1000 Australian Human Communication Science (HCS) researchers, engaged that community in over 60 different workshops, seminars and conferences, integrated that community with like international communities,
and augmented the success rate of Australian HCS grant applications, including significant multi-site, interdisciplinary projects such as the ‘Thinking Head’ and the ‘Big Australian Speech Corpus’. The focus that binds this community is the manner in which humans communicate with each other and with computers and machines via codified means — speech and text, music and sound.
The HCSNet experience made it clear that there is a thriving HCSNet community with vast potential for cross-disciplinary research connectivity (i) to provide new insights into old problems by approaching them from different disciplinary perspective s or with a hitherto untried method, and (ii) to apply novel combinations of old ideas or methods of analysis from different disciplines to new problems (see for instance a mcompendium of 30 HCSNet research papers, Dale, Burnham & Stevens, 2011). On
the other hand, the HCSNet experience has also made it abundantly clear that one of the main impediments to the quantum leap that is required for HCS
research to bloom is the difficulty for a researcher from one discipline to apply the tools and techniques of another discipline, or to explore data collected under one paradigm via a completely different analytical perspective.
created the intellectual space between universities, disciplines, paradigms and methods for HCS research to flourish. Above and Beyond HCSNet, the ‘HCS vLab’ will create the virtual space for infrastructure that enables easy access to shared tools and data, and overcomes the resource limitations of individual desktops. It will be a virtual laboratory that makes it possible for HCS researchers from a diverse range of disciplines to access an amalgamation of existing data sets (corpora) and selected analytical tools collected and generated from their midst and then, in this project, put into the cloud. Most importantly, it will enable the guided use of workflow tools and options to allow researchers to cross disciplinary boundaries.
Consider the Music Researcher who wishes to analyse auditory-visual cues that Indonesian singers provide to elicit particular emotions in their audience. Our researcher has in mind someinstances of particular songs, but then where to? How can auditory-visual records of appropriate songs be found, the words transcribed, and the phonetic nuances annotated and aligned with the auditory-visual record and the emotion in the face at any particular time? In HCS vLab a consolidated cross-corpus search would produce candidate songs (probably mainly from the PARADISEC corpus in this case), the text from use of a transcription tool such as Transcriber analysed with the ParGram grammar for Indonesian, the emotions information from the DeMobLib tool, with EMU affording
a prosodic analysis (usually reserved for speech) of the songs. Without HCS vLab our researcher would have had to access PARADISEC, run a search, look around for the tools to analyse it, and probably be unaware of the range of tools that would be of use. With HCS vLab our researcher has searched a complex of corpora, and been guided through a complex of analytical tools and provided with a rich multimodal description of relevant songs. Moreover, the resultant metadata can be stored, as can the sequence of operations, and later be applied to other songs; the HCS vLab will house an ever-expanding enriched data set that will endure, not only for this researcher but for any of the national and international HCS researchers to use and augment. This example demonstrates various strengths of the vLab – multimodal, cross-cultural data and analyses, affective and lexical material, and focus on unique features of the Asia-Pacific region. The same process can be applied to any of the data available through the HCS vLab, with similar or quite different emphases. While this example is somewhat esoteric, more mainstream applications include research supporting automatic speech recognition (taxi ordering, directory assistance etc.), hearing aids and cochlear implants, interactive learning programs for children with learning disabilities, automatic melody recognition, forensic determination of origin and background of particular accents, computer-based information retrieval based on music, speech, sounds, or visual patterns, and psycholinguistic studies of second language learning and pedagogy.
The HCS vLab framework will be built by Intersect, a leading eResearch organisation in NSW with a proven track record in building research software infrastructure. Intersect will be responsible for the technical design of the vLab and will manage the software development lifecycle. The Intersect Project Manager will report to the HCS vLab Project Manager; the project will be governed by the HCS vLab Steering Committee consisting of a blend of senior stakeholders from partner organisations along with a number of technical advisors. The vLab will be delivered in 3 development releases over 14 months, and will incorporate 11 tools and 7 corpora. HCS vLab will be built upon the ground prepared by HCSNet and the raw cross-disciplinary potential that HCSNet seeded. Corpora have been collected and tools have been developed by a range of technically-skilled HCS researchers, and each begs for a common ground, but due to the local needs of individual researchers and the impossibility of one lab or even one institution providing the required person- and computational-power these corpora and tools are isolated on PCs and servers around the country.
HCS vLab will:
- integrate 7 existing corpora, e.g., the 100TB ClueWeb corpus; the 3000-hour ARC-LIEF-funded AusTalk audio-visual speech corpus, the ANDS-funded Australian National Corpus (text & speech), the PARADISEC corpus of speech, text & song in indigenous & endangered languages, and the AMC corpus of Australian music.
- integrate 11 existing tools, e.g., Natural Language Tool Kit (NLTK) for text analytics; EMU for speech analysis and interactive waveform labelling; PsySound3 with physical and psychoacoustic algorithms for sound & music analysis; Johnson-Charniak parsers to generate parse trees; & DeMoLib, for lip-tracking video analysis.These wide-ranging integrations are only possible because of the cohesion of a >1200-strong HCS research community who eagerly await the development of a vLab that will allow full realisation of the collaborative power of HCS research; a community that includes: Bruce Croft & Mark Sanderson (RMIT), top Information Retrieval (IR) researchers in the world; Denis Burnham & Catherine Stevens (UWS, speech & music experts, respectively) leaders on large ARC projects – HCSNet, the Thinking Head, and for Burnham, the Austalk corpus, Janet Fletcher, Andy Butcher & MarijaTabain (UniMelb, Flinders, LaTrobe), world experts in Australian indigenous languages; Steve Cassidy, Mark Johnson, Robert Dale (Macq), and Steven Bird (UniMelb) who are at the hub of language technology research in Australia; and Michael Wagner (UC), Roberto Togneri, Mohammed Bennamoun (UWA) & young stars Trent Lewis (Flinders) & Roland Göcke (UC), who are pioneering AV face and voice analysis. Over and above the realisation of a heuristic HCS research environment, HCS vLab will be:
- accessible: the variety of HCS tools will be accessible to non-technical researchers via workflow tools, stored protocols, and interactive GUIs, while retaining capacity for more sophisticated analyses.
- interoperable: the generic HCS vLab infrastructural support will allow incorporation of HCS corpora from various platforms (e.g., Windows, Mac OS, Linux), and interoperability with other major systems at home, e.g., the already-funded NeCTAR Humanities National Infrastructure (HuNI) vLab, and internationally by virtue of our Product Owner’s intimate domain knowledge and wide collaborations.
- sustainable: 13 universities, 3 organisations, and 47 key investigators have provided $423K in cash and $1.9M in-kind (1.7 times > the request to NeCTAR; with 5 Gold, 5 Silver, 4 Bronze & 2 Other members – see our Partner Model, Appendix E) with partner cash supporting sustained operational development and development of capabilities and reach including future plug-in of additional tools and corpora, and specialist user support well beyond the formal conclusion of HCS vLab construction.
Without the HCS vLab the promise of HCSNet and the careful priming of the HCS community will have been in vain. The HCS vLab is the natural progression for a strong receptive and ambitious research community that is on the move and wants to keep moving. It will create the avenue by which HCS research can make a quantitative and qualitative leap to enhanced capability, collaboration and output, to travel well beyond the geographical confines of individual labs, and well above the disciplinary confines of speech, language, music, or sonics alone into an interdisciplinary and heuristic Human Communication Science cloud-space. B.1.1 Summary for Public Release Applications in automatic speech recognition (taxi ordering, directory assistance etc.), hearing aids and cochlear implants, interactive learning programs for children with learning disabilities, automatic melody recognition, forensic determination of origin and background of particular accents, computer-based information retrieval based on music, speech, sounds, or visual patterns, psycholinguistic studies of second language learning and pedagogy all depend upon research in Human Communication Science (HCS). Human Communication Science encompasses the areas of speech science, speech technology, computer science, language technology, behavioural science, linguistics, music science, phonetics, phonology, and sonics and acoustics. In turn HCS research depends upon datasets (corpora) of speech, music, text, faces, sounds, and specialised tools by which to search, analyse and annotate these data. Australia boasts a strong and active community of HCS researchers who have developed a wealth of corpora and tools relevant to HCS research. However, these researchers tend to analyse their own corpora of data using their own analysis tools in relative isolation. Yes, these researchers meet and share their knowledge, at national/international conferences but (a) relatively infrequently and (b) in discipline-centred contexts. While HCS research in Australia is blooming, especially due to the highly successful Australian Research Council funded HCS Network from 2005-2009, and related research projects, research conducted in isolation entails inefficient repetition of analysis of local data sets. HCS research in Australia, and successful further real-life applications, requires going beyond the isolated desk-PC-lab-university-bound model of research into a new research environment. Such an environment will eradicate the waste involved in repeated unshared analyses; ignite the research spark that affords the serendipity of new tool-corpus combinations; and dramatically improve scientific replicability by moving corpora and tools and the analyses conducted with these into an easy access, shared, in-the-cloud, public, replicable environment. The HCS virtual Laboratory (HCS vLab) will connect HCS researchers, their desks, computers, labs, and universities and so accelerate HCS research and produce emergent knowledge that comes from novel application of previously unshared tools to analyse previously difficult to access data sets. The HCS vLab infrastructure will overcome resource limitations of individual desktops; allow easy access to shared tools and data; and provide the guided use of workflow tools and options to allow researchers to cross disciplinary boundaries. The HCS vLab will be:
- accessible to non-technical researchers via workflow tools, stored protocols, and interactive GUIs, while retaining capacity for more sophisticated analyses;
- interoperable by incorporating HCS corpora from various platforms (e.g., Windows, Mac OS, Linux), and ensuring compatability with other with major systems in Australia and internationally by virtue of our Product Owner’s intimate domain knowledge and wide collaborations; and
- sustainable: 13 universities, 3 organisations, and 47 key investigators have provided $423K in cash and $1.9M in-kind to support sustained operational development and development of capabilities and reach including future plug-in of additional tools and corpora, and specialist user support well beyond the formal conclusion of HCS vLab construction.
The HCS vLab will create the avenue by which HCS research can make a quantitative and qualitative leap to enhanced capability, collaboration and output, to travel well beyond the geographical confines of individual labs, and well above the disciplinary confines of speech, language, music, or sonics alone into an interdisciplinary and heuristic Human Communication Science cloud-space.
B.2 Research Community Profile
HCS vLab will be built by and for the Human Communication Science research community. The Human Communication Science community was formally established with the creation of HCSNet, and HCS researchers are active in various professional organisations as well as in Universities and Research Organisations, as set out below.
