CSIRO HAIL Seminars

Conversational Agents for Intelligent Environments

Claude Sammut — Wed, 15 july 2009 11:48 +1000

For several years, we have been developing scripting languages for building conversational agents. Over time, these languages have evolved to include middleware that allows the conversational agent to interface with devices embedded in the environment, for example, audio- visual equipment, lights, cameras, microphones and other computers, including mobile ones, i.e. robots. In this talk , we will discuss some of the issues in designing this set of software. This includes questions of appropriate language constructs to support conversational context, representation and reasoning with background knowledge and requirements for interacting with external devices. We will also discuss problems of grounding language for use in robots.

Do we simply accept the Australian Internet filtering scheme or do we need to ask hard questions?

Bjorn Landfeldt — Wed, 17 june 2009 10:48 +1000

The Australian government is currently planning to introduce mandatory Internet content filtering. The move would clearly set Australia aside from all other democratic nations in the world. The initial purpose of this scheme was to protect Australian children from accessing unsuitable material such as child pornography. Even though there is widespread consensus in society that such material is undesirable and potentially harmful, the issue of filtering is extremely complicated and it is far from evident that the proposed scheme will achieve its goal. In addition, over the past few months evidence has been presented indicating that the side effects of such filtering could have severe negative impact on society. Associate Professor Bjorn Landfeldt was part of a team that studied the feasibility of implementation of Internet filtering at the ISPs. The report was commissioned by the Howard government and handed to the current minister in charge of this issue, Senator Stephen Conroy. In this talk, Professor Landfeldt will detail some of the major difficulties associated with ISP level content filtering, some of the possible side effects and discuss why such filtering may not be effective. He will also give examples of the many difficult moral questions such filtering inevitably raises and demonstrate the need for a comprehensive public debate an the issue before legislation and implementation takes place.

The future Australian National Corpus and the Web

Michael Haugh — Tue, 09 june 2009 17:15 +1000

2008 saw the launching of the “Australian National Corpus Initiative”. Linguists and language technologists from around Australia committed themselves to a sustained push for the creation of a major addition to Australia’s research infrastructure in the humanities and social sciences, namely, a massive online database of spoken and written language in Australia, in all its forms and diversity and over time (Burridge, Haugh, Mulder and Peters forthcoming). One potential challenge to such an endeavour arises from the fact that a massive online source of language data already exists, namely the World Wide Web. In this talk I first consider why Web as Corpus cannot be seen as an adequate substitute for a national corpus. I then discuss the relative merits of Web as Corpus versus the creation of Specialised Web Corpora, and the relative potential of each to contribute to the future Australian National Corpus. I next outline two case studies investigating the relative utility of Web as Corpus versus Specialised Web Corpora in studies of specific pragmatic phenomena, namely im/politeness in emails between academics and students (Haugh forthcoming a), and “taking the piss” in everyday interaction amongst Australians (Haugh forthcoming b). I conclude by briefly discussing the implications of the various types of data considered in this talk for the ways in which such data is ingested and searched in the future Australian National Corpus.

Reducing semantic drift in biomedical lexicon bootstrapping

Tara McIntosh — Tue, 02 june 2009 15:19 +1000

Extracting biomedical semantic lexicons from raw text is critical for overcoming the knowledge bottleneck in many bio-NLP tasks. In this talk, I will present the Weighted Mutual Exclusion Bootstrapping (WMEB) algorithm for simultaneously extracting precise biomedical semantic lexicons and patterns for multiple categories. WMEB is capable of extracting larger lexicons with higher precision than previous techniques, successfully reducing semantic drift by incorporating new weighting functions and a cumulative pattern pool. Unfortunately, semantic drift still dominates in later iterations, as erroneous terms eventually shift a category's direction. We present two novel approaches for reducing semantic drift further in WMEB - unsupervised bagging, and utilising distributional similarity to detect and censor potential semantic drifts.

Empirical studies of speech and linguistic futures for cognitive load measurement

Dr Fang Chen — Tue, 05 may 2009 16:04 +1000

Cognitive load refers to the capacity and resources used in working memory while learning – completing tasks where novel information and novel processing are required. In complex and time-critical situations, system users can experience extremely high cognitive load, which can interfere with task completion. An understanding of the users’ cognitive load will enable us to alleviate these problems by implementing strategies to adjust the system’s behavior, support, and resources needed as per their cognitive burden and help them complete the task effectively. Moreover, for complex collaborative tasks where many users have to cooperate to solve task-related problems, understanding cognitive demands can be very helpful. However, cognitive load is difficult to measure, particularly across individual users, and there is clearly room for innovation in this area. Certain speech features have been shown to change under high levels of load and are good candidates for cognitive load indices for usability evaluation and automatic adaptation of an interface or work environment. Several speech-based user study and analysis will be presented in which we explore the behaviour of speech features in natural speech. Our other study presents a speech content analysis approach to the measurement of cognitive load which employs users’ linguistic and grammatical features of speech to determine their experienced level of cognitive load. We present the analyses of several linguistic features extracted from the speech data collected involved in highly time-critical and data-intense bushfire management tasks around Australia. Some preliminary results of relationship between cognitive load and linguistic features in collaborative work will also be presented.

