Monte Carlo techniques for phrase-based translation

Recent advances in statistical machine translation have used approximate beam search for NP-complete inference within probabilistic translation models. We present an alternative approach of sampling from the posterior distribution defined by a translation model. We define a novel Gibbs sampler for sampling translations given a source sentence and show that it effectively explores this posterior distribution. In doing so we overcome the limitations of heuristic beam search and obtain theoretically sound solutions to inference problems such as finding the maximum probability translation and minimum risk training and decoding.     

Negative results for software effort estimation

More than half the literature on software effort estimation (SEE) focuses on comparisons of new estimation methods. Surprisingly, there are no studies comparing state of the art latest methods with decades-old approaches. Accordingly, this paper takes five steps to check if new SEE methods generated better estimates than older methods. Firstly, collect effort estimation methods ranging from “classical” COCOMO (parametric estimation over a pre-determined set of attributes) to “modern” (reasoning via analogy using spectral-based clustering plus instance and feature selection, and a recent “baseline method” proposed in ACM Transactions on Software Engineering). Secondly, catalog the list of objections that lead to the development of post-COCOMO estimation methods. Thirdly, characterize each of those objections as a comparison between newer and older estimation methods. Fourthly, using four COCOMO-style data sets (from 1991, 2000, 2005, 2010) and run those comparisons experiments. Fifthly, compare the performance of the different estimators using a Scott-Knott procedure using (i) the A12 effect size to rule out “small” differences and (ii) a 99 % confident bootstrap procedure to check for statistically different groupings of treatments. The major negative result of this paper is that for the COCOMO data sets, nothing we studied did any better than Boehms original procedure. Hence, we conclude that when COCOMO-style attributes are available, we strongly recommend (i) using that data and (ii) use COCOMO to generate predictions. We say this since the experiments of this paper show that, at least for effort estimation, how data is collected is more important than what learner is applied to that data.     

Better Living Through Geometry

Mark Weiser described ubiquitous computing as “invisible, everywhere computing that does not live on a personal device of any sort, but is in the woodwork everywhere”[1]. The EasyLiving project is concerned with development of an architecture and technologies for ubiquitous computing environments that allow the dynamic aggregation of diverse I/O devices into a single coherent user experience. Though the need for research in distributed computing, perception and interfaces is widely recognised, the importance of an explicit geometric world model for enhancing the user’s experience of a ubiquitous computing system has not been well-articulated. This paper describes three scenarios that benefit from geometric context and introduces the EasyLiving Geometric Model.     

From insect to Internet: Situated control for networked robot teams

Ant-like systems take advantage of agents' situatedness to reduce or eliminate the need for centralized control or global knowledge. This reduces the need for complexity of individuals and leads to robust, scalable systems. Such insect-inspired situated approaches have proven effective both for task performance and task allocation. The desire for general, principled techniques for situated interaction has led us to study the exploitation of abstract situatedness – situatedness in non-physical environments. The port-arbitrated behavior-based control approach provides a well-structured abstract behavior space in which agents can participate in situated interaction. We focus on the problem of role assumption, distributed task allocation in which each agent selects its own task-performing role. This paper details our general, principled Broadcast of Local Eligibility (BLE) technique for role-assumption in such behavior-space-situated systems, and provides experimental results from the CMOMMT target-tracking task. This revised version was published online in August 2006 with corrections to the Cover Date.     

AutoTutor: Incorporating Back-Channel Feedback and Other Human-Like Conversational Behaviors into an Intelligent Tutoring System

This paper describes our recent attempts to incorporate human-like conversational behaviors into the dialog moves delivered by an animated pedagogical agent that simulates human tutors. We first present a brief overview of the modules comprising AutoTutor, an intelligent tutoring system. The second section describes a set of conversational behaviors that are being incorporated into AutoTutor. The behaviors of interest involve variations in intonation, head movements, arm and hand movements, facial expressions, eye blinking, gaze direction, and back-channel feedback. The final section presents a recent empirical study concerned with back-channel feedback events during human-to-human tutoring sessions. The back-channel feedback events emitted by tutors are mostly positive (63%), mostly verbal (77%), and immediately follow speech-act boundaries or noun-phrase boundaries (83%). Tutors also deliver back-channelevents at a very high rate when students are emitting dialog, about 13 events per minute. Conversely, 88% of students' back-channel feedback events are head nods, and they occur at unbounded locations (63%).     