1. The Network for Human Communication Science (HCSNet)
The HCSNet community is a broad-reaching interdisciplinary mix of researchers from right across Australia, whose research spans speech, language and music and sonics. The community banded together and obtained funds from the Australian Research Council (ARC) for an ARC Research Network (RN0460284, 2004-2009, $2M, see http://www.hcsnet.edu.au/), a network that greatly facilitated research and research collaboration in the > 1200 community of HCS researchers. HCSNet aimed to build Australia’s reputation as a leader in communication science and technology via advances in its priority areas of ‘Speech’, ‘Effective Human-Computer Interfaces’, ‘Next Generation Search Technology’, ‘Human Communication Disorders’, and ‘Human and Machine Perception and Action’. This succeeded: of the 47 researchers on this Virtual Lab bid, all but the 5 researchers who were not in Australia at the time were HCSNet members; and as a result of collaborations formed and projects hatched in the > 60 HCSNet workshops and conferences during the life of HCSNet, the HCSNet community spawned and continues to incubate various multidisciplinary multi-institutional projects, such as the following: The Thinking Head (ARC/NH&MRC Special Initiatives, TS0669874, 2006-2012, thinkinghead.edu.au) – Research on auditory-visual speech, dialog, speech and speaker recognition, human-machine interaction, avatar and robot development. 11 Chief Investigators (CIs) from 6 Australian universities; 4 Partner Investigators (PIs) from 3 international institutions. Forensic Voice Comparison (ARC Linkage, LP100200142, 2010-2013) – Making demonstrably valid and reliable forensic voice comparison a practical everyday reality in Australia. 3 UNSW CIs, partners in Spain and China, industry partners including Australian Federal Police (AFP), National Institute of Forensic Science Australia (NIFS), NSW and Queensland Police and others. The Big ASC (Australian Speech Corpus, ARC LIEF, 2010-2012, LE100100211, austalk.edu.au/) – Large (1000 speakers x 3 hours speech) auditory-visual speech corpus from 17 sites across Australia. 29 CIs, 11 Australian universities in every state of Australia, 1 international PI. DADA-HCS (ARC SRI e-Research Support, 2005-2006, SR0567319) – Distributed Access & Data Annotation for Human Communication Sciences. 9 CIs, 5 Australian universities and institutions. Other funded projects involving HCS researchers preceded HCSNet and set the scene for the growing zeitgeist in human communication science, e.g., See Hear! The Multimodal Recording and Analysis Facility – new interfaces for analysis of complex visual and auditory scenes, and creation of a research tool for sound and music analysis (ARC LIEF: LE0668448, 2006, 12 CIs from 4 Australian universities, psysound.wikidot.com/start). Such projects have cemented HCS cross-disciplinary links by focussing research effort and harnessing cross-disciplinary research expertise. The result is a mature Human Communication Science research community au fait with the approaches and strengths of other disciplines, but a community yearning for a vehicle to transport these approaches to new lands.
2. The Australasian Speech Science & Technology Association (ASSTA)
ASSTA (est. 1988) is the peak speech science and technology body in Australasia, and the meeting ground for engineers, computer scientists, cognitive scientists, psycholinguists, language technologists, phoneticians, linguists, forensic speech scientists, and speech pathologists via instruments such as its biennial Australasian Speech Science & Technology conference and its international counterpart, Interspeech, which ASSTA hosted in 2008. ASSTA has financially backed HCS endeavours (HCSNet; Forensic Voice Comparison; the Big ASC above), and will financially and academically support the current proposal. Along with other professional bodies (see B.8), ASSTA supports HCS research in Australia via research and conference funding (especially for early career researchers).
3. HCS Community members in Universities and Research Organisations
University of Western Sydney (UWS): UWS, and in particular the Marcs Institute, has played a leading role in the establishment of the HCS community; it was the lead institution on the HCSNet, Thinking Head, See Hear!, and Big ASC projects (see above), and provided and continues to provide cash and infrastructure support for these and like projects. UWS is committed to eResearch, and is a strong supporter of interdisciplinarity and eResearch initiatives. UWS will provide $175K cash and $528K in-kind and ongoing support for the project, support that can only facilitate HCS vLab uptake among HCS researchers. UWS and Marcs Institute will continue to play a leading role in maintaining the HCS community both by its contribution to research and its lead role in this project and promoting the use of HCS vLab. Marcs Institute, elevated from a UWS Centre to one of its 4 Institutes in 2011, is led by the Project Lead on this bid, Prof Denis Burnham. Marcs comprises 51 researchers and 24 higher degree students, who conduct behavioural, neuroscience, and computational research on human-human and human-machine communication in normal, heightened and degraded contexts in 5 programs: Speech & Language, Music Cognition & Action, Bioelectronics & Neuroscience, Multisensory Processing, and Human-Machine Interaction; and in ARC FOR codes areas 1701, 1702, 2004, 1904, 0801, 0906, and 0903. Marcs has current public funding of $6,664,807, comprising 6 ARC Discovery grants, 1 Discovery Early Career Research Award grant, 1 ARC/NHMRC Special Research Initiative, and 1 ARC Linkage Infrastructure, Equipment and Facilities grant across a range of areas that will use and promote the HCS vLab in areas such as auditory-visual speech and cognitive processing; speech perception, regional accents, reading and language acquisition; reverse engineering of the brain; acoustic factors in music perception; human-machine interaction; and corpus studies. Moreover, Marcs has established collaborations with researchers from psychology, linguistics, music, education, computer science, engineering, and various industry partners, in over 20 major research institutes and over 30 additional individuals in Australasia, North America, the UK, and Europe, which will further add to the reach of HCS vLab to the HCS community. Further Universities and Research Organisations: Macquarie University and the lead institution, UWS, have a long history of collaboration and project co-leadership. The convenor of HCSNet, Prof Robert Dale, directs the Macquarie Centre for Language Technology and was a senior investigator in the Thinking Head, Big ASC and DADA projects; A/Prof Felicity Cox and HCS vLab Product Owner, Associate Prof Steve Cassidy are major players in the Big ASC project. Together Macquarie is a major node of HCS research. Other universities in the project all have a history of involvement in HCS research through HCSNet, one or more of the above multi-disciplinary projects, and in their own HCS projects. These HCS community universities contribute to HCS research in the specialist areas as set out hereafter. ANU: Phonetics, Indigenous Languages; Canberra: Speech Forensics, AV Speech and Speaker Recognition; Flinders: Computer Science and AV speech; Melbourne: Engineering, Phonetics; Sydney: Indigenous Song, Music, Speech Pathology; Tasmania: Cognitive Science, Psychology; UNSW: Speech Science, Music, Emotion; UWA: Speech and Speaker Recognition; RMIT: Information Retrieval (IR) for very large databases, dialog and multi-agent systems, models in AI and computer science; UNE: Psycholinguistics, regional languages, logic in child language; LaTrobe: Language diversity, minority languages, Australian indigenous languages, data-oriented and theory-oriented approaches; NICTA: Machine Learning and NLP. The wide geographical and disciplinary spread of the 14 partner research organisations across Australia, along with the active research profiles of the 47 individual researchers in the bid, with their ongoing Higher Degree Research (HDR) student load and numerous significant national and international collaborations, provide a strong scaffold of support and uptake for this project.
B.3 Development Organisation Profile
1. Intersect Australia
Intersect Australia Ltd is a not-for-profit company limited by guarantee, owned and funded by its members, the universities in NSW, state government departments, and other organisations undertaking research in NSW. Intersect has a strategic focus on national research infrastructure. Intersect is a member of The National Computational Infrastructure (NCI), and the Australian Access Federation (AAF). Intersect has undertaken and is undertaking many projects deploying data capture and management solutions for the Australian National Data Service (ANDS). The software Intersect develops integrates with infrastructure provided through these bodies. Since its establishment, Intersect has demonstrated that it is one of Australia’s leading eResearch organizations in having the capability and capacity for undertaking eResearch projects.
Capacity and Capabilities
- Intersect has approximately 50 staff. It has established a capacity and capability to develop, deploy and support substantial and complex eResearch infrastructure, that is unique in Australia. This capability is built on a company culture which emphasises a focus on the client and of engineering excellence. Intersect has built a team that delivers eResearch solutions on time, on budget and of value.
- Intersect’s Engineering Division brings together many years of commercial experience in developing large scale IT systems across many sectors such as academia, government, banking and enterprise security tools. The team of 30 staff includes user interface designers, specialist test engineers, software and systems engineers and project managers.
- Intersect’s Services Division staff have backgrounds in publically funded research, commercial research and development, and commercial information technology service provision. The team of eleven staff is responsible for outreach and engagement before, during and after development commences. They provide capability to carry out stakeholder management, requirements gathering, and product ownership.
- The Operations Division, comprising five staff, has been centrally involved in systems integration projects and the transitioning of projects from development to commissioning and ongoing hosting.
Track record and relevant experience
- Intersect has undertaken and successfully delivered many eResearch projects (at the time of writing approximately 25), including projects with development and integration budgets in excess of $1 million. These projects provide solutions and infrastructure to research efforts across a range of disciplines. These projects include:
- analysis and integration projects for many NCRIS capability areas (e.g. AMMRF, PHRN, AAL) analysis and development projects for non-university research bodies (ANSTO, NSW Office of Environment and Heritage)
- software development projects funded through PfC capabilities (e.g. 11 ANDS data capture projects for 4 universities) software development projects funded directly by our membership (e.g. Rainfall) strategic software development projects funded by Intersect (e.g. Genomic Data Repository, Australian Schizophrenia Research Bank)
- In the vocabulary of NeCTAR, a number of these projects would fit within the parameters of eResearch tools (e.g. ANDS data capture projects, ASRB) or virtual laboratories (e.g. PHRNi). Intersect is currently delivering six ANDS-funded projects for its members.
Please see the attached letters of support for Intersect’s track record as a development partner.
Approach to quality standards
Intersect has not sought formal certification under any standards (e.g. ISO 9000). Intersect follows a three part method for achieving quality:
- “Say what you are going to do”. Starting from the concept stage of this project, and continuing throughout the project, Intersect keeps the customer informed of what they are doing and how they are doing it. Intersect has processes covering Consultation, Business Analysis, Project Management and Software Engineering. Intersect works with clients to tailor these to suit their needs and the project’s needs.
- “Do what you said you were going to do”. Intersect follows their processes. If issues are encountered then Intersect talks to the customer to find agreeable solutions.
- “Prove it”. Intersect keeps the smallest quality record possible, as documented in a quality management plan.
This plan is written during the elaboration stage of development (see Item B.15), in conjunction with the stakeholders and NeCTAR.
Support and warranty mechanisms
Intersect issues a formal 3-month warranty for all projects; the warranty commences after user-acceptance testing has completed. During this period all defects are fixed at no cost to the customer. Intersect’s on-going defect rate is less than one new defect discovered per month, across 25 deployed systems. In practice, the defect rate has been low enough that Intersect has fixed the majority of defects that come to light after the warranty period has expired. Additional feature requests are carried out on a fee-for-service basis.
- RMIT. The Information Retrieval (IR) group in the School of Computer Science & IT at RMIT University is recognised as an international leader in the development of search engines: producing 28 A or A* journal/conference papers in the last ERA period. The ISAR group has extensive experience building open source search engines, creating the Zettair and MG systems, both widely used. In a separate strand of research, the group also evaluates search engine effectiveness. Based on citations to its last five years of publication outputs, the RMIT IR group was placed in the world’s top 15 IR research groups by Microsoft’s Academic Search (http://academic.research.microsoft.com/); and No. 1 in Australia. Three members of the RMIT group will contribute.