Blended Interaction Spaces for Distributed Team Collaboration

Kenton O'Hara — Tue, 21 april 2009 14:35 +1000

The talk will discuss work recently conducted in the HxI intiative aimed at developing new workspace environments to support complex collaboration between multiple teams distributed across multiple geographic locations. The work draws inspiration from recent high-end video conferencing systems such as HP Halo, Cisco Telepresence and Polycom TPX that have demonstrated enhanced user experience traditional remote conferencing systems. These systems combine camera positioning, display arrangement and table design to create geometrically correct representations of the remote sites that provide near correct eye gaze and spatial representation of non-verbal gestures. In essence, they provide the visual sensation of remote parties sitting around the same conference table whereby the different sites become blended. Such systems work well for simple communication tasks across distance but provide little support for the sharing, visualisation and interaction with multiple information sources that are an essential part of intense collaborative work. Drawing on the philosophy of these systems, the talk will discuss research around the development of an ecology different "blended space" environments that provide the experience of sitting and working round the same table for more complex collaboration tasks. As well as appropriately combining camera views, display arrangements and furniture, we will introduce interactive workspaces into the environment in the appropriate ways that maintain eye gaze, spatial representation of non-verbal gestures and geometrically coherent reference to a shared interactive workspace. The use of different interaction mechanisms in these spaces such as horizontal multitouch surfaces will be discussed.

Cloud computing - Mapping data-intensive problems onto warehouse-sized data centres

Dr J. Craig Mudge — Thu, 9 april 2009 15:29 +1000

Every time you do a Google search, you use a form of cloud computing, since you are using the Internet to access storage and computers located in massive warehouse-sized data centres. More generally, cloud computing allows an organisation to outsource the management and location of its IT equipment to such facilities. Because they use commodity PCs and disks, shared power and cooling, and 100% automation of operations., overall costs are a factor-of-five less. The Academy of Technological Sciences and Engineering has formed a Working Group on “Cloud computing at peta-scale” to construct a roadmap for Australia to guide effective use of this emerging disruptive technology in business, government, and science. Its membership of fifteen is drawn from Fellows and others in universities, CSIRO, and NICTA, with strong engagement from Google, IBM, and Microsoft Research in the U.S. This seminar will characterize cloud computing, raise some geo-political issues in the use of clouds located abroad, and provide a status report on some of the projects in the Working Group.

Using Linguistically Motivated Features in Document Retrieval for Question Answering

Luiz Augusto Pizzato — Wed, 11 march 2009 14:33 +1000

In this talk I will present the work conducted for my thesis where I investigated the impact of using linguistic features in the Information Retrieval (IR) stage of Question Answering (QA) systems. We hypothesise that techniques that are commonly used in the final answer extraction stage can improve the overall results of a QA system when adopted in the earlier IR stage. In particular, we study the use of the following information: i) named entities in a pseudo-relevance feedback process; and ii) semantic relations between words of questions and text sentences. The study of the use of named entities is inspired by the common practice of filtering out sentences that do not contain the expected answer type. We consequently introduce a pseudo-relevance feedback that inserts entities of the correct answer type in the original query. Our experiments show that this technique leads to a query drift and the final results do not improve with respect to a query without additional feedback. To study the use of relational information, we design an IR framework that is more efficient (in both speed and memory consumption) than standard approaches based on relational databases and on the concatenation of word pairs at the indexing stage. The resulting framework allows a multi-layer index that uses an extension to the standard vector space model as a ranking strategy. The resulting ranking strategy improves precision, without compromising the overall recall, by including linguistic word relations. We present the Question Prediction Language Model (QPLM), a model of relational information that borrows concepts from Semantic Role Labelling (SRL) but is designed for the fast generation of annotation and its use for indexing and retrieval. The results are of quality comparable to SRL and indicate that linguistic information encoded in the form of semantic relations does enhance the retrieval quality of text and the final accuracy of QA systems.

Cues as Mediators of Human Performance: Implications for Training and System Design

Mark Wiggins — Thu, 5 november 2008 15:59 +1000

Despite the fact that cues are often referred to in explanations of expert behaviour, they have been difficult to identify and validate in complex operational environments. This presentation explores some of strategies that we have developed to identify cues, and validate the use of cues amongst experts. We also examine applications of cue-based strategies in training, checklist development, and as a basis for system design. Finally, we explore individual differences in the acquisition of cue-based relationships and the implications for selection and the operation of complex, advanced technology systems in the future.

The Future of Email

Gabor Cselle — Wed, 15 October 2008 15:59 +1000

Email clients haven't moved forward since the mid-1990s. Most applications have added superficial features, but the basics remained unchanged: Folders, lists of disconnected emails sorted by arrival time. Clients have no sense of priority, urgency, workflows, or connectedness. Their search features are simple and are sometimes painfully slow. Users today are bombarded with email and find popular email clients hard to use and inefficient. How did we get here? How do we get out of it? This talk will show new ideas of improving the email experience for overloaded users. Gabor will also talk about commercial opportunities in this field, illustrated with his own experience in Silicon Valley at his previous startup, Xobni, the maker of a popular Outlook-based application.