Nonsystematic Search and No-Good Learning

Nonsystematic search algorithms seem, in general, to be well suited to large-scale problems with many solutions. However, they tend to perform badly for problems with few solutions, and they cannot be used for insoluble problems, since they are incomplete.Here we present a new algorithm, learn-SAT, that, although based on nonsystematic search, is complete. Completeness is realized through a process of no-good learning, learning-by-merging. This requires exponential space in the worst case. We show, nevertheless, that learn-SAT performs very well on certain SAT problems that are tightly constrained or insoluble. Indeed, its performance generally approximates the best SAT algorithms and does much better at lower clause densities. Learn-SAT also maintains much of the efficient performance of nonsystematic search for large-scale problems with many solutions, at least relative to backtrack search algorithms.These results indicate that the burden on memory, imposed by no-good learning, is not generally a problem for learn-SAT. This is perhaps surprising in view of previous work. What is even more surprising is the scalability of learn-SAT. For some types of problem it scales very much better than the nearest competitive algorithm. There are other types, however, for which this is not the case.The performance profile of learn-SAT emerges from an experimental methodology related to the one outlined by Mammen and Hogg in 1997.     

IRS-III: A broker-based approach to semantic Web services

A factor limiting the take up of Web services is that all tasks associated with the creation of an application, for example, finding, composing, and resolving mismatches between Web services have to be carried out by a software developer. Semantic Web services is a combination of semantic Web and Web service technologies that promise to alleviate these problems. In this paper we describe IRS-III, a framework for creating and executing semantic Web services, which takes a semantic broker-based approach to mediating between service requesters and service providers. We describe the overall approach and the components of IRS-III from an ontological and architectural viewpoint. We then illustrate our approach through an application in the eGovernment domain.     

Analysis of the human immune response to vaccinia by use of a novel protein microarray suggests that antibodies recognize less than 10% of the total viral proteome

Control of smallpox by mass vaccination was one of the most effective public health measures ever employed for eradicating a devastating infectious disease. However, new methods are needed for monitoring smallpox immunity within current vulnerable populations, and for the development of replacement vaccines for use by immunocompromized or low-responding individuals. As a measure for achieving this goal, we developed a protein microarray of the vaccinia virus proteome by using high-throughput baculovirus expression and purification of individual elements. The array was validated with therapeutic-grade, human hyperimmune sera, and these data were compared to results obtained from individuals vaccinated against smallpox using Dryvax. A high level of reproducibility with a very low background were apparent in repetitive assays that confirmed previously reported antigens and identified new proteins that may be important for neutralizing viral infection. Our results suggest that proteins recognized by antibodies from all vaccinees constituted <10% of the total vaccinia proteome.     

Three-dimensional spinal motion and risk of low back injury during sheep shearing

Sheep shearers are known to work in sustained flexed postures and have a high prevalence of low back pain (LBP). As sustained posture and spinal movement asymmetry under substantial loads are known risk factors for back injury our aim was to describe the 3D spinal movement of shearers while working. We hypothesised that thoraco-lumbar and lumbo-sacral movement would be tri-axial, asymmetric, and task specific. Sufficient retro-reflective markers were placed on the trunk of 12 shearers to define thoraco-lumbar and lumbo-sacral 3D motion during three tasks. Thoraco-lumbar movement consistently involved flexion, left lateral flexion, and right rotation. Lumbo-sacral movement consistently involved right lateral flexion in flexion with minimal rotation. Shearers therefore work in sustained spinal flexion where concurrent, asymmetric spinal movements into both lateral flexion and rotation occur. These asymmetric movements combined with repetitive loading may be risk factors leading to the high incidence of LBP in this occupational group.     

ProteinParser--a community based tool for the generation of a detailed protein consensus and FASTA output

Comparison of bioinformatic data is a common application in the life sciences and beyond. In this communication, a novel Java based software tool, ProteinParser, is outlined. This software tool calculates a detailed consensus, or most common, amino acid at a given position in an aligned protein set, whilst also generating a full consensus protein FASTA output. A second application of this software tool, computing a consensus amino acid given a tolerance threshold, is also demonstrated. The phytase and the common bacterial beta-lactamase proteins are analysed as 'proof of concept' examples. Consensus proteins, as generated by ProteinParser, are regularly utilised in the selection of residues for protein stabilisation mutagenesis; however, this widely applicable software tool will find many alternative applications in areas such as protein homology modelling.