- NICTA. NICTA is Australia’s centre of excellence for ICT research, with over 600 researchers and students, including many world leaders in their fields. NICTA’s mission is to deliver outstanding ICT research outcomes, and to create wealth for Australia through the application of that research. NICTA performs research in a number of areas, including Machine Learning and Control and Signal Processing, which include outstanding researchers in Language Technology and Text Retrieval. NICTA’s Engineering and Technology Development team is expert at transitioning research into innovative technology solutions to real problems.
- RMIT/NICTA Software Engineering. An important part of the HCS vLab’s toolkit will be the ability to assemble “processing pipelines” involving multiple tools processing the same data sequentially. Based on the combined expertise of the NICTA and the RMIT staff involved in this project, a Research Engineer will be employed via sub-contract to help the Intersect engineering team build a flexible component architecture for the HCS vLab compatible with UIMA, an emerging standard for wrapping components for processing language, speech, video, and other unstructured data.
B.4 Operational Organisation Profile
Intersect Australia Ltd is also the Operational Organisation.
Capacity and capabilities
Intersect provides hosting, operations, outreach, L&D and support for our members’ and affiliates’ eResearch needs. Intersect hosts its production infrastructure with commercial hosting partners ‘ac3’. The hosting is located at the Global Switch data centre in Ultimo. Intersect hosting partners provide managed services, including backup, system monitoring and logging, and core network connectivity for all services. The Intersect systems administration team performs network management, user-level support and troubleshooting, and general systems administration. Intersect has an access agreement with AARNet; all systems hosted by Intersect are “on-net”. Issues with services hosted at ac3 are notified to the head of Intersect’s systems administration team. Operations, advocacy, L&D and support are built on a full-time team of ten eResearch Analysts, an HPC specialist, a data management specialist and five systems administrators. The team is currently responsible for:
- Operations of and merit-based allocation to HPC facilities, both through Intersect’s own McLaren service, hosted at ac3, and Intersect’s partner share of NCI.
- Hosting data and applications on behalf of Intersect members’ researchers. We host off-the-shelf applications, customized open-source applications, bespoke software developed by Intersect, data accessible via the DataFabric, and data management systems. Hosted systems include Confluence, Jira, OpenClinica and OpenCDMS.
- Hosting hardware on behalf of Intersect clients, for example the SAX Institute.
- Providing first-tier support for HPC users, as well as users of national services such as the ARCS Data-Fabric, ARCS Grid Computing Service, and AAF’s authentication services.
Intersect’s team of Analysts has a background in publicly funded research, commercial research and development, and commercial information technology. They provide both one-off and ongoing assistance to research groups, and combine experience across a large number of research disciplines. Intersect’s system administration team brings together commercial- and research-based experience supporting and integrating administrative as well as research systems. Intersect has applied to the RDSI NoDe program and has a plan to build and operate an RDSI node. Intersect has applied to the NeCTAR Research Cloud program, and has a plan to build and operate a research cloud node. Both services will operate along side our existing infrastructure, hosted in commercial hosting facilities.
Track record and relevant experience
- Intersect provides operational support for HPC, valued at approximately $1m annually, to its membership. Over the last three years, Intersect has provided support directly to more than 100 research projects comprising hundreds of researchers. In the last year Intersect provided specialist ongoing support, in the form of scripting, troubleshooting, compilation of software, and design of experiments, to approximately 30 research projects.
- Intersect hosts applications or application spaces for approximately 30 groups across seven institutions. Hosting is provided through a combination of in-house infrastructure and in partnership with ac3.
- Systems administrators at Intersect have been involved in the roll-out and support of national services, through most of the PfC programs, including ARCS, the AAF and ANDS.
- Intersect’s eResearch Analysts provide ongoing support comprising engagement and outreach, issue management, and learning and development to hundreds of researchers each year.
Please see the attached letters of support for Intersect’s track record as an operating partner.
Approach to quality standards
Intersect has systems and procedures in place that provide quality control and assurance to their customers. The tools used in these systems and procedures include: ‘Jira’ to raise and track external and internal issues; ‘Nagios’ to automate service monitoring and raise alerts; ‘Cacti’ for usage trend analysis and reporting; ‘Splunk’ for log management, troubleshooting and forensics; and ‘Confluence’ to document and manage procedures and processes. There is an end-to-end process for raising and resolving support issues, including a process by which support issues are prioritized and escalated.
Support and warranty mechanisms
Intersect issues a formal 3-month warranty for all projects; the warranty commences after user-acceptance testing has completed. During this period all defects are fixed at no cost to the customer. In practice, the defect rate has been low enough that Intersect has fixed the majority of defects that come to light after the warranty period has expired. Additional feature requests are carried out on a fee-for-service basis. The systems administration team provides user support for hosted services. The systems administration team works with the customer to configure the initial setup of their service. This work includes making a decision on the appropriate hosting arrangements (e.g. local, Intersect, commercial) based on the required level of service. After commissioning, issues are tracked using Jira, and issues are triaged into support and defect cases. Defects are escalated to the engineering team. Responsibility for each support issue is assigned to a case manager, who looks after the reporter until the issue is either resolved or escalated. All services are monitored automatically using Nagios. Outages and anomalies are reported to the systems administration team, and emails are sent to the owners of the service using mailman. Scheduled outages are negotiated with the customer as they are required e.g. for upgrade of hardware or maintenance releases of software. Services are promoted through Intersect’s website, newsletter, marketing collateral and outreach program. Outreach activities include promotion of services, including through an Intersect-maintained tools register (soon to be integrated with CAUDIT’s eResearch portal), as well as providing one-on-one assistance to researchers deciding on and using services in their research. Training is supported though Intersect’s learning and development (L&D) program. Existing L&D material has been developed for services including: interactive and self-paced training courses (covering e.g. HPC, Google’s REFINE); written material available through our website (e.g. guides to Evo, Collaborative Authoring); and an emerging program of web-casts demonstrating concepts and usage of tools (e.g. Subversion).
B.5 Other Participants
In addition to UWS (project leadership and management, access to the AusTalk (BigASC) (Burnham), and the AMC (Dean in association with the John Davis, CEO of the Australian Music Centre) databases, and the ParseEval (Shaw) tool, the following institutions or groups will be involved. Details of involvement and contribution are given in Part D3.
- Intersect – Project development, business analysis, and co-investment.
- Macquarie U – Specialists in Phonetics and in Language Technology. Access to the EMU, AusNC (Cassidy), and the Johnson-Charniak parser (Johnson) tools; adaptation of the audio aspects of AusTalk (Cox).
- U of Melbourne – Specialists in Phonetics, Linguistics and Engineering. Adaptation and testing of the NLTK (Bird), PARADISEC tools and the speech component of the PARADISEC corpus (Thieberger).
- Sydney U – Specialists in Music and Speech Pathology. Adaptation of the PARADISEC (Music aspects) corpus (Barwick); user testing and feedback (Arciuli), adaptation and testing of PsySound3 (Carbrera).
- ANU – Specialists in Phonetics and Indigenous Languages. Adaptation of Indigenous languages and text aspects of PARADISEC corpus (Simpson), adaptation of the Indonesian corpus (Arka, Mistica), user testing and feedback (Ishihara).
- Flinders U – Specialists in Computer Science and Auditory-Visual (AV) Speech. User testing and feedback (Powers, Lewis); adaptation of AV aspects of AusTalk and advice on AV aspects of the project (Lewis).
- UNSW – Specialists in Speech Science, Music, Emotion in speech and music, and in Forensic Speech Science. User testing and feedback (Epps, Ambikairajah, Cabrera, Emery).
- UWA – Specialists in Robust Speech, Speaker Recognition, and 3D audio-visual speech and speaker recognition. Adaptation and testing of the modifications to HTK (Togneri); Visual and 3D processing for recognition (Bennamoun); User testing and feedback, audio-visual feature processing advice (Togneri, Bennamoun).
- U Canberra – Specialists in AV speech, Automatic Speech Recognition, Forensics. Adaptation of AVOZES corpus and of the DeMoLib liptracker tool (Goecke); User testing and feedback (Wagner, Goecke).
- U Tasmania – Psycholinguistics and reading studies. User testing and feedback (Kemp).
- RMIT – Information Retrieval and Natural Language Processing (Sanderson, Croft).
- UNE – Psycholinguistics, regional languages, logic in child language. User testing and feedback (Khlentzos, and the Language and Cognition Research Centre).
- LaTrobe – Language diversity, minority languages, data-oriented and theory-oriented approaches. Integration post-project of the VisLab for the remote use of scientific instruments and imaging of scientific data (Schembri); User testing and feedback (Tabain).
- NICTA – Specialists in Machine Learning and Control and Signal Processing, including Language Technology and Text Retrieval. Support for interoperability with UIMA, the Unstructured Information Management Architecture, an emerging standard for wrapping components for processing language, speech, video, etc. data (Cavedon, Vespoor).
- AusNC Inc. – The Australian National Corpus – Adaptation of AusNC corpora and tools for HCS vLab; contribution of expertise on the licensing of corpora for online use and the core technical platform developed to ingest corpus data and meta-data into a unified online format (Haugh, Cassidy, Goddard).
- ASSTA – the Australasian Speech Science and Technology Association – Peak body on speech science research in Australasia. Promotion of the HCS vLab through electronic bulletins, the ASSTA Newsletter and the biennial Speech Science and Technology (SST) conference; advice from speech science and technology experts as required; liaison and HCS vLab promotion with international counterpart, The International Speech Communication Association ISCA.
B.6 Key Personnel
- Project Director, Professor Denis Burnham, Inaugural Director (1999- ) of Marcs Institute, UWS conducts research in behavioural and speech science with collaborators from music cognition, linguistics, phonetics, engineering, computer science and creative arts. He is President, Australasian Speech Science & Technology Association (ASSTA, 2002- ); Member, ISCA (International Speech Communication Association) International Advisory Council and Interspeech Steering Committee; and Co-Founder, Auditory-Visual Speech Perception Association. Burnham has held over 30 externally-funded grants (over 20 as Leader): he has led ARC Linkage projects with industry partners Australian Caption Centre, and with Cochlear Ltd; and large interdisciplinary projects, e.g., the $2M 5-year ARC Research Network on Human Communication Science (HCSNet) (Triumvirate CI); and Leader of the $3.4M 5-year ARC and NH&MRC Special Initiative Thinking Head, the ARC $1M Big Australian Speech Corpus (Big ASC), and most recently Seeds of Literacy, a 5-year $750K ARC Discovery.
- Project Manager, Dr Dominique Estival has a PhD in Linguistics and extensive experience in academic research and commercial project management for Language Processing. Following research, industry and academic positions in the USA, Europe and Australia, she took up Project Management roles: Team Leader, R&D, Syrinx Speech Systems, a Sydney speech recognition company developing automated telephone dialogue systems; Senior Research Scientist, Natural Language technologies, human-computer interfaces and multi-lingual processing with the DSTO (Defence Science & Technology Organisation); and Senior Manager, Projects and Research, managing language processing research for US-government-funded and commercial projects at Appen P/L, a company providing speech and language databases for language applications. Estival is currently Project Manager, the Big ASC (Australian Speech Corpus) at UWS, where she has managed rollout of software and hardware to 17 Australian sites for AV recording of the AusTalk corpus. Estival is a founding member of the Australasian Language Technology Association (ALTA) and in 2008 established the Australian Computation and Linguistic Olympiad (OzCLO). Dr Estival has a wealth of experience in academia and industry, including project management of large collaborative projects and will, therefore, be employed at the Manager level.