Hybrid Multi-Step Disfluency Detection

Sebastian Germesin — Wed, 1 October 2008 15:09 +1000

Previous research has shown that speech disfluencies - speech errors that occur in spoken language - affect NLP systems and hence need to be repaired or at least marked. The talk presents our experiences with a hybrid approach that uses different detection techniques for this task where each of these techniques is specialized within its own disfluency domain. A thorough investigation of the used disfluency scheme, led us to a detection design where basic rule-matching techniques are combined with machine learning approaches. The aim was both to reduce computational overhead and processing time and also to increase the detection performance. In fact, our system works with an accuracy of 92.9% and an F-Score of 90.6% while working faster than real-time.

Relevance assessment: are judges exchangeable and does it matter?

Paul Thomas — Tue, 9 September 2008 14:25 +1000

Reusable test collections for information retrieval rely on relevance judgements, decisions on which documents are good answers to each query in the test. We investigate to what extent people making relevance judgements are interchangeable. Analysis shows low levels of agreement between judges, and we report on experiments to determine if this is sufficient to invalidate the use of a test collection. We find that both system scores and system rankings are subject to consistent but small differences across different relevance judges. It appears that collections are not completely robust to changes of judge when these judges vary widely in task and topic expertise. This has implications for anyone building a reusable test collection. (This is joint work with colleagues at NIST, Microsoft, and CWI.)

Integral Methodological Pluralism

Tim Mansfield — Thu, 27 August 2008 14:16 +1000

Many of us in HCI and Interaction Design work in multi-disciplinary teams with other professionals and are presented with methods, practices and data from fields with which we are not familiar. While a detailed understanding of the methods of other disciplines is something only gained from extensive collaborative work, I have wondered if it's possible to answer fairly simple questions about whether a situation is being adequately investigated, are we using the right mix of methods, are we getting enough different kinds of data or are we getting lots of different kinds of data from a very similar perspective. Ken Wilber is a syncretic, integrative philosopher from the United States who claims that it is possible to answer these questions. His work is well-known in ecology and consciousness studies and in some psychology departments, but less known to social scientists. One of his most interesting contributions in the last decade is the notion of Integral Methodological Pluralism, which seeks to clarify meaningfully distinct knowledge realms in human experience and to characterise the kinds of approaches to gaining knowledge in each of these realms.

Authoring Interactive Drama

Nicolas Szilas — Thu, 12 August 2008 14:53 +1000

Interactive Drama is often considered as the Graal of the new media. It aims at creating a new dramatic art/entertainment form where the user acts as a main character in a story. S/he is able to perform a large range of actions in the story, interact with other characters in order to influence the story. In this talk, we will review the various existing approaches to reach that challenging goal. A prototype developed by the author will be demonstrated. We will discuss authoring, a specific issue emerging in the field. Very few works in Interactive Drama have been produced so far, because prototypes are technically quite difficult to handle by authors. What is the point of developing sophisticated technology if it cannot be used by the creative authors? Is it just a temporary problem of a lack authoring tools, or shouldn't we design our systems differently, by thinking in authoring terms first? Finally, this specific debate around Interactive Drama might also be relevant in Artificial Intelligence (AI) in general. The future of AI might be neither designer-centred nor user-centred but author-centred.

Evaluating Information Retrieval in Context

Paul Thomas — Thu, 08 August 2008 15:03 +1000

Established evaluation methodologies for information retrieval are not well suited to the task of comparing systems in many real settings. We have implemented a comparison tool which support this task. By offering a working search interface the tool permits study of real information needs as they occur, uses the documents actually available at the time of the search, and records judgements taking into account the instantaneous needs of the searcher. We report experiments, to validate the tool and explore potential biases, and ongoing and future applications.

Logic, Automated Reasoning and Applications

Ulrich Furbach — Thu, 06 August 2008 15:08 +1000

In this talk the state of the art in Automated Reasoning is shortly depicted. The focus is mainly on 1st order predicate logic, but there will also be a discussion of knowledge representation and model checking issues. Besides the development of calculi for automated reasoning the research of the AI Group in Koblenz also put emphasis on applications. Some examples ranging from academic prototypes to the development of products are given. This includes natural language processing, personalized information systems and RoboCup.

THE OMNIUM PROJECT: Investigating the potential of Online Creative Collaboration

Rick Bennett — Fri, 11 April 2008 13:24 +1000

Omnium’s ongoing research focuses on exploring the notion of online collaborative creativity (OCC) and how the Internet can be best used to help geographically distanced?individuals interact and work creatively together from any location worldwide. Omnium has since formed a series of fully-online creative communities, facilitated major global and fully-online collaborative projects as well as designed and written some ground breaking e-learning courses.

Automatic Cognitive Load Detection Based on Speech Features

Dr Fang Chen — Fri, 14 March 2008 13:24 +1000

This talk focuses on potential speech feature indices. It presents attempts to induce controlled levels of load and solicit natural speech, and the use of machine learning in the development of a speech classifier that is able to detect different levels of load in the speech signal. The user studies confirm a significant variation of speech features in different cognitive load levels.

