The construction of spoken dialog systems today relies heavily on appropriate corpora, but corpus selection is more an art than a science.As interaction style properties govern many aspects of dialog, they have the potential to be useful for relating and comparing corpora.This paper overviews a recently-developed model of interaction styles and shows how it can be used to identify relevant corpus differences, estimate corpus similarity, and flag likely outlier dialogs.

There is increasing interest in modeling style choices in dialog, for example for enabling dialog systems to adapt to their users.It is commonly assumed that each user has his or her own stable characteristics, but for interaction style the truth of this assumption has not been well examined.I investigated using a vectorspace model of interaction styles, derived from the Switchboard corpus of telephone conversations and a broad set of prosodic-behavior features.While most individuals exhibited interaction style tendencies, these were generally far from stable, with a predictive model based on individual tendencies outperforming a speaker-independent model by only 3.6%.The tendencies were somewhat stronger for some speakers, including generally males, and for some dimensions of variation.

This paper describes a collaborative project between the Australian Urban Research Infrastructure Network (AURIN), the Australian Bureau of Statistics (ABS), the University of Queensland eResearch Lab and a number of social science research centres across Australia - that provides programmatic access to ABS Census data sets to enable its re-use within a range of research projects. The project successfully demonstrates machine-to-machine access to 2011 Census data through a federated data hub model that dynamically delivers statistical datasets to the research community. Using an SDMX web services approach, combined with advanced data analytics and visualization services available through the AURIN workbench, social scientists are able to manipulate and visualize national census data overlaid with other related data sets through a sophisticated mapping interface. Integrated statistical analysis services (R-based) also enable social scientists to quantify correlations between different demographic and socio-economic parameters. A number of socio-economic use-cases are presented that illustrate how the system enables researchers to understand and quantify changes in industry sectors, labor force needs, employment, population needs and disadvantage, over space and time. The paper also outlines problems and limitations revealed through the demonstrator projects, lessons learnt and areas that will require further effort to deliver optimum access to national census data sets and associated e-social science infrastructure for both the Australian and global social science community.

The field of Spatially Integrated Social Science (SISS) recognizes that much data of interest to social scientists has an associated geographic location. SISS systems use geographic location as the basis for integrating heterogeneous social science data sets and for visualizing and analyzing the integrated results through mapping interfaces. However, sourcing data sets, aggregating data captured at different spatial scales, and implementing statistical analysis techniques over the data are highly complex and challenging steps, beyond the capabilities of many social scientists. The aim of the UQ SISS eResearch Facility (SISS-eRF) is to remove this burden from social scientists by providing a Web interface that allows researchers to quickly access relevant Australian socio-spatial datasets (e.g. census data, voting data), aggregate them spatially, conduct statistical modeling on the datasets and visualize spatial distribution patterns and statistical results. This paper describes the technical architecture and components of SISS-eRF and discusses the reasons that underpin the technological choices. It describes some case studies that demonstrate how SISS-eRF is being applied to prove hypotheses that relate particular voting patterns with socio-economic parameters (e.g., gender, age, housing, income, education, employment, religion/culture). Finally we outline our future plans for extending and deploying SISS-eRF across the Australian Social Science Community.

When one tries to use the Web as a dictionary or encyclopedia, entering some single term into a search engine, the highly ranked pages in the result can include irrelevant or useless sites. The problem is that single‐term queries, if taken literally, underspecify the type of page the user wants. For such problems automatic query expansion, also known as pseudo‐feedback, is often effective. In this method the topndocuments returned by an initial retrieval are used to provide terms for a second retrieval. This paper contributes, first, new normalization techniques for query expansion, and second, a new way of computing the similarity between an expanded query and a document, the “local relevance density” metric, which complements the standard vector product metric. Both of these techniques are shown to be useful for single‐term queries, in Japanese, in experiments done over the World Wide Web in early 2001.