- Product Owner, Associate Professor Steve Cassidy, Macquarie University is a Computer Scientist whose research covers the creation, management and exploitation of language resources. Cassidy has a PhD in Cognitive Science and has worked in both Linguistics and Computer Science departments. He is the main author of the Emu speech database system which is widely used in the creation and analysis of spoken language data for acoustic phonetics research. He has been involved in the standardisation of tools and formats for the exchange of language resources starting with his work on the Emu system and more recently as an invited expert on the ISO TC 37 working groups on annotation interchange formats and query languages for linguistic data. He has been instrumental in establishing the Australian National Corpus as an umbrella organisation to manage language resources in Australia and is an active collaborator with similar projects in the US and Europe. Cassidy will act as Product Owner for this project and as such will act as a conduit between the development team and the prospective users around Australia as well as ensuring that the product is interoperable with related international efforts.
- eResearch Analyst, Peter Bugeia (Intersect, UWS) is the eResearch Analyst for the University of Western Sydney. He has 27 years IT experience across a wide range of industries including medicine, banking, finance and media. He has worked in commercial, not-for-profit and public sectors and has held various roles from Senior Software Engineer and Test Manager to Project Manager, Enterprise Architect and Business Analyst.
- Intersect Project Manager. Georgina Edwards is Intersect’s technical development manager assigned to manage the software development aspects of the project. She has over 10 years experience in commercial software design and development, and has worked in banking and finance as well as eResearch. Edwards has a BE (Hons) in IT & Telecommunications from the University of Adelaide. She has experience building web applications in a range of languages, and is also an experienced agile practitioner. Edwards will work closely with the Project Manager and Product Owner to ensure proper and effective integration between the stakeholder community and the software engineering team. Edwards will also work with Intersect’s management team comprising of Dr Ian Gibson (CEO), Rodney Harrison (Engineering Manager), Dr Joe Thurbon (Services Manager), Shane Youl (IT Manager) to ensure that the development and operations of the project are staffed appropriately and executed efficiently.
The HCS vLab will constitute a collaborative environment for access and analysis of human communications data by HCS tools. It will provide resources to create new annotations for existing data, and a space for researchers to store new data and tools for use by the research communities. The overall structure of the environment is shown in Figure 1. The HCS vLab is designed to make use of national infrastructure – including data storage, discovery and research computing services. It incorporates existing eResearch tools adapted to work on shared infrastructure, and orchestrated by a workflow engine with both web and command line interfaces. Dr Estival will be Project Manager and will be working with A/Prof Steve Cassidy, the Product Owner. The Project Manager will be part of the Steering Committee and have oversight of the development undertaken at both Intersect and RMIT/NICTA.
The HCS Virtual Laboratory will provide researchers with an integrated environment in which to select and perform analysis on Corpora through a suite of pre-installed tools. The HCS vLab will be designed for use by both IT-technical and non-IT-technical researchers, with user interaction through a Web interface. General functionalities of the HCS vLab are as follows:
- Users can browse lists of corpora containing Human Communications data and of pre-installed HCS Tools, including corpora from the already-funded NeCTAR Virtual Lab, HuNI (the Humanities Networked Infrastructure) (see Tools and Corpora below, and also B.18).
- Users can select either a single corpus for search and analysis or several corpora to perform a federated search. Some users will also be able to select and add their own data sets for search and analysis.
- Users then select one or more tools by which to analyse the selected data. The system will display runtime options for the selected tools. Users will then be able to choose and save the most appropriate options for their analysis. The system will ensure only valid options are selectable.
- Once one or more corpora or data sets have been selected and tool options have been chosen, the user can invoke execution of the tools and the vLab will run the tools in their execution environment. Some tools may be configured to run in multiple execution environments, and a special HPC execution environment will be available for compute-intensive tools (Intersect has provided an initial in-kind allocation of HPC computing time for the HCS vLab). During execution, vLab will copy and/or make available data and files to the selected tool transparently to the user.
- The user will be able to monitor and control execution as it proceeds, and terminate if necessary; the user will be able to request a change in the computing resources assigned to an executing tool.
- Once execution is complete, the user will be able to view the results through the Web interface.
- Tools will either automatically add results to the Annotation and Record-Level Metadata Stores and/or the user will make these updates manually through an Annotation Service. Annotations and metadata will be private to the originating researcher until the researcher chooses to make them publicly available.
- The UWS Research Data Catalogue, managed by the UWS library (established with the help of ANDS Metadata Stores and Seeding the Commons programmes) will play a key role in describing and disseminating descriptions of corpora, tools (as services) and annotations (new data sets).
- A command line interface will also be available for users.
- For Desktop Tools, the user will be able to interact with the application once it has been initiated.
- The user will be able to chain the execution of Tools together, and capture and share these workflows.
- As some of the Corpora data are sensitive, Corpora will only be available to researchers who are appropriately authorised for that particular corpus. This will be checked through the “User Registration” service.
- Users will be able to request the addition of Tools and Corpora to the vLab, and to store their own private data, subject to the level of commitment to the Virtual Lab from their organization as described in the Partner Model, with appropriate service support and regulation for authority.
The HCS vLab, in particular the Workflow Engine, will be built around the Galaxy open source workflow management system (see also B.11). The proposed Technical Architecture of the HCS vLab is as follows:
- An instance of the Workflow engine will be run on a virtual machine in the Intersect, NeCTAR, or other Cloud.
- The Standard Execution Environments will be pre-configured virtual machines that have one or more tools installed. The HPC Execution Environment is likely to be a standard operating environment with no virtualisation.
- The Tool & Corpora Definitions, Record-Level Metadata Store and Annotation Store will be cloud-based databases which are available to all virtual machines that form part of the vLab.
- The Corpora will live on the Intersect RDSI Node, or some other RDSI node, closely located to HPC compute resources.
- Within an execution environment, tools will be run in either the foreground or in batch mode, depending on what is appropriate for the tool and the environment.
Tools and Corpora
The HCS vLab will integrate existing corpora that house human communications data, consisting of language and music data, in the three most common modes in which these are represented – audio, auditory-visual and text. The corpora to be made available in this project are all corpora which our participant members have either established or are caretakers for, so we have direct access to these. They are presented below in the order in which they will be incorporated into the framework (see also B.13) and further details regarding platforms, UIs, input and output etc. are provided in Appendix B. The order of incorporation of these corpora into the framework is chosen by consideration of the joint factors of maturity of each corpus and the amount of work required to adapt them for incorporation into the HCS vLab.
- PARADISEC (the Pacific and Regional Archive for Digital Sources in Endangered Cultures, including Indigenous languages music, and speech) (5.1TB);
- AusTalk AV speech corpus from the BigASC project (7TB);
- the Australian National Corpus (incorporating the Australian Corpus of English (ACE), Australian Radio Talkback (ART), AustLit, Braided Channels, Corpus of Oz Early English (COOEE), Email Australia, Griffith Corpus of Spoken English (GCSAusE), International Corpus of English (Australia contribution is ICE-AUS), the Mitchell & Delbridge corpus, and the Monash Corpus of Spoken English (5TB);
- AVOZES visual speech corpus (15GB);
- Australian Music Centre archive (extremely large collection of sound and text: over 30,000 items by 530 artists);
- Colloquial Jakartan Indonesian corpus (audio and text 32.5TB);
- The ClueWeb09 dataset (100TB).
The HCS vLab will also integrate existing HCS tools for the analysis of music, speech and written text and make them accessible to non-technical researchers, while maintaining a command line functionality for more sophisticated analyses. Nine of these are tools that have been developed by our participant members and these are listed below in the order in which they will be incorporated, by consideration of the joint factors of maturity of the tool, the amount of work required to adapt the tool for incorporation into the HCS vLab and the order in which corpora will be incorporated (see B.13).
- EOPAS (PARADISEC tool) for text interlinear text and media analysis;
- NLTK (Natural Language Toolkit) for text analytics with linguistic data;
- EMU for search, speech analysis, interactive labelling of spectrograms and waveforms;
- AusNC Tools: KWIC, Concordance, Word Count, statistical summary and statistical analysis on a user-defined subset of content;
- Johnson-Charniak parsers, to generate full parse trees for text sentences;
- ParseEval, tool to evaluate the syllabic parse of consonant clusters;
- HTK – modifications, a patch to HTK (Hidden Markov Model Toolkit, http://htk.eng.cam.ac.uk/) to enable missing data recognition;
- DeMoLib software for video analysis; and
- PsySound3 (physical and psychoacoustical algorithms) of complex visual and auditory scenes.
- ParGram (grammar for Indonesian).
- The INDRI tool for information retrieval with large data sets.
For each of the corpora and tools listed above, there is a HCS expert who will work in-kind for 3 weeks (15 effort days) with the Project Manager, the Product Owner and Intersect to incorporate these corpora and tools into the HCS vLab (see B.13).
Collaboration with HuNI (an already funded NeCTAR Virtual Lab)
The Virtual Lab will query the HuNI virtual lab using a protocol to be negotiated (e.g. OAI-PMH or Atom) for information about corpora known to HuNI which may be of use to VL users. Appropriate corpora with sufficient metadata can then be loaded into the HCS vLab and used with the suite of HCS vLab Tools. Resulting data will be stored in the HCS vLab Annotation store, and the existence of the new data advertised back to HuNI via an appropriate mechanism. Metadata exchange will use an appropriate standards, such as OAI-PMH with discipline and corpus-appropriate metadata schemas (EAC-CPF for information about parties, OLAC for linguistics resources, MARC for bibliographic description). Figure 2 shows a potential workflow where an HCS vLab user is able to acquire data from HuNI, transcribe it, and publish it so that a HuNI user can access the original HuNI object along with the transcript using one of the HCS web-based tools, giving them access to new data has been created in the HCS vLab environment. Similarly, HuNI users will be able to access key HCS vLab tools, such as time-aligned transcription tools, which save data in standard reusable formats, rather than using ad hoc solutions as is currently often the case.
Figure 2: A potential scenario involving interoperable metadata, tool and data exchange between HCS vLab and the HuNI vLab
RESEARCH COMMUNITY NEEDS & BENEFITS
B.8 Target Research Community
The primary target research community to use the HCS vLab is encompassed by HCSNet (see B.2), an Australian community with >1200 active members from Speech, Language, Music and Sonics areas and around 45,000 international members. More specifically, the international music community that wants to have the AMC not only more easily available but open to different kinds of searches is estimated at around 10,000. The research communities that would benefit from Text and Speech corpora and associated tools can be estimated at 20,000 (linguists, speech scientists, behavioural scientists and language technologists). For video and visual analysis, the international community is estimated at around 15,000 researchers and is steadily growing. The Australian and international HCS community also intersects and overlaps with various professional bodies as set out below and, via these, with their international counterparts.