Ergonomic requirements, methods, and standards for user centered software systems

Dr Dominique Scapin — Fri, 7 March 2008 13:24 +1000

This talk aims at stressing the human aspects in the design and evaluation of software systems, from the point of view of engineering ergonomics. It does not aim at exhaustiveness, but at providing a large panorama of the main issues. After a few definitions and statements on the needs for a user centered approach, the talk describes the main ergonomic requirements to be taken into account. Then, after a survey on usability methods and standards, the choice of ergonomic methods is discussed within the software life cycle steps and other parameters of the context. The talk concludes with an outline of the major issues to be retained.

Embedding Expert Users in the Interaction Design Process

Dr Alastair Weakley — Fri, 26 February 2008 13:24 +1000

This presentation will describe an approach to interaction design that evolved from having a group of expert users as clients to being a participatory interaction design project. We will detail the user- centred design and development of an application with pliable interaction that enables the search, display and filtering of email data beyond the standard functionalities available in conventional email software. We will describe how one of the expert users adopted the role of 'embedded researcher' while working as part of a team of interaction designers outside of the users' organizational environment. This process is part of an iterative cycle of interaction design emerging from the Virtual Communities project at the Australasian Co-operative Research Centre for Interaction Design (ACID).

Being There Together: The Future of Social Interaction in Virtual Environments

Dr Ralph Shroeder — Fri, 22 February 2008 13:24 +1000

This presentation will provide an overview of different types of shared virtual environments and how people interact in them. This includes immersive (Cave-type) and PC-based environments, small and large groups, and collaborative instrumental tasks as well as socializing. Various aspects of interaction - navigation, communication, avatar appearance, and comparisons with real world interaction - will be highlighted and results from a number of studies of these topics summarized. The argument is made that we can anticipate what the future of 'being there together' will look like, both technically and socially. The implication is that we can use this forecasting to improve the social and technological shaping of shared virtual environments.

The Text Attribution Tool: author profiling for English emails

Dr Dominique Estival and Dr Tanja Gaustad van Zaanen — Fri, 9 November 2007 13:24 +1000

The Text Attribution Tool (TAT) aims at automating the analysis of texts for the purpose of author profiling and identification. It provides probabilities for the author's basic demographic traits (gender, age, geographic origin, level of education and native language) as well as for five psychometric traits. The TAT has been developed for the purpose of language-independent author profiling and has now been trained on two email corpora, English and Arabic. In this talk, we will describe the email data which was collected for the project, the ways this data is processed and analysed, and the experimental setup used for classification with the TAT. We will describe the overall TAT system and the Machine Learning experiments resulting in classifiers for the different author traits before presenting our results for the demographic and psychometric traits using the English email data. Results are very promising for all ten traits examined.

Annodex and The Future of Online Media

Shane Stephens — Wed, 12 September 2007 15:49 +1000

Current multimedia offerings on the World Wide Web present media as an indivisible binary blob - accessing a particular section within a resource is typically not possible. In general, providing access within media is a hard problem - video and audio formats are typically compressed at variable bit-rates; video formats often require deep references to earlier parts of the file; and searching across multiple media streams (e.g. video and audio) requires careful consideration about synchronization. The Annodex platform is a set of specifications and technologies, produced by CSIRO and released under an open-source license, that attempts to redress this deficiency for Ogg-based media. Annodex adds two basic concepts to a media resource - a textual "track" that contains time-aligned annotations and is part of the resource; and a means of addressing time points and time ranges within media in a streaming fashion. With pervasive access within media, it becomes possible to create a range of enhanced, media-centric applications - for example, search engines that can direct users directly to the relevant section of videos; reference sites for primary multimedia sources; online video-remixing applications; community annotated video collections; etc. This talk will provide a brief demonstration of the technology, discuss some of the problems which needed to be solved, and overview some current and proposed web applications centered around Annodex. Future research questions and engineering challenges will also be examined.

From Monomodal to Multimodal: Affect Recognition Using Visual Modalities

Hatice Gunes — Tue, 28 August 2007 15:49 +1000

Affective computing has emerged with the aim to enable affective human-computer interaction by designing machines and interfaces that will sense, recognise, understand and interpret human emotional states via language, speech, facial and bodily gesture and respond accordingly. Although much progress has been achieved in the last decade, one major present limitation of affective computing has been that most of the research on emotion recognition has focused on one single sensorial source, or modality, at a time and especially the face display. While it is true that the face is the main display of a human's affective state, other sources can improve the recognition accuracy. As natural human-to-human interaction is multimodal, the single sensory observations are often ambiguous, uncertain, and incomplete. Despite this fact, the research community has only recently started proposing emotion recognition systems using affective multimodal data. This talk will introduce recent advances in multimodal affect recognition by focusing on visual modalities. The talk will start by defining affect and emotions and provide a brief historical background of the research field. The problem domain of multimodal affective computing will be discussed next, by focusing on background research, data collection, data annotation, synchrony between modalities, data integration or fusion, information complementarity or redundancy, and information content of modalities. A number of representative systems introduced within the last 5 years analysing monomodal face or body display will be presented. The talk will then cover the representative systems recognising affective bimodal/multimodal data from visual modalities. The limitations of the current systems will be summarized and the features of an 'ideal' multimodal affect analyser will be discussed in order to provide an insight for the future of affective computing.

