Bootstrapping a statistical speech translator from a rule-based one

We describe a series of experiments in which we start with English to French and English to Japanese versions of an Open Source rule-based speech translation system for a medical domain, and bootstrap correspondign statistical systems. Comparative evaluation reveals that the rule-based systems are still significantly better than the statistical ones, despite the fact that considerable effort has been invested in tuning both the recognition and translation components; also, a hybrid system only marginally improved recall at the cost of a los in precision. The result suggests that rule-based architectures may still be preferable to statistical ones for safety-critical speech translation tasks.

Facilitatory and interfering effects of neighbourhood density on speech production: Evidence from aphasic errors.

In a system where tens of thousands of words are made up of a limited number of phonemes, many words are bound to sound alike. This similarity of the words in the lexicon as characterized by phonological neighbourhood density (PhND) has been shown to affect speed and accuracy of word comprehension and production. Whereas there is a consensus about the interfering nature of neighbourhood effects in comprehension, the language production literature offers a more contradictory picture with mainly facilitatory but also interfering effects reported on word production. Here we report both of these two types of effects in the same study. Multiple regression mixed models analyses were conducted on PhND effects on errors produced in a naming task by a group of 21 participants with aphasia. These participants produced more formal errors (interfering effect) for words in dense phonological neighbourhoods, but produced fewer nonwords and semantic errors (a facilitatory effect) with increasing density. In order to investigate the nature of these opposite effects of PhND, we further analysed a subset of formal errors and nonword errors by distinguishing errors differing on a single phoneme from the target (corresponding to the definition of phonological neighbours) from those differing on two or more phonemes. This analysis confirmed that only formal errors that were phonological neighbours of the target increased in dense neighbourhoods, while all other errors decreased. Based on additional observations favouring a lexical origin of these formal errors (they exceeded the probability of producing a real-word error by chance, were of a higher frequency, and preserved the grammatical category of the targets), we suggest that the interfering effect of PhND is due to competition between lexical neighbours and target words in dense neighbourhoods.

Normative Topographic ERP Analyses of Speed of Speech Processing and Grammar Before and After Grammatical Treatment.

Typically developing (TD) preschoolers and age-matched preschoolers with specific language impairment (SLI) received event-related potentials (ERPs) to four monosyllabic speech sounds prior to treatment and, in the SLI group, after 6 months of grammatical treatment. Before treatment, the TD group processed speech sounds faster than the SLI group. The SLI group increased the speed of their speech processing after treatment. Posttreatment speed of speech processing predicted later impairment in comprehending phrase elaboration in the SLI group. During the treatment phase, change in speed of speech processing predicted growth rate of grammar in the SLI group.

Toward a novel forensic intelligence model: systematic profiling of false identity documents

False identity documents represent a serious threat through their production and use in organized crime and by terrorist organizations. The present-day fight against this criminal problem and threats to national security does not appropriately address the organized nature of this criminal activity, treating each fraudulent document on its own during investigation and the judicial process, which causes linkage blindness and restrains the analysis capacity. Given the drawbacks of this case-by-case approach, this article proposes an original model in which false identity documents are used to inform a systematic forensic intelligence process. The process aims to detect links, patterns, and tendencies among false identity documents in order to support strategic and tactical decision making, thus sustaining a proactive intelligence-led approach to fighting identity document fraud and the associated organized criminality. This article formalizes both the model and the process, using practical applications to illustrate its powerful capabilities. This model has a general application and can be transposed to other fields of forensic science facing similar difficulties.

Forensic processing of false identity and travel documents to produce intelligence: a novel approach to assist in fighting organized crime

The production and use of false identity and travel documents in organized crime represent a serious and evolving threat. However, a case-by-case perspective, thus suffering from linkage blindness and a limited analysis capacity, essentially drives the present-day fight against this criminal problem. To assist in overcoming these limitations, a process model was developed using a forensic perspective. It guides the systematic analysis and management of seized false documents to generate forensic intelligence that supports strategic and tactical decision-making in an intelligence-led policing approach. The model is articulated on a three-level architecture that aims to assist in detecting and following-up on general trends, production methods and links between cases or series. Using analyses of a large dataset of counterfeit and forged identity and travel documents, it is possible to illustrate the model, its three levels and their contribution. Examples will point out how the proposed approach assists in detecting emerging trends, in evaluating the black market's degree of structure, in uncovering criminal networks, in monitoring the quality of false documents, and in identifying their weaknesses to orient the conception of more secured travel and identity documents. The process model proposed is thought to have a general application in forensic science and can readily be transposed to other fields of study.