- The Australasian Speech Science and Technology Association (ASSTA, est. 1972, www.assta.org/) is the peak body for speech scientists in Australia and New Zealand and is a cash contributor to this bid. ASSTA has various Corporate Members (Appen Butler Hill, Cochlear Pty Ltd., the HEARing CRC, Spectral Dynamics) and 110 members from the disciplines of engineering, computer science, cognitive science, psycholinguistics, language technology, phonetics, linguistics, forensic speech science, and speech pathology. ASSTA runs the biennial Speech Science & Technology conference, provides HDR student travel assistance for international conference presentations, and funds speech science and technology research events and initiatives. ASSTA has close ties to ISCA, the International Speech Communication Association, which attracts > 1000 registrants to its annual conference, Interspeech. ASSTA members make up the core set of researchers working in the various manifestations of speech science in Australasia. Almost all the members were active participants in HCSNet, and ASSTA members are at the forefront of grant-getting, publication, and PhD supervision in speech science in Australasia.
- The Australian Linguistics Society (ALS, est. 1967, www.als.asn.au/) with more than 450 members, is the peak organisation for linguists in Australia. It runs the annual ALS conference and a biennial Australian Linguistic Institute (ALI). Many ALS members participated in HCSNet activities. The main international counterpart is the LSA (Linguistic Society of America) with around 4,500 members, and there are individual Linguistics organisations in many countries around the world, with an estimated combined total of 10,000.
- The Australasian Language Technology Association (ALTA, est. 2002, http://www.alta.asn.au/) with around 240 members promotes research in Computational Linguistics and Natural Language Processing. It is a founding regional organisation of the Asian Federation of Natural Language Processing, runs the annual ALTA workshops, and manages funds for OzCLO (the Australian Computational and Linguistics Olympiad). Its main international counterpart is the ACL (Association for Computational Linguistics), with around 5000 members and there are more specialised organisations, such as AMTA (Association for Machine Translation in the Americas), EAMT (European Association for MT), AFNLP (Asian Federation of Natural Language Processing), with an estimated combined total of 10,000 researchers around the world. Most ALTA members participated in HCSNet activities.
- The Australian Music and Psychology Society (AMPS, est. 1996, www.ampsociety.org.au/), with 200 members, is a member of the Asia-Pacific Society for the Cognitive Sciences of Music (APSCOM) and has links with ESCOM, the European Society for the Cognitive Sciences of Music and SMPC the (US) Society for Music Perception and Cognition. AMPS runs workshops, seminars and conferences, and provides HDR student travel assistance for international presentations, e.g., at the annual International Conference on Music Perception and Cognition which AMPS hosted in Sydney in 2002. Many AMPS members participated in HCSNet activities.
B.9 Needs and Impact
Needs The HCS research community has produced a number of corpora and repositories for human communications data and many HCS tools to manipulate, process and analyse these data. But the use of and access to these data is hampered by two constraints. First, as in so many other disciplines, the amount of data available is growing exponentially, making it increasingly infeasible for any one researcher or research laboratory to maintain up-to-date locally stored datasets. Even where the storage capacity to do this exists at a given institution, the many-ways replication that local copying strategies encourage only introduces version inconsistency problems that outweigh the advantages of redundancy. Cloud-based storage, with appropriate backup procedures, is widely accepted as the best way forward, ensuring that all users see the same data. Similarly, cloud-based analytical tools relieve individual researchers and sites of the need to maintain up-to-date versions of software, outsourcing software infrastructure tasks that are not, and should not be, core business for a researcher. The second problem is unique to interdisciplinary endeavours. It is already well-recognized that a major impediment to interdisciplinary interaction is the fact that we ‘speak different languages’: two researchers aiming to cross a disciplinary divide need to take the time to understand how each uses terminology in often subtly different ways. A related phenomenon is present where ‘the rubber hits the road’ in terms of actual software and data usage: although a researcher in one discipline may have much to gain, and much to offer, from the use of tools and corpora developed in another discipline, all too often the hurdles to making this a reality are immense, posing a learning curve that has characteristics not dissimilar to the culture shock one faces when moving to a new country. Cloud-based tools and support cannot erase the difficulties here, but they provide an opportunity for interfacing and modularising that make it easier to overcome them. A further key concern that our proposal addresses is that of replicability. At a time when conventional science is subject to disruptive forces—debates over open access models, attacks on the peer review process, and a sense of public distrust—this element of the scientific process remains indisputably indispensable. But replicability has always been hard, and gets harder as experiments require ever-more technically sophisticated tools, and make use of ever-larger data sets. This problem has been recognized in Big Science, whether dealing with web click histories, consumer purchasing patterns or astronomical data. But it is all too easy to overlook the importance of replicability in sciences, like Human Communication Science, that rely on ‘mid-size’ data sets. Our HCS vLab proposal addresses this problem head-on, by providing a cloud-based platform that supports user-defined experimental workflows using standardised public-domain data sets and cloud-friendly revisions of existing desktop analysis tools. In this regard, our aim is to develop and provide a world-leading best-practice platform for scientific replicability in the human communication sciences. To address these issues we need a connected repository for data and integrated access to tools. The HCS community will benefit from (i) easy access to data collected in other disciplines, e.g., speech and video researchers accessing the indigenous language PARADISEC data, linguists accessing the AusTalk transcripts, and the speech/language components of music in the Australian Music Corpus; and (ii) the use of tools developed in neighbouring disciplines to process and manipulate their own data, e.g., linguistic tools to process the language components of the AMC, visual analysis and lip-tracking tools to analyse the video components of AusTalk, and musical and acoustic analysis tools to examine rhythm and melody in speech corpora. The potential impact of making these tools and resources easily available to the vastly distributed Australian research community is vast. This is perhaps most obvious where the resources and tools have a direct connection to commercially-valuable technological developments, as in the areas of speech and language technology, and music processing software. Australia’s small size means that we will always struggle to compete against the major players in the US and Europe, and increasingly in Asia, but HCS vLab will enable a pooling of resources, data and tools that will encourage a higher degree of collaboration amongst our researchers, and allow us to do more with less. Perhaps less obvious are the more niche areas of research that are all too often ignored in the push for short-term technological wins. Here, HCS vLab provides an opportunity to enfranchise and strengthen more isolated research activities. For example:
- Thieberger (U Melbourne) has built a corpus of recordings and time-aligned transcripts of South Efate, a language from Vanuatu. This rich set of material could be accessed by others, but currently there is no platform to make it available. It requires streaming media linked to texts with an annotation module to allow researchers interested in prosody, narrative structure or musicology to access and annotate the material, safe in the knowledge that both the primary material and the new annotations will all have persistent locations.
- PARADISEC has many legacy audio speech and music recordings that are not annotated beyond a gross one-line description per audio-tape. In some cases it is not even clear what language is represented. Annotation would be greatly facilitated by crowd-sourcing; HCS vLab would provide an online space with an easy-to-use GUI where native language speakers and other researchers could access and annotate the recordings.
- There is a wealth of Aboriginal language dictionary and text material that once formed the Aboriginal Studies Electronic Data Archive (ASEDA) digital archive collected between the late 80s and early 2009 by the Australian Institute of Aboriginal and Torres Strait Islander Studies (AIATSIS). The establishment of the HCS vLab would make possible collaborative work with AIATSIS to make the ASEDA digital archive (also available as AUSTLANG) more visible.
- There is an increasing awareness that the temporal dimension of speech encodes a substantial amount of information about linguistic structure. Extracting this information requires experimental methods for acquiring high temporal resolution speech movement data, software for data analysis and computational modelling tools for linking the data to linguistic structure, but there is no accessible platform by which such analyses can be effected and no standards for analysis have been established, and many of the analytical tools remain inaccessible to the broader research community. HCS vLab will provide access to Shaw’s ParseEval, a tool that allows the syllable structure to be ‘read’ from speech data. Other tools, such as DemoLib will provide analysis of audio-visual speech.
Much has been written and said about the imperative to make research data widely available, especially when its creation is publicly funded; but the commitments signed up to in ARC proposals are more honoured in the breach than in the observance. This is in no small part due to the difficulty of the mechanics of making data available. HCS vLab will provide a platform for researchers in the human communication sciences that overcomes this problem, allowing increased leverage of existing investments and, in the process, making the data accessible to a wide range of existing tools in a streamlined fashion. For instance, with regard to the AusNC, HCS vLab will provide web-based tools for the display and analysis of annotations on linguistic material; natural language processing tools for text; automatic tools to support the generation and classification of annotations from audio and textual material; and tools for transcoding and streaming of audio and video in HMTL 5 ready format; and HCS vLab will make available the Ethnographic E-Research Online Presentation System www.eopas.org to present interlinear glossed text and media for other corpora. In addition, with the assistance of ongoing infrastructure support from UWS after the project is completed and HCS vLab is established, there are new tools that could be incorporated into HCS vLab, e.g., with the Australian Music Centre, an IR (information retrieval) tool for music and other acoustic data; and musical scores in XML form for musical academic research. Impact The Australian HCS community responded positively to the formation of HCSNet by attending the 60 HCSNet workshops and seminars, by an increase in successful large grants in the HCS area (see B.2) and by significant high impact journal publications (see the top 30 papers to come out of HCSNet in Dale, R., Burnham, D. & Stevens, C.J. (2011) Human Communication Science: A Compendium). Through HCSNet, the ARC has already invested in the development of a strong interdisciplinary community that has been widely recognized overseas; HCS vLab is an opportunity to both build upon, and reach far beyond HCSNet to provide that community with the tools and resources that it needs to leverage our distributed research capacity. The user base is ready and waiting. The impact of the HCS vLab on HCS research in Australia will be significant, far-reaching and sustainable. It will take HCS research capability to the next level, beyond the individual modalities of speech, language and music, and above that which can be accomplished on the ground in individual centres and institutes. HCSNet provided the opportunity for applying novel combinations of old ideas or methods of analysis from different disciplines to new problems. For instance, in the HCS Compendium, Butavicius and Lee (2011) describe a multi-disciplinary approach involving visual perception, human-computer interaction and cognitive modelling to the problem of assisting users to find relevant information in very large data sets; and Copland et al. (2011) married an exisiting psycholinguistic semantic priming task with fMRI to address the role of dopamine on neurotransmitters in semantic processing. However, the HCSNet experience also made it abundantly clear that one of the main impediments to such quantum leaps in HCS research was the difficulty for a researcher from one discipline to apply the tools and techniques of another discipline, or to explore data collected under one paradigm via a completely different analytical perspective. As HCSNet proved to be a model for inter-disciplinary collaboration, the HCS vLab will be an exemplar for other research communities. The impact and benefits of the HCS vLab activity will be tracked via the UWS Service Desk. The following measures will be used to report on utilisation and uptake by the research community: and measured through usage count (number of users logged in, number of tools used, number of queries) and user surveys distributed and collected at each of the 3 vLab phases (see B.13).
- Project / developmental measures:
- number of functional tests performed by researchers
- number of researchers who have participated in requirements gathering and testing
- Production measures:
- number of researchers with login accounts
- number of actual researcher login to VLab
- transaction counts: e.g. searches performed, invocation of tools, annotations generated and committed to the annotation store.
Future Developments Once the tools described in section B.7 have been incorporated, other well-known public access tools, such as ELAN (professional tool for the creation of complex annotations on video and audio resources); CLAN (Computerized Language ANalysis, for Conversational Analysis); PRAAT (scientific software program for the analysis of speech in phonetics); and The Field Linguist’s Toolbox (data management and analysis tool for field linguists), will be considered for addition to the HCS vLab. See Section B.18 for a list of projects which will be interoperable with the HCS vLab and for a list of additional corpora and tools which project partners already plan to add. Project partners have already indicated their intention to add the following:
- U. Sydney
- ExSite9 – a tool for efficient adding of metadata for digital research data as it is collected in the field.