An interdisciplinary approach to CSCW research: An example of collaboration between engineers and sociologists

Professor Hideaki Kuzuoka — Wed, 15 August 2007 14:33 +1000

In order to improve the practice of CSCW research, we need to consider the process of system development, analyze system usage, and redesign the improved system based on the implications from the analysis. In working towards these goals, we have formulated an international research group consisting of engineers and sociologists, and have reported the initial results of our research at conferences such as CSCW, CHI, and ECSCW. In this seminar, through an introduction of our research in chronological order, I will discuss how engineers and sociologist have been effectively working together within CSCW research.

Enriching and using large-scale lexico-semantic resources

Dr Eneko Agirre — Thu, 9 August 2007 17:36 +1000

This talk presents a research line for the semantic processing of natural language. Current limitations of semantic interpretation systems stem from the pervasive ambiguity in natural language. Our working hypothesis is that a complex mixture of resources (ranging from hand-made knowledge bases, lexicons and hand-tagged corpora to raw corpora) and machine learning strategies built by different researchers need to be integrated in order to advance the state-of-the-art. An architecture using the EuroWordNet design will be presented, with some examples of integrating both hand-made resources and automatically learned relations. Finally, some promising applications on CLIR are presented, as well as the ongoing SemEval 07 / CLEF 2008 exercise on evaluating the contribution of WSD to CLIR.

Games and learning: Towards a more formal approach

Dr Nicolas Szilas — Thu, 2 August 2007 16:21 +1000

Using games for learning has become the new trend in Educational Technologies. Because games are fun and promote user's activity, games would be the perfect candidate to replace the good old fashioned pedagogy of our old universities. However, when looking in more detail, few scientific experimentations demonstrate the effective benefit of games for learning. Besides, some uses of games in real educational settings tend to prove that the reality of "games for learning" is much less encouraging than papers or books on the field claimed. Given that context and with the general goal to encourage the design of games dedicated to learning in various fields of knowledge, we have felt the need to improve the theoretical grounding to the idea that games could be used as educational tools. In this talk, I will provide a theoretical account of what a game is and to which extent it is related to learning. Several types of articulations between learning and playing will be deduced, which are useful both for analysing existing educational games and designing new games.

Computer corpora and language description

Professor Pam Peters — Tue, 20 July 2007 16:25 +1000

This presentation examines ongoing challenges for automatic analysis of (i) standard general language and (ii) specialised sublanguages, where additional layers of meaning (denotative and connotative) are still crucial for sophisticated NLP systems. Computational techniques based on small, purpose-designed corpora have been used by linguists since the 1960s to quantify lexical and grammatical elements of standard English, and to support intercomparisons between varieties of English. Corpus frequencies can show numerous syntactic divergences between major varieties such as British and American, as reported in the Longman Grammar of Spoken and Written English (Biber et al. 1999). Interesting differences have likewise been found in quantitative studies of new varieties of English such as those of Australia, New Zealand, Singapore, Philippines (Hundt, 2006), e.g. syntactic variables such as the patterns of agreement for collective nouns. However the subtle polysemy of most high frequency words still requires discretionary analysis, to separate common and distinctive senses of words -- despite the availability of mutual information tools. This problem affects usage of the function words and phrases of English, e.g." in case" in British and American English, as well as new usages found in ex-colonial Englishes, e.g. Singaporean use of "until". Conjunctions/prepositions like these define the logical relationships between the content-bearing clauses/phrases of the sentence and are the key to their interpretation. Other regional differences, e.g. the British preference for "about" v American for "around" are more cosmetic. They nevertheless serve to geolocate the text to some extent -- give it a regional tinge which may or may not matter to its writers and readers, and may or may not be reinforced by other more obvious though less frequent regionalisms of the "sidewalk"/"pavement" kind. Computer-based techniques for profiling specialised forms of language, aka sublanguages, also go back several decates in research by information engineers such as Bross, Shapiro and Anderson (1972) on the language of hospital surgeons. The identification of specialised terms and constructions is based on the principle that they occur with much greater frequency in technical texts than those intended for general reading (e.g. newspapers). A corollary of this is that the corpus needed to profile the terminology of a specialism need not be so large as that needed to support research on the lexis of the standard language (McEnery and Wilson, 1996/2001). Comparative frequency data from general and specialised corpora are effective in identifying the technical terminology of a discipline such as anatomy (Chung, 2003). However the terminology of different academic disciplines is rather variable in scope, and experimental research has shown that the density and structure of terms is quite different in texts from anatomy and, say, applied linguistics (Chung and Nation 2003). A key conceptual issue is whether to include in the inventory of terms only those which are distinctive to the discipline (the traditional terminological approach), or to embrace also those terms which are special uses of everyday words, e.g. "menu" in computer science (the descriptive terminologist's approach). The latter are essential for comprehensive coverage and professional training, but again they raise problems of polysemy for automatic analysis of corpora. The presentation will demonstrate the combination of computational and discretionary techniques, involving both linguists/lexicographers and disciplinary specialists, which is currently being used at the Dictionary Research Centre to build online termbanks of specialised expressions for academic disciplines in science and social science at Macquarie University (the TermFinder project).