Today there are many services which provide information over the phone using a prerecorded or synthesized voice. These voices are invariant in speed. Humans giving information over the telephone, however, tend to adapt the speed of their presentation to suit the needs of the listener. This paper presents a preliminary model of this adaptation. In a corpus of simulated directory assistance dialogs the operator 's speed in number-giving correlates with the speed of the user's initial response and with the user's speaking rate. Multiple regression gives a formula which predicts appropriate speaking rates, and these predictions correlate (.46) with the speeds observed in good dialogs in the corpus. An experiment with 18 subjects suggests that users prefer a system which adapts its speed to the user in this way.

In human-to-human interaction, people sometimes are able to pick up and respond sensitively to the other's internal state as it shifts moment by moment over the course of an exchange. To find out whether such an ability is worthwhile for computer human interfaces, we built a semi-automated tutoring-type spoken dialog system. The system inferred information about the user's \scare{ephemeral emotions}, such as confidence, confusion, pleasure, and dependency, from the prosody of his utterances and the context. It used this information to select the most appropriate acknowledgement form at each moment. In doing so the system was following some of the basic social conventions for real-time interaction. Users rated the system with this ability more highly than a version without.

Ratings of user satisfaction, although fairly easy to elicit for today's spoken language systems, can be more elusive for systems which operate at near-human levels of performance. This problem can be alleviated by adding a `relistening ' phase before eliciting judgements: in this phase the user listens to a recording of himself interacting with the system while consulting a transcript of that interaction. This technique allows more sensitive judgements of system quality by avoiding problems arising from attention limits. 1 EVALUATING PERSONABLE SYSTEMS Speech technology has advanced to the point where there exist systems which allow motivated users to accomplish useful tasks. However there are also potential applications where users are not so motivated and where there is no clear criterion of task-completion, including applications in education, commerce, and entertainment. In these situations it will be important for interaction with the system to itself be pleasant, otherwise the...

Non-lexical speech sounds (conversational grunts), such as uh-huh, un-hn, mm, and oh, are common in English. In human dialogs these sounds are important in conversation control and for conveying attitudes. Spoken dialog systems may make use of these sounds to achieve concise, smooth, relaxed interactions. Doing so is, however, a challenge, because most algorithms used in spoken language processing were devised for words, but grunts are different from words both phonetically and semantically. For example, the phonetic inventory is different, superimposition of phonemes occurs, the set of conversational grunts is productive rather than finite, and the meanings are compositional and involve sound-symbolism. 1 PURPOSE This paper discusses non-lexical speech sounds (`conversational grunts' for short) in dialog, and points out possibilities and problems for their use in spoken language processing. 2 GRUNTS IN A CORPUS OF CONVERSATIONS This section presents some facts about the frequency ...

On the basis of corpus analysis, we havemade directiongiving dialog system which adjust the pace of dialog without using speech recognition. And we evaluate the naturalness of the resulting conversations by experiments. Then the system showed good performance. And also the possibility of prosody to compensate for the weak points of speech recognition. 1

This paper characterizes the methodology of Artificial Intelligence (AI) by looking at research in speech understanding, a field where AI approaches contrast starkly with the alternatives, particularly engineering approaches. Four values of AI stand out as influential: ambitious goals, introspective plausibility, computational elegance, and wide significance. The paper also discusses the utility and larger significance of these values.

In human language there seems to be a direct link between perception and action for back-channel feedback, that is, the grunts and other acknowledgments produced by a listener in a conversation. In particular, in Japanese, detection of a low pitch region in the speaker's utterance often cues the listener to produce back-channel feedback. This fact has been exploited in a system that can interact with unsuspecting speakers and appear to converse, in real time. This finding suggests that Brooks' 1991a, 1991b methodology for the study and synthesis of physical behavior is relevant also for social behavior. If so, then real-time responsiveness, rather than better phoneme recognition,is a top priorityfor the construction of spoken language systems and for the scientific study of language.

Looking to the future, generators will have more knowledge of language and will have to deal with inputs that are very rich in information. As a result, several problems will become more acute, including selecting what to say at the subproposition level and dealing with interaction among goals and dependencies among choices. This paper explains how these problems arise and why they are hard to handle within traditional architectures for generation. It also discusses why these issues have not been well addressed, including the current lack of demanding applications, excessive emphasis on linguistic traditions, the use of reverse engineering to determine generator inputs, and the tendency to research only one issue at a time.