- NABU – a catalog system for research collections with the ability to provide streaming access to media objects as well as providing a management environment.
These tools could work together with EOPAS: 1) ExSite9 for assembling and linking data prior to submission to repository; 2) NABU for managing the data repository and providing access to it where possible; 3) EOPAS for providing fine-grained online access to annotated media in the repository
- U. Canberra:
- UCBN – a broadcast news database with audio-visual video sequences (both text dependent and text independent subsets). There are around 20 speakers and total footage is around 6 hours of recording about 100 MB.
- MultimodalFusionLibrary – a software Library in C# for audio visual processing with several pre-processing, feature extraction, learning and classification algorithms mainly for stereoscopic and 3D video sequences.
- The AV face cover corpus through our MOU with the BBfor2 (Bayesian Biometrics for Forensics) Euro-funded project (BBfor2 is funded by the EC as a Marie-Curie ITN-project (FP7-PEOPLE-ITN-2008) under Grant Agreement number 238803.)
- The ABC corpus of infant-directed speech.
B.10 Broader adoption
The flexible HCS vLab framework coupled with the ongoing support by the UWS eResearch Unit will facilitate the inclusion of future corpora and tools. For example:
- A new (awarded 2012) Discovery (Best, Shaw, w/ PIs Hay, Foulkes, Docherty, Evans: ‘You came here TO DIE?!’ will collect audio recordings of a carefully-constructed corpus of words and phrases in 5 English regional accents (AusE, NZEng, Cockney, York and Newcastle-upon-Tyne). That corpus, plus earlier sets of Jamaican and American English materials collected in a current ARC Discovery project at UWS ‘How strict is the Mother Tongue?’ will be added as a database in HCS vLab. The pending project includes plans to run computational modelling on that corpus to determine relationships among pronunciations of words in the 5 accents, and to relate those modelled parameters to real-human word recognition across accents.
- A current ARC Discovery project at UWS (DP DP110105123, 2011-2015, The Seeds of Literacy) will generate a corpus of caretaker speech to 100 children at 3-monthly intervals from when they are 6-month-old infants to 5-year-old children and this will be added as a database in HCS vLab.
- A recently awarded (2012) ARC LIEF grant ‘A Living Archive of Australian Indigenous Languages’ through ANU is designed to digitise vernacular literature from NT Aboriginal schools, and the HCS vLab would provide a suitable home for these data.
- At ANU there are plans to augment a current fledgling corpus of Spanish with new data in order to conduct comparative research. Again the HCS vLab would provide a suitable home for these data.
In addition, a number of international labs across the world will welcome HCS vLab, for example:
- The New Zealand Institute for Language, Brain and Behaviour (NZILBB), a close partner of Marcs Institute, has both databases and tools they have developed for recordings of NZ English and NZ Maori and Maorian-English, which could be included as part of HCS vLab in the future.
- The University of Southern California is currently developing Electomagnetic Articulograph and real-time dynamic Magnetic Resonance Imaging speech databases and tools, and members of that project, Shri Narayanan and Mike Proctor are some of the many international collaborators of Australian HCS researchers.
- The recent NSF grant for ‘The Language Acquisition Grid: A Framework for rapid Adaptation and Reuse’ project led by Nancy Ide of Vassar College will provide a grid-based development environment for building natural language processing pipelines, facilitating application building and experimentation. As an international collaborator, the HCS vLab Product Owner, Steve Cassidy, will ensure that the Language Application Grid is compatible with HCS vLab infrastructure through the use of shared data standards and application interfaces.
B.11 Value adding
ANDS – Australian Research Data Commons The HCS vLab will be integrated with the Australian Research Data Commons. The project will manage collections of research data, and enable the selective export of descriptions of the data collections via OIA-PMH. AAF – Authentication The HCS vLab will provide authentication against the AAF through the SAML2 protocol. This will allow appropriately authorized individuals to authenticate using their institutional credentials. Where that is inadequate, dual authentication systems will be used, as outlined in the Federation’s technical documentation. Where possible, international collaborators will be authenticated in the system using the existing reciprocal arrangements between the AAF and its international partner organizations.
NCI – National Computational Infrastructure
The HCS vLab will integrate with NCI by submitting jobs to the HPC system. Authentication will be based on the credentials of the user logged in to the proposed system. SSH will be used as the protocol. Access to the HPC system assumes that the user of the proposed system has CPU time allocated to them on the HPC system, for example through the national Merit Allocation Scheme.
RDSI – Research Data Storage Infrastructure
The HCS vLab will host human communications datasets of national significance, as described in items 2 and 3 of this section. We plan to host these datasets on RDSI on the Intersect node. We anticipate strong alignment between this project and the RDSI ReDs programme. We are planning to host this project on the proposed Intersect Research Cloud node, which is to be co-located with the proposed Intersect RDSI node, and Intersect’s existing infrastructure. Research Cloud The HCS vLab will be built using the OpenStack Cloud API, to allow integration with the NeCTAR Research Cloud. Galaxy-GDR The HCS vLab acknowledges the Galaxy-GDR Integration NeCTAR project which will provide designs and insights that will be useful to this project, and vice versa. The Australian National Corpus The Australian National Corpus was funded by the Australian National Data Service (ANDS) to collect together a number of existing language corpora on a single technical platform that unified the many data and meta-data formats and provided a uniform set of interfaces to browsing and searching these data sets. The ANDS project developed an ingestion process that could be used with a wide variety of data types including text, audio and video collections, bringing them into a single standardised data store that is then exposed via a web interface. Another outcome of the project is a legal and ethical framework for sharing language resources within the research community. The Australian National Corpus platform will provide a starting point for the storage of data and meta-data in the HCS vLab. It will be extended to allow ingestion of new data types and to support federated storage of resources and meta-data. The Humanities Networked Infrastructure (HuNI) The Humanities Networked Infrastructure (HuNI) is one of the NeCTAR Virtual Labs being established. The HCS vLab and HuNI are complementary to each other and a compatible feed pipeline will be built between the two (see B.18).
HCS vLab Project Organisation
The authority structure for the HCS vLab project reflects the partnership model as described in Appendix E. Two key governance groups will be formed: a Steering Committee and a Stakeholder Group. A number of key project roles will interact with these groups, as described below. The Steering Committee is a small, senior group with overall responsibility for and authority over the project and its resources. The Steering Committee is charged with ensuring the project achieves its objectives, that stakeholders receive the benefits of the HCS vLab, and that these benefits are sustainable into the future. An important aspect is that the Steering Committee is responsible for trading off between the interests of the Stakeholder Group when necessary. The Committee consists of:
- Project Director – Professor Denis Burnham, UWS
- UWS eResearch Manager – Dr Peter Sefton, UWS
- Senior Partner Representatives – Representatives from the major partners in the project:
- Occasional members: other stakeholder representatives have been identified as having particular expertise across areas of HCS and will be called upon to for particular meetings as required.
- A representative from NeCTAR will be invited to attend as an observer.
The Steering Committee will meet monthly. The HCS vLab Project Manager (Dr Dominique Estival) and the Intersect Project Manager (Georgina Edwards) will be required to attend Steering Committee meetings to report progress and resolve issues. The Steering Committee will be provided expert technical advice through the Product Owner and the Intersect Project Manager. The Steering Committee is expected to become the Executive User Group once the project has completed. Responsibilities:
- The Project Director is the Chair of the Steering Committee, and is responsible for overall direction of the project and liaising with UWS executive.
- The HCS vLab Project Manager will be responsible for overall project planning, tracking and control, change management, stakeholder communication both within and external to the project, vLab marketing, user support, and acceptance testing. The Prince 2 project methodology will be used, leveraging off templates available through UWS Information Technology Services’ Project Management Office. These methods will be tailored somewhat to integrate with Intersect’s agile Scrum software development method. The vLab Project Manager will work with the Steering Committee to establish key success factors, metrics and project quality gates for assessing project performance.
- The Intersect Project Manager will be responsible for software development and co-ordinating all requirements analysis and software development activities associated with the project. Intersect will be responsible for producing the solution design, a working vLab application, and for integrating tools and corpus’ into the vLab. Intersect employs an agile approach to software development – see Appendix C for more information.
- The RMIT / NICTA Software Developer will be responsible for building software interfaces which enable the vLab to interoperate with other applications and vLabs as described in B.7 Infrastructure.
- The Product Owner will be the main point of interface between the software engineering team and members of the stakeholder working group, as per Intersect’s software methods which are described in Appendix C. The Product Owner (see B.6) will be responsible for the injection of domain-specific information into the framework, and will work closely with the vLab Project Manager to gain input from external stakeholders as and when required.
The Stakeholder Group comprises as wide a collection of representatives of end-users as possible to ensure the widest possible engagement. Importantly, this group will provide subject matter expertise in defining the functional requirements for the vLab, and will participate in user testing. The primary stakeholder group has already been identified: they are the HCS Tool Authors, HCS Corpus Caretakers and HCS academic experts from across the country, who will interact with the Intersect software engineering team through the Product Owner. They will, along with academic experts in particular areas of HCS and their doctoral students, test releases of the HCS vLab (see Section B.15 for a more detailed description). There are currently 47 academics (see Appendix A) who are committed to participating in the Stakeholder Group. Further stakeholders will be nominated during the elaboration phase, and will include members of the Steering Committee.
B.13 Project Scale, Key Deliverables and Acceptance Criteria
Project Scale Please see Part D3.2 for more detail. A summary is given below:
|Project Duration||12 elapsed months|
|Total contingent effort||190 effort months|
|FTE (during project)||14.3|
|Funding Requested from NeCTAR||$1.3M|
|Co-investment (during project)||$1.2M|
|Additional Co-investment (post-project) to 31/12/2014||$1.1M|
- The effort estimates have been built according to Intersect standard estimation approaches.
- A standard 20% contingency has been applied to the estimate.
- Project risk profile is Moderate. Overall mitigation strategy is to build the system Design To Cost
- FTE is a combination of project management personnel, product ownership, software developers, testers, researcher developers, and user representatives.
- $423,000 of the co-investment is in cash. This will be used to part-fund the Product Owner role, and for additional software development.
- Co-investment is from the following partners: UWS, Macquarie, Melbourne, Sydney, ANU, Flinders, RMIT, NICTA, UNSW, UWA, UNE, UC, LaTrobe, UTas, ASSTA, AusNC Inc.