Linguistics and the development of community in computer mediated conversation

Dr Barbara Kelly — Tue, 5 June 2007 16:20 +1000

How is online language different from face-to-face language, and how is the enormous popularity of social networking websites such as MySpace and Facebook affecting how we communicate? With the explosion of the internet there has been a lot of interest in the types of communications which occur in online communities. Language use in computer mediated communication (CMC) shares many linguistic features typically found in written language, such as punctuation and capitalization, while attempting to incorporate information common in casual face-to-face spoken communication (FFC), such as prosodic marking and subject and copula omission (Cherny 1999, Herring 1996, 2005). Dialogue participants are creative in their use of different paralinguistic communication tools such as emoticons like <:> suggesting positive associations, and acronyms like for 'laughing out loud'. While use of these strategies helps lessen communicative differences between FFC and CMC by attempting to make written communication seem like speech, electronically mediated conversations have often been analyzed as exhibiting fewer social norms than have face to face encounters occurring in real time (Brennan and Ohaeri 1999). Typically this is attributed to the lack of physical co-presence in CMC. Conversation participants are considered to be less socially responsible and more de-personalized in non-face-to-face communications. They are more individual in their interactions and behave less like socially responsible members of a community. However, determining what constitutes a community in CMC is somewhat problematic since most claims rely on subjective assessments, either of the researchers or of the participants. In this talk I present an analysis of CMC discourse interactions among participants in a conversation-like CMC system. Data for the study comes from samples of on-line conversations over a six-month period from different online message boards occurring within a large multi-national organization. Although the topics for the boards are discrete and unrelated, several participants post to more than one board. I present evidence that participants in the different message board systems use non-standard written linguistic features which suggest more casual, face-to-face-like interactions. These include omission of pronouns and copulas, non standard punctuation such as , <@?@>, stylistics including emphasis markers such as *really* and capitalization to indicate YELLING. Using a Social Network Analysis framework I show that participants who exhibit the most frequent uses of non-standard linguistic features are those system users who have the most dense social networks within the online community in which they are posting. That is, they have links to several other participants in the system either through other on-line communications or through face-to-face communication. Extending the results of Kelly and Halverson (2003) I show that several users who have dense relationships with other users in one message board and use non-standard linguistic features in this system exhibit little use of non-standard language in their postings on other bulletin boards in which they do not have such dense social networks. This work presents a promising approach for confirming the classification of online discourse communities via their linguistic behavior and their relationships as determined through a social network analysis.

24/7 Personal Heart Monitoring and Rehabilitation System using Smart Phones and Wireless Sensors

Dr Valerie Gay and Dr Peter Leijdekkers — Tue, 22 May 2007 15:35 +1000

This presentation describes our work in progress regarding personalized heart monitoring using smart phones. Our research combines ubiquitous computing with mobile health technology. We use wireless sensors and smart phones to monitor the wellbeing of high risk cardiac patients. The smart phone analyses in real-time the ECG data and determines whether the person needs external help. Depending on the situation the smart phone can automatically alert pre assigned caregivers or call the ambulance. It is also used to give advice (e.g. exercise more) or to reassure the patient based on the sensors and environmental data.

NLP of Medical Records

Professor Jon Patrick — Tue, 8 May 2007 17:39 +1000

The medical record contains most of its worthwhile content in natural language and so building a domain specific processor would seem to be a useful and straightforward task. Unfortunately that is far from the real situation. The effective extraction of useful content from the medical record requires coming to terms with large medical ontologies, serious gradations of grammaticality, and documents of different registers, for example discharge summaries, clinical notes, and pathology reports. The motivations for processing the medical record are also diverse and leave open a series of design questions about the nature of how to target the NLP for the post processing systems. Some of these processing systems are decision support systems to guide patient care, auditing patient care against clinical protocols, extracting knowledge from published texts, or for epidemiological analysis. The task is further complicated by the need to extract the data from large legacy hospital information systems and in some cases to bolt on new technology to old systems. These problems will be discussed with illustrations from the large range of projects oriented around the needs of our health collaborators.

GLEU: Automatic Evaluation of Sentence-Level Fluency

Dr Mark Dras — Mon, 30 Apr 2007 11:21 +1000

In evaluating the output of language technology applications -- MT, natural language generation, summarisation -- automatic evaluation techniques generally conflate measurement of faithfulness to source content with fluency of the resulting text. We have developed an automatic evaluation metric to estimate fluency alone, by examining the use of parser outputs as metrics, and we show that they correlate with human judgements of generated text fluency. From this we have developed a machine learner based on these, which performs better than the individual parser output metrics, approaching a lower bound on human performance. We have also investigated the effect on the metric of different language models for generating sentences, and show that while individual parser metrics can be 'fooled' depending on generation method, the machine learner provides a consistent estimator of fluency. This is joint work with Andy Mutton and Stephen Wan.

Designing Creativity Support Tools

Dr Shigeki Amitani — Tue, 10 Apr 2007 14:36 +1000

The goal of designing creativity support tools is to make more people more creative more often, enabling them to successfully cope with a wider variety of challenges and even straddle domains. Some tasks may be routine, such as doing computations or searching databases, while others require innovative leaps to identify associations, discover correlations, or recognise opportunities. In this seminar, I introduce fundamental concepts in the research field of creativity support and show examples of creativity support tools from my previous and current works.