Forensic intelligence framework-Part I: induction of a transversal model by comparing illicit drugs and false identity documents monitoring

Forensic intelligence is a distinct dimension of forensic science. Forensic intelligence processes have mostly been developed to address either a specific type of trace or a specific problem. Even though these empirical developments have led to successes, they are trace-specific in nature and contribute to the generation of silos which hamper the establishment of a more general and transversal model. Forensic intelligence has shown some important perspectives but more general developments are required to address persistent challenges. This will ensure the progress of the discipline as well as its widespread implementation in the future. This paper demonstrates that the description of forensic intelligence processes, their architectures, and the methods for building them can, at a certain level, be abstracted from the type of traces considered. A comparative analysis is made between two forensic intelligence approaches developed independently in Australia and in Europe regarding the monitoring of apparently very different kind of problems: illicit drugs and false identity documents. An inductive effort is pursued to identify similarities and to outline a general model. Besides breaking barriers between apparently separate fields of study in forensic science and intelligence, this transversal model would assist in defining forensic intelligence, its role and place in policing, and in identifying its contributions and limitations. The model will facilitate the paradigm shift from the current case-by-case reactive attitude towards a proactive approach by serving as a guideline for the use of forensic case data in an intelligence-led perspective. A follow-up article will specifically address issues related to comparison processes, decision points and organisational issues regarding forensic intelligence (part II).

Atypical coordination of cortical oscillations in response to speech in autism.

Subjects with autism often show language difficulties, but it is unclear how they relate to neurophysiological anomalies of cortical speech processing. We used combined EEG and fMRI in 13 subjects with autism and 13 control participants and show that in autism, gamma and theta cortical activity do not engage synergistically in response to speech. Theta activity in left auditory cortex fails to track speech modulations, and to down-regulate gamma oscillations in the group with autism. This deficit predicts the severity of both verbal impairment and autism symptoms in the affected sample. Finally, we found that oscillation-based connectivity between auditory and other language cortices is altered in autism. These results suggest that the verbal disorder in autism could be associated with an altered balance of slow and fast auditory oscillations, and that this anomaly could compromise the mapping between sensory input and higher-level cognitive representations.

New genes for focal epilepsies with speech and language disorders.

The last 2 years have seen exciting advances in the genetics of Landau-Kleffner syndrome and related disorders, encompassed within the epilepsy-aphasia spectrum (EAS). The striking finding of mutations in the N-methyl-D-aspartate (NMDA) receptor subunit gene GRIN2A as the first monogenic cause in up to 20 % of patients with EAS suggests that excitatory glutamate receptors play a key role in these disorders. Patients with GRIN2A mutations have a recognizable speech and language phenotype that may assist with diagnosis. Other molecules involved in RNA binding and cell adhesion have been implicated in EAS; copy number variations are also found. The emerging picture highlights the overlap between the genetic determinants of EAS with speech and language disorders, intellectual disability, autism spectrum disorders and more complex developmental phenotypes.

Alzheimer Disease Diagnosis based on Automatic Spontaneous Speech Analysis

Alzheimer’s disease (AD) is the most prevalent form of progressive degenerative dementia and it has a high socio-economic impact in Western countries, therefore is one of the most active research areas today. Its diagnosis is sometimes made by excluding other dementias, and definitive confirmation must be done trough a post-mortem study of the brain tissue of the patient. The purpose of this paper is to contribute to improvement of early diagnosis of AD and its degree of severity, from an automatic analysis performed by non-invasive intelligent methods. The methods selected in this case are Automatic Spontaneous Speech Analysis (ASSA) and Emotional Temperature (ET), that have the great advantage of being non invasive, low cost and without any side effects.

On cumulant techniques in speech processing

This paper analyzes applications of cumulant analysis in speech processing. A special focus is made on different second-order statistics. A dominant role is played by an integral representation for cumulants by means of integrals involving cyclic products of kernels.

Spontaneous Speech and Emotional Response modeling based on One-class classifier oriented to Alzheimer Disease diagnosis

The purpose of our project is to contribute to earlier diagnosis of AD and better estimates of its severity by using automatic analysis performed through new biomarkers extracted from non-invasive intelligent methods. The methods selected in this case are speech biomarkers oriented to Sponta-neous Speech and Emotional Response Analysis. Thus the main goal of the present work is feature search in Spontaneous Speech oriented to pre-clinical evaluation for the definition of test for AD diagnosis by One-class classifier. One-class classifi-cation problem differs from multi-class classifier in one essen-tial aspect. In one-class classification it is assumed that only information of one of the classes, the target class, is available. In this work we explore the problem of imbalanced datasets that is particularly crucial in applications where the goal is to maximize recognition of the minority class as in medical diag-nosis. The use of information about outlier and Fractal Dimen-sion features improves the system performance.