Key Deliverables and Acceptance Criteria The 7 corpora, and the 11 HCS Tools developed by HCS members to be made available in the HCS vLab project (see also B.7) will be incorporated at different phases (N=3) of the project in an order determined by consideration of the joint factors of maturity of the corpus and the amount of work required to adapt the corpus for incorporation into the HCS vLab. Table 1: Key Deliverables and Acceptance Criteria
|D1||Start date – contract signed|
|D2||Problem Statement complete||Signed off by Steering Committee; submitted to NeCTAR|
|D3||Communications plan||Signed off by Steering Committee; submitted to NeCTAR|
|D4||Project plan; Quality plan; User stories;HCS vLab Architecture Staff hired Prototype released||Signed off by Steering Committee; submitted to NeCTAR Contracts signedTested by collaborators and HDRs|
|D5||VLab V1: Standard Execution Environment.Basic Workflow platform: User Registration, Corpus Browse, Tool Browser, Tool Execution Service. RDSI; Tool and Corpora Data. Usable with 2 Corpora (PARADISEC and AusTalk) and 4 sets of Tools (EOPAS PARADISEC tools, NLTK, EMU, Johnson-Charniak parsers).||Tested by collaborators and HDRs Browse corpora and read data Use tools on data Signed off by Steering Committee; submitted to NeCTAR|
|D6||Service Desk operational; Impact and benefits tracking in place; Reports: Impact and benefits; Project Status.||Tested by collaborators and HDRs Signed off by Steering Committee; submitted to NeCTAR|
|D7||VLab V2: Execution Environment: NeCTAR Research Cloud; Basic UIMA Data Bus. Workflow: Security and Access Control; Annotation Services; Federated Search UWS Research Data Catalogue; Annotation Store. Additional corpora (AusNC, AVOZES, AMC) and sets of tools (AusNC Tools, HTK, DeMoLib, PsySound3, INDRI). Reports: Impact and benefits; Project Status||Tested by collaborators and HDRs Perform federated / cross corpora search Create annotations Signed off by Steering Committee; submitted to NeCTAR|
|D8||VLab V3: HPC Execution Environment.Workflow: Corpus Management Service, Tool Management Service, Workflow Capture. Record-level Metadata Store; Access to HuNI Vlab Data and Tools Additional tools (ParseEval, NuancesWithMidi, ParGram), corpora (Jakarta Indonesian, Forensic). Reports: Impact and benefits; Project Status||Tested by collaborators and HRDs Access HuNI Data and Tools Create and reuse workflows Signed off by Steering Committee; submitted to NeCTAR|
|D9||Post-implementation Review (PIR)||PIR conducted; report signed off by Steering Committee and submitted to NeCTAR;Practical Completion Certificate accepted by NeCTAR.|
|D10||Service Levels met and reported to NeCTAR as defined.|
B.14 Staged Deployment
|Milestone No.||Name of Service/Deliverable||Date of deployment for pilot use||Date of deployment as production service|
|M1||Sub-contract signed, project started||N/A||01/12/2012|
|M2||Elaboration Phase Complete Prototype released||01/02/2013||15/03/2013|
|M3||HCS vLab Version 1 Operational||15/03/2013||01/06/13|
|M4||HCS vLab Version 2 Operational||N/A||1/09/2013|
|M5||HCS vLab Version 3 Operational||N/A||1/12/2013|
|M6||Final Admin Closure||N/A||31/12/2013|
|M7||Application Support and Maintenance, User Group to Dec 2014||31/09/2014||31/12/2014|
|M8||Application Support and Maintenance, User Group to Dec 2015||31/09/2015||31/12/2015|
B.15 Project Approach
A detailed description of Intersect’s project approach is in the attached document “Intersect Software Development Process” (Appendix C) and this is a summary of that document. Intersect’s approach to running the project rests on two principles: 1) Ongoing communication with the governance group, reporting to them and soliciting input; and 2) Continual integration of a tested and deployed product. A project goes through four stages: Concept; Elaboration; Development and Deployment. These four stages will be spread across the 14 months of the HCS vLab project as set out below. The Stakeholders Group and the Steering Committee will be involved in all four stages of the project, with the Project Manager and the Product Owner ensuring communication between the Steering Committee, the Stakeholder Group and the Intersect development team and overseeing the integration and testing of the HCS Corpora and Tools. HCS end-user testers, drawn from the Stakeholder Group, will be responsible for testing the software after functionality has been developed. More specifically, for the HCS vLab, the Stakeholder Group will comprise: HCS Tool Authors: The author or developer of each of the 11 HCS tools to be integrated in the HCS vLab will be responsible for adapting their particular tool to the HCS vLab environment, with the assistance of the Product Owner interacting with the Developer, Intersect. HCS Corpus Caretakers: For each of the 7 corpora to be incorporated, at least one HCS Corpus Caretaker (for instance AusTalk and PARADISEC have 2 and 3 developers respectively) will be responsible for adapting their particular corpus to the HCS vLab environment, with the assistance of the Product Owner interacting with the Developer, Intersect. HCS Sprint Testers: The HCS Tool Authors and HCS Corpus Caretakers will be joined by another 26 academic experts in particular HCS disciplines to provide in-kind co-investment of 3 occasions of 2 days of testing the HCS vLab functionalities during the sprints in the development phase – see 3 below. These people will be deployed when tools and/or corpora of interest in their particular area have been incorporated (see Appendix B for a list of the tools and corpora). A total of 15 Higher Degree Research (HDR) students from the 15 co-investing universities and research institutions (see Appendix A) will be assigned to test the functionalities of the HCS vLab during the sprints in the development phase – see 3 below. To provide continuity, each HDR student will act as tester for the three consecutive sprints in a particular development phase (see Figure 1 below) and then as a tester in the first sprint of the next development stage.
1. Concept Stage
During the concept stage, the development team and the stakeholders come to an agreement on a concise definition of the key problem to be solved. This statement identifies the nature of the problem, the group whom the problem impacts, the impact, and the properties of a solution. The outcome of this stage is a problem statement. The nature of the NeCTAR RfP means that proposed projects are well into the concept stage by the time they are proposed. The key outcome of the concept stage is an agreement on the problem statement.
2. Elaboration Stage
The elaboration stage is used to bootstrap the development stage. During this stage, the artefacts uses to monitor and steer the project are created, including the initial user stories, a product backlog and a burn-up chart. In addition, the team evaluates and settles on the key technology choices for the project. The elaboration phase concludes when the team can articulate the key technical risks and approaches to removing those risks; and when the team can identify the key roles on the project (especially key stakeholders) and key constraints of the project (e.g. dates). Key to the agile process is that these decisions may change at a later stage. By the end of this stage, a project management plan and a quality management plan have been developed.
3. Development Stage
The development phase consists of two-week sprints where the team works to complete the stories in that sprint. Usually planning, execution and review overlap, as shown below. The defining characteristic of the development stage is that at the end of each sprint, we have a potentially shippable increment of the product. If the customer wishes, we can deploy this to production systems for use. Figure 3: Illustration of when planning, executing and reviewing of sprints occurs Prior to the start of a sprint, the sprint is ‘planned’. A set of stories from the backlog is selected and elaborated. This elaboration includes defining detailed acceptance criteria: these criteria define when the story is ‘done’ (i.e. acceptable to the end-user and integrated into the product). The selected stories are the ones that will be implemented during the sprint. During the sprint, the sprint is ‘executed’. Selected stories are developed, tested and accepted by the stakeholders. At the end of each sprint the sprint is ‘reviewed’ and we have a demonstration of the functionality developed during the sprint. Progress is then evaluated in terms of stories implemented, stories remaining and budget spent. Subsequent to the completion of a sprint’s review, the user-testing team is responsible for testing the product against acceptance tests, and reporting defects found. During subsequent sprints, reported defects are assessed and planned along with other user stories. The contribution of the partners to the development of the HCS vLab is as follows:
- 15 days of effort to help Intersect integrate each tool or corpus contributed into the HCS vLab.
- 30 days of effort involved in vLab requirements gathering and user testing.
- 3 days of effort involved in project governance.
- 30 days of effort involved in user group meetings and post-project user support activities (Jan 2014 – Dec 2015).
4. Deployment Stage At the end of the last sprint in the Development phase, we have a fully functioning, integrated and tested piece of software. Unlike traditional waterfall development, there is no ‘final testing’ or ‘acceptance testing’ phase – the software has been tested by its users throughout the project. During the Deployment phase the system is deployed to production in accordance with the hosting arrangements.
B.16 Quality Control
Quality Control and Acceptance testing are integrated in Intersect’s development approach, and are ongoing from the start of the project. This mitigates the risks of “big-bang” integration and acceptance testing. Part of the quality control is integrated with other processes, including test-driven development and writing acceptance tests before implementing a user story. Additional testing (e.g. non-functional requirements, risk-mitigation testing) is performed based on the quality management plan (see above). For each user story implemented during development, quality assurance is managed by interaction between the Product Owner (as a representative of the governance group), the developer responsible for that story, and the senior test-engineer on the project. During planning for the sprint, the Product Owner is responsible for defining the acceptance criteria for the story, with support from a test engineer. During the execution of the sprint, the developer responsible for the implementation of the story defines automated unit tests, to guard against regressions. Prior to the completion of the sprint, the test engineer is responsible for signoff of the user story against the acceptance criteria. Where possible, the acceptance tests are automated, using frameworks such as Cucumber. During the review of the sprint, the Product Owner (representing the governance group) validates that the stories implemented during have the correct functionality. Subsequent to the demonstration, the end-user testers are responsible for end-user testing of the system as implemented to date, as described above. A deliverable is comprised of multiple user stories. Formally, the Product Owner, acting as a proxy for the governance group, is responsible for the acceptance of a deliverable. This responsibility comprises: ensuring that a set of user stories covering the deliverable are defined, elaborated, tested and reviewed.
Wherever possible, commissioning testing is combined with Acceptance testing. That is, we deploy the system early and often, and acceptance tests are run against a production system. This will be the NeCTAR Research Cloud, as available, including the lead node at the University of Melbourne during development.
B.17 Risk and Issue Management
|This project has an external dependency on the development of the NeCTAR research cloud. Delays in access to the cloud increase the risk of difficulty in deployment and commissioning, c.f. Commissioning Testing, aboveMitigation: The project has the ability to use Intersect storage and HPC in the event of NeCTAR delays.||Moderate||Moderate|
|This project has an external dependency on the development of the RDSI project. Delays in access to the storage on RDSI increase the risk of difficulty in deployment and commissioning.Mitigation: The project has the ability to use Intersect storage and HPC in the event of NeCTAR delays.||Moderate||Moderate|
|This project depends on further development of open-source software. There is a risk that the governance of that software will not incorporate our changes, resulting in fractured support for that software.Mitigation: The project members will engage with the open-source software developers to ensure acceptance of our changes.||Low||Low|
|This project depends on further development of open-source software. The functionality of that project has been assessed and is appropriate. There is a risk that the code base of the project is more difficult than anticipated to develop in support of our intended functionality.Mitigation: In most cases, there will be close collaboration between project members and the original developers of the software.||High||Moderate|
|This project proposes a distributed development effort. There is a risk that the communications overhead of this effort is greater than expected, impacting on the speed of development. We estimate the likelihood of this risk as low.Mitigation: In addition to regular video-conferences throughout the project, the proposal makes provisions for a face-to-face meeting of all project members in the Initial Phase and 4 visits by the Project Manager and Product Owner to the 4 main centres (Sydney, Melbourne, Canberra and Perth).||Low||Low|
Other possible risks include:
- Ability to manage and arbitrate data access, including privacy concerns
- Impact on a lack of standards, or tools being integrated using non-standard formats
- IP position for tools being integrated
- Competing research project requirements, making it difficult to arrive at a consolidated position during the execution of the project
Approach to Risk Management
The HCS vLab project will maintain a risks register, including the impact, likelihood and mitigation strategy for each risk, as it is identified. The risks register will be a part of the regular reporting to both the governance body and NeCTAR. For technical risks, risk level is accounted for in prioritising user stories – more risky stories are, other things being equal, attempted before less risky stories. For staffing-related risks, use of Intersect as a development partner with significant capacity ameliorates the risk of departure of key staff. For external/dependency risks, responsibility for identification and management of the risk profile of the project rests with the Project Manager.