Mining Syntactically Annotated Corpora

Dr Gosse Bouma — Tue, 3 Apr 2007 16:38 +1000

Using a robust and accurate wide-coverage parser, large syntactically annotated corpora can be constructed easily. In this talk, we will review a number of application areas where such corpora for Dutch have been found useful: to study the distribution of certain syntactic constructions (i.e. word order in indirect object constructions, the distribution of focus particles inside PPs, (alleged) extraction of PPs from NPs, etc.), to acquire lexical and ontological information (ranging from support verb constructions to definition sentences), and for relation extraction and question answering. An issue in all applications is the development of tools for searching, extracting, and combining information from treebanks stored in XML. Recently, we have started to use XQuery, a generic XML query language based on XPath, for a such tasks. An interesting feature of the language is the module system, which allows the definition of treebank-specific functions that can be used to support advanced extraction tasks.

Multimedia Applications and New Media Arts

Associate Professor Brigitte Kerherve — Tue, 27 Mar 2007 18:27 +1000

In the last two decades, multimedia research community had made major achievements in different areas such as content analysis, processing and retrieval; systems and networks as well as multimedia tools and applications. Nevertheless, producing multimedia content integrating texts, images, videos and audio is still complex and time-consuming. For example, in their everyday work, new media artists and creators face the complexity and difficulties of multimedia art creation, installation, delivery and archiving. In this talk, we will present our collaborative work with colleagues from the School of Visual and Media Arts at UQAM in the framework of Hexagram, the Institute for Research/Creation in Media Arts and Technologies. We will present different research projects where we investigate mobility, adaptability and representation for new forms of narrative in audio-video artwork.

Self-Organizing Map for Visual Analytics

Dr. Masahiro Takatsuka — Wed, 14 Mar 2007 15:14 +1000

A Self-Organizing Map is a type of Artificial Neural Networks and has been primarily used for 1) Associative Memory and 2) Data Visualization. In either cases, the main function of the SOM is to model multivariate data. This modeling corresponds to learning the statistical structure of the multivariate data in a much simpler form (typically in a 2D space). During this process, the SOM will try to keep the topological structure of the original data as much as possible. Hence, it achieves its non-linear topological mapping from the high-dimensional space to the much lower (typically 2D) space. With this topological mapping capability, it allows us to visualize complex high-dimensional data structures. Although the original form of the SOM does provide a useful visualization mechanism to a certain extent, it falls short of requirements in Visual Analytics. Visual Analytics is a emerging discipline, and it aims to utilize Visualization and User Interfaces in order to improve an Analytical Reasoning process. In this talk, SOM's shortcomings will be explained and some possible solutions and future challenges will be presented.

Wikipedia and Question Answering

Dr. George Ferizis — Tue, 13 Feb 2007 15:10 +1000

Information retrieval as observed by the wider community typically involves a user entering a set of keywords and receiving a set of documents that address their information needs. Question answering does more than this. It invites the user to enter a natural language question as a query. Using the extra information in the natural language query, a question answering system returns only information specific to the user's query. Wikipedia is an online encyclopaedia made up of user-contributed content. Wikipedia is an interesting corpus for research as it contains a large amount of open-domain information with a large amount of user-contributed classification information and a large amount of descriptive metadata in the form of multiple titles per document. In this talk we present a system that performs question answering using data in Wikipedia. We also discuss properties of Wikipedia that make question answering using the corpus interesting.

Engineering an Effective Social Interface for Sharing Digital Photographs

Trent Apted — Tue, 30 Jan 2007 15:20 +1000

The digital camera has changed the nature of photography and the photographic process, and the need for more suitable ways to share digital photographs is well understood. Digital photos may be printed, but this adds an expense, and reduces the search and distribution capabilities of photos in digital form, so some opt not to print at all. Thus some of the traditional methods of sharing photographs, such as photo albums, have been lost; along with the emotional attachment and social interaction that is possible with physical items. We have recently begun to see hardware support for the "tabletop" user interface, offering a number of new ways for humans to interact with computers. Our interface, currently in the form of a high-fidelity prototype, leverages the multi-user and direct-interaction capabilities of the tabletop, as well as aspects of multimodal and pervasive computing to provide a natural interface for sharing digital photographs in a collocated social setting. However, designing such an interface and making it effective is not a straightforward task. Such an interface needs to overcome problems of clutter and orientation inherent in tabletop interfaces, and provide an experience that allows the users to ignore the computer and conduct their dialogue unhindered for a truly social interaction. This talk will discuss our approach taken to engineer a highly usable and versatile software platform, and the high-quality graphical environment used in our prototype. The development is driven by a number of application "modes" allowing multiple, collocated users to load, share, search, browse, augment (with audio), tag, annotate, associate, crop, combine and present digital photographs on the tabletop interface. This will also be covered, as well as a facility for "live" photos -- interactive images updated from a remote computer screen.

Quiet Interfaces that Help People Think

Professor Sharon Oviatt — Tue, 12 Dec 2006 16:00 +1000

As technical as we have become, modern computing has not permeated many important areas of our lives, including mathematics education which still involves pencil and paper. In the present study, twenty high school geometry students varying in ability from low to high participated in a comparative assessment of math problem solving using existing pencil and paper work practice (PP), and three different interfaces: an Anoto-based digital stylus and paper interface (DP), pen tablet interface (PT), and graphical tablet interface (GT). Cognitive Load Theory correctly predicted that as interfaces departed more from familiar work practice (GT > PT > DP), students would experience greater cognitive load such that performance would deteriorate in speed, attentional focus, meta-cognitive control, correctness of problem solutions, and memory. In addition, low-performing students experienced elevated cognitive load, with the more challenging interfaces (GT, PT) disrupting their performance disproportionately more than higher performers. The present results indicate that Cognitive Load Theory provides a coherent and powerful basis for predicting the rank ordering of users? performance by type of interface. In the future, new interfaces for areas like education and mobile computing could benefit from designs that minimize users? load so performance is more adequately supported.

Interactive lecturing - teaching and learning in wireless networks

Dr Anja Wessels — Tue, 5 Dec 2006 13:15 +1000

The conventional lecture scenario implicates fundamental didactic problems due to a lack of interactivity and opportunity for feedback. Particularly in large meetings, it is difficult for the lecturer to respond to individual questions or remarks or to maintain the students' attention permanently to enable deeper cognitive processes. From the students' point of view, communication with the instructor is very limited in these typical mass meetings. In this talk the learning paradigm of the interactive lecture is introduced which is supported by Wireless LAN. The use of this transmission technology allows interactivity and adaptivity in lectures through bi-directional synchronous communication between all participants. A quiz service for example allows a posting of questions, an evaluation of the students' answers and a graphical presentation of the results. Academic researchers and professionals in the field of education, psychology, wireless technologies and engineering might benefit from attending and experiencing the talk.

Cognitive Load Measurement through Multimodal Behavior Patterns

Dr Fang Chen — Tue, 24 Oct 2006 19:21:00 +1000

Multimodal interfaces expand the communication channel between the system and the user allowing users to express themselves more naturally and interact with complex information with more freedom of expression. One of the many cited advantages of multimodal interfaces is their ability to facilitate effortful complex tasks over unimodal interfaces. These strategies often result in changes to the way multimodal constructions are planned and executed. Cognitive load refers to the amount of mental effort imposed by a particular task and has been associated to the limited capacity of working memory. I will start with an overview of the state of the art in cognitive load measurement. Recent research has shown that users' multimodal constructions exhibit significant changes as they self-manage their cognitive load when faced with tasks of increasing complexity. Our research focuses on extending the accepted benefits of multimodal interaction by using it to detect fluctuations in cognitive load will be stressed. The primary advantage of this approach is that cognitive load can be determined implicitly by monitoring variations of specific multimodal features during day to day tasks. Such unobtrusive measures may help determine user's cognitive load in real time and adapt information content selection and presentation (multimodal output generation) accordingly, in order to ensure optimal user performance. In this talk, I will describe an experiment designed to identify the relationships between combined speech and manual gesture input structures and users' cognitive load. The two input modalities are very familiar to users and psychologically closely interrelated, both in terms of planning and execution. Assessing a user's cognitive load implicitly through their multimodal behaviour requires identifying a number of indices that reliably reflect fluctuations. Our hypothesis is that variations in redundant and complementary multimodal constructions can reflect cognitive load changes experienced by the user. The feasibility of using rates of redundant constructions or even complementary constructions in multimodal input as an index of cognitive load is supported by the results of our study. I will illustrate multimodal patterns that may be monitored to detect cognitive load variations based on symptomatic behavioural features. I will conclude with a discussion on the enormous impact such methods may bring to the design of human computer interaction systems, but highlight the current limitations of the pattern acquisition methodology. Directions for future work will also be addressed.

The Language of Weblogs: A study of genre and individual differences

Dr Scott Nowson — Tue, 10 Oct 2006 19:00:00 +1000

This talk gives an overview of a linguistic investigation of individual differences in online personal diaries, or 'blogs.' There is substantial evidence of gender differences in language (Lakoff, 1975), and to a lesser extent linguistic projection of personality (Pennebaker & King, 1999). New forms of computer-mediated communication - unrestricted as to content and form guidelines - allow users total freedom for individual expression. This makes weblogs the perfect choice for exploring linguistic expression of individual differences. This talk will present findings from a study to identify linguistic features which can be used to differentiate between not only gender, but personality type. However, it is further recognised that whilst people are capable of recognising personality generally, their ability varies considerably across specific traits. In an attempt to develop automatic personality recognition, early results in machine classification by a selection of the identified linguistic features will also be discussed and compared to human performance.

Computing with Meaning by Operationalising Socio-cognitive Semantics

Robert McArthur — Tue, 3 Oct 2006 19:00:00 +1000

Robert McArthur from the CSIRO ICT Centre discusses his work based on the framework of conceptual spaces. Dr Peter Gärdenfors has proposed the framework of conceptual spaces to represent meaning. Between the symbolic representation (like first order logic) and subconceptual representation (like neural networks), the conceptual representation uses a geometric paradigm for modelling semantics. This talk describes the outcomes of bringing together Gärdenfors' theory with existing and modified cognitive algorithms, and applying them to mailing list, email and blog data. The meanings in the semantic spaces were exploited in the areas of charting the topical ebbs and flows in an online community, visualising a computational sense-of-self of people with chronic illness, managing expertise within enterprises, the discovery of social networks using tacit knowledge, and finally the uncovering and use of deeper notions of user context.