B.18 Standardisation and Interoperability
There is a significant move towards interoperability of language resources and tools in the international arena as national and international groups collaborate more frequently on large scale projects. There are three primary areas of focus for standardisation that are relevant to this project: meta-data, annotation standards and tool interoperability. Meta-data standardisation has long been a focus of the language resources community with well established standards such as OLAC and IMDI used to describe resources in various meta-data repositories. More recent efforts have established the ISOcat data category registry that is used to register vocabularies used at various levels in language resources. A significant EU effort just getting underway is META-NET (Multilingual Europe Technology Alliance) which aims to establish a platform for resource sharing around Europe and has already made significant contributions relating to meta-data management. The current ANDS funded Australian National Corpus project is developing a hybrid meta-data model suitable for describing language corpora which can be exported to the ANDS Research Data Collection via the RIF-CS standard vocabulary. The HCS vLab will make ARDC-compliant use of RIF-CS at the collection level, through automatic export of selected data set descriptions to ARDC as well as by allowing import of data that others have ‘advertised’ on the ARDC. The standardisation of annotation formats and the semantics of various annotation schemes has been a focus of the ISO TC 37 (Terminology and Other Language and Content Resources) working groups. This group has established, for example, standards for Morphosyntactic annotation (MAF), a Lexical Markup Framework for use in dictionary-like resources (LMF) and an interchange format for annotation (LAF/GrAF). These formats and standards are now gradually being adopted by projects around the world and will provide for increased inter-operability between language resources. Project member Steve Cassidy is an active member of a number of ISO TC 37 working groups and was recently invited to help establish a working group to standardise query languages for language resources. Internationally: A number of projects in the EU and US aim to develop standard web-service architectures for defining and managing work-flows that process audio or textual resources using tools such as parsers, taggers, speech recognisers etc. One important project that we are associated with is the US NSF funded project “The Language Application Grid: A Framework for Rapid Adaptation and Reuse” managed by Nancy Ide and James Pustejovsky which names the Australian National Corpus as one of a number of international collaborators. The HCS vLab will work closely with these international partners to ensure that we are building compatible and interoperable toolsets. In Australia:
- NICTA will contribute to provide support for interoperability with UIMA, the Unstructured Information Management Architecture, an emerging standard for wrapping components for processing language, speech, video, and other unstructured data. Dr. Verspoor (NICTA) was a member of the OASIS standard committee for UIMA and is an editor of the published standard.
- The Humanities Networked Infrastructure (HuNI) is a NeCTAR Virtual Lab established in the first round of funding. HuNI is primarily concerned with developing sustainable, relevant and enabling infrastructure for Australian humanities researchers and cultural custodians and it involves researchers and custodians working with cultural datasets. There is some overlap in the kinds of data being managed by each network and we believe that there is significant scope for HuNI to benefit from some of the workflows that will be developed within HCS vLab. We will work closely with HuNI to ensure that data can be exchanged where appropriate to maximise the value of this infrastructure to the audiences of both virtual laboratories.
- La Trobe University is making available the Humanities and Social Sciences Visualisation Laboratory (VisLab). VisLab allows for the remote use of scientific instruments and imaging of scientific data, creating a capability for interactive research collaboration, visualisation and imaging. For instance, the city of Melbourne is a research hub in the phonetics of endangered languages, yet the largest speech research facilities are at the University of Western Sydney and Macquarie University in Sydney. Using the VisLab, phonetics staff members and PhD students in Melbourne would have access to expensive equipment such as the 3D Carstens EMA machine at UWS (worth approximately $100K), without having to travel to Sydney for the fiddly and time-consuming trial-and-error stage of data acquisition. Such virtual access will save resources in the long term, since errors are less likely in the data collection stage if there has been sufficient time for testing.
PLEASE COMPLETE DETAILS IN ATTACHMENT D3 MILESTONE AND FUNDING TEMPLATE
SERVICES AND SUPPORT
B.19 Service Levels
The service levels are set out in Table 4. Please refer to B.20 for further details of these services. Table 4: Service level that will be offered for each service
|Application Hosting||24×7 365 days per year, with 99.9% availability|
|Business Services||For automated workflows: as for Application Hosting, aboveFor manual workflows: initial response next business day|
|Virtual Service Desk||Initial response: next business day|
|Application Support||Initial response: next business day|
B.20 Operations and User Support
With reference to the services in Table 4, the following operations and user support will be provided:
Intersect will provide the application hosting support service for the HCS vLab application. Each service hosted by Intersect has an individual who is ultimately responsible for the provision of that service – the service owner. Services hosted by Intersect are monitored, and backed up, and user support is provided during business hours. All services are automatically monitored for availability using a ‘shallow monitoring’ approach (“Is the service alive?”). Where appropriate, services are monitored using a ‘deep monitoring’ approach (“Is the service responding sensibly?”). Unexpected outages are:
- published via a mailing list – the makeup of the mailing list being maintained by the service owner;
- escalated to the systems administration team.
Scheduled outages, including for upgrades of software and hardware, are requested by the systems administration team. These outages are publicised and negotiated with the user base of the service by the service owner.
The following business services have been identified:
- User Registration, Research Data Annotation Service, Tool Execution Service, Federated Search, Corpora Management and Commissioning Service, Tool Management and Commissioning Service, Workflow Capture Service, ANDS Publication Service.
- UWS Marcs will be responsible for servicing these requests. The project is aiming to implement a self-service model. However, it is likely that not all services will be fully automated; some services may require human co-ordination. The requests will originate from application workflows. Service levels are as specified in B.19.
Virtual Service Desk
UWS Marcs and Intersect will jointly maintain a virtual service desk for users. Intersect will provide a single point of access for end-users of the service to lodge and track issues and other service requests. UWS Marcs will provide tier-1 (first response and triage) and tier-2 (business workflow and problem resolution) support to users. UWS and Intersect will provide tier-3 (software defect resolution) support for the vLab framework proper, whereas the tools themselves will be supported by contributing project partners. Jira will be used to track all issues and service requests. UWS is committed to support the service desk until 2014.
The application support service (e.g., account creation and access control, troubleshooting) will be carried out by UWS Marcs tier-1 support. In addition to Virtual Service Desk, UWS Marcs and Intersect will provide learning and development (L&D) and outreach programs. Support for the project during its operation will include:
- Development and delivery of training modules for the use of the outcomes of the project through UWS’ and Intersect’s learning and development activities,
- Publicity and on-going assistance with the outcomes of the project, through Intersect’s outreach activities. This work is distributed amongst the team of 9 research analysts.
The HCS vLab will be operationalised throughout the course of the project and is expected to attract a broad and active user group both within the research communities, and by external users. After the project is completed, UWS Marcs, with its partners, will take responsibility for supporting and hosting the HCS vLab. As researcher requirements are expected to be diverse with new corpora and new tools becoming available, there will be strong drivers in place for further development and improvement to occur. UWS Marcs will:
- Explore opportunities to expand the HCS vLab service desk.
- Negotiate with its partners to secure additional funding for software development ($150,000 has already been allocated).
- Engage and influence UWS to provide further operational and application support for the HCS vLab.
- Explore opportunities to develop a transactional cost model for external parties to gain access to the HCS vLab.
- Explore opportunities to develop a data mining and analysis service for external parties.
- Apply for further infrastructure grants to expand the HCS vLab further.
- Facilitate the development of training courses for adding tools and corpora to the HCS vLab – on-line and face to face.
- Commercialise the system and make it available to new and emerging research centres and external commercial enterprises for a fee. UWS would commit to reinvesting funds back into the system and establish a product user group.
- Hold conferences and forums to further the concept of the HCS vLab. Resulting funds reinvested.
There are already very strong links between the HCS vLab partners and the research community (see B.2) and it is expected that the infrastructure built will encourage the integration of future databases and analysis tools. Several partners are already planning to add corpora and tools to the HCS vLab (see B.9 and B.10). In addition, A/Prof. Drew Khlentzos, from the Language and Cognition Research Centre at UNE, has proposed the integration of a new logic database comprising core logical principles governing the main logical operators that are expressed in most (if not all) of the world’s languages, which should be available for incorporation into the HCS vLab within the next 2 years.
B.22 IP, Licensing and Access
Intersect currently makes software available under an open source licence. Intersect’s IP in NeCTAR projects will be made available to Australian publicly funded researchers under the same conditions. All the corpora and tools in the HCS vLab will be available to researchers and users under appropriate licensing, according to guidelines published from the Australian National Data Service.
B.23 Communications and Engagement
There are three periods during which customer satisfaction is solicited by Intersect.
- During the concept and elaboration stages (see B.15) the governance and stakeholder groups represent the customer. They are engaged frequently and in depth through this phase by the developer, Intersect. Project plans and quality management plans are reviewed by the group, as are the initial list of user stories that define the initial direction of the project. There is no separate process by which satisfaction is measured – it emerges from the collaborative nature of the project.
- During the development and deployment stages (see B.15), again the governance and stakeholder groups represent the customer. The stakeholder group is engaged via sprint-end demonstrations and end-user testing, and can affect the course of the project during sprint planning. Rather than attempt to measure satisfaction, Intersect will engage continuously, and allow the customer regular opportunities to effect change in the project’s direction. The Project Manager reports to the Steering Committee monthly. The governance group is invited to provide input at that point, including their level of satisfaction.
- After deployment of every project carried out by Intersect, the Development Team, Product Owner and Project Manager conduct a ‘lessons learned’ forum. Representatives from the Steering Committee can attend and provide input. At this forum, the outputs of the project and the process by which they were derived are critically analysed.
After 6 months, the governance group are consulted for their satisfaction with the process and outputs of the project. Intersect uses a standard form to solicit the group’s input. Formal channels, as outlined above, go some way to measuring customer satisfaction. Intersect also makes active use of a variety of informal channels to solicit feedback, including consulting DVCRs or PVCRs and CIOs of universities making use of the service, attending university eResearch committee meetings, and through the outreach conducted by eResearch analysts. These informal channels provide an important complement to the formal ones. Successful engagement under Intersect’s development model requires that the stakeholder group is broadly representative of the eventual end-user community. Further, the model requires a significant investment of time from members of the stakeholder group over the period of the project.
B.24 Constraints and Dependencies
External Party Capability Required Date first required Milestones or Deliverables dependent on that capability ANDS Seeding The Commons Project SC20 Research Data Catalogue 01/02/2013 M3 RDSI Node Data Storage 01/02/2013 M3 Research Cloud Computing Environment 01/01/2013 M4 AAF Interoperability 01/01/2013 M4 Griffith and AusNC Inc. Expertise and Support for integrating the Aus NC corpora and tool sets 01/05/2012 M4 HuNI Interoperability 01/08/2013 M5