Joseph Mariani

On the use of time compression for word-based recognition

Dynamic time warping is a very efficient technique in dealing with the problem of time distorsion between different pronunciations of any given worm. However when, in a word-based recognition system(isolated or connected words), time normalisation is solely based on a DR-matching process, much processing time is necessitated. Another characteristic is that the general constraints used to optimise the DP-matching algorithm impose a severe limit on acceptable time distorsions. In this paper, we evaluate the interesting aspects of non-linear time compression methods which carry out a first time normalisation prior to DP-matching in word-based recognition. We describe three non-linear compression techniques, which have come under consideration during the study of recognition systems developed at LIMSI. These non-linear compression methods are compared to linear compressions in an isolated word recognition framework for different vocabularies.

A dynamic time warp VLSI processors for continuous speech recognition

We present a VLSI processor designed to compute dynamic time warping algorithms for speech recognition with extreme rapidity. This processor works as a coprocessor in a classical system including a standard microprocessor and a digital signal processor. It uses its own local memory for reference utterances and intermediate results. It has been designed to give maximum efficiency on continuous speech recognition applications with or without syntax constraints. Its flexibility permits software optimisation and its use in a large number of different applications. We use a sequential approach for DTW computations and work along the time axis. All the calculations are carried out on each frame of the unknown utterance as soon as it arrives from the DSP and DTW computations therefore take place in real time. Response time is in hundredth of second; intermediate results are obtained before the end of the sentence. A system using this chip will be able to carry out continuous speech recognition in real timee on a vocabulary of 300 references. Many of those chips can be used in parallel on a single system.

Natural Interaction with Robots, Knowbots and Smartphones

These proceedings presents the state-of-the-art in spoken dialog systems with applications in robotics, knowledge access and communication. It addresses specifically: 1. Dialog for interacting with smartphones; 2. Dialog for Open Domain knowledge access; 3. Dialog for robot interaction; 4. Mediated dialog (including crosslingual dialog involving Speech Translation); and,5. Dialog quality evaluation. These articles were presented at the IWSDS 2012 workshop.

Spoken Dialogue Systems Technology and Design

Spoken Dialogue Systems Technology and Design covers key topics in the field of spoken language dialogue interaction from a variety of leading researchers. It brings together several perspectives in the areas of corpus annotation and analysis, dialogue system construction, as well as theoretical perspectives on communicative intention, context-based generation, and modelling of discourse structure. These topics are all part of the general research and development within the area of discourse and dialogue with an emphasis on dialogue systems; corpora and corpus tools and semantic and pragmatic modelling of discourse and dialogue.

Crowdsourcing for Language Resource Development: Criticisms About Amazon Mechanical Turk Overpowering Use

This article is a position paper about Amazon Mechanical Turk, the use of which has been steadily growing in language processing in the past few years. According to the mainstream opinion expressed in articles of the domain, this type of on-line working platforms allows to develop quickly all sorts of quality language resources, at a very low price, by people doing that as a hobby. We shall demonstrate here that the situation is far from being that ideal. Our goal here is manifold: 1- to inform researchers, so that they can make their own choices, 2- to develop alternatives with the help of funding agencies and scientific associations, 3- to propose practical and organizational solutions in order to improve language resources development, while limiting the risks of ethical and legal issues without letting go price or quality, 4- to introduce an Ethics and Big Data Charter for the documentation of language resource

Evaluation of time compression for connected word recognition

Recently, several studies have shown the interesting aspects of nonuniform sampling of the filtered speech signal in the context of an isolated word recognizer. This paper investigates the effect of nonuniform sampling for connected word recognition. Three nonlinear time compression techniques are evaluated, one that brings all reference utterances down to one same length, and others, for which the utterance length is variable. The nonuniform sampling approach is compared to the uniform one by opposing the three non-linear methods to two linear time compression ones. The results show that the variable length trace segmentation technique gives the best scores under all conditions, and that the uniform sampling approach can therefore be advantageously used in connected word recognition processes.

Developing Language Technologies with the Support of Language Resources and Evaluation Programs

The role of language resources and language technology evaluation is now recognized as being crucial for the development of written and spoken language processing systems. Given the increasing challenge of multilingualism in Europe, the development of language technologies requires a more internationally distributed effort. This paper first describes several recent and on-going activities in France aimed at the development of language resources and evaluation. We then outline a new project intended to enhance collaboration, cooperation, and resource sharing among the international language processing research community.

The language resource Strategic Agenda: the FLaReNet synthesis of community recommendations

Abstract The main purpose of this paper is to serve as a landmark for future research and in particular for future strategic, infrastructural and coordination initiatives. It presents a preliminary plan for actions and infrastructures that could become the basis for future initiatives in the sector of Language Resources and Technologies (LRTs). The FLaReNet Language Resource Strategic Agenda presents a set of recommendations for the development and progress of LRT in Europe, as issued from a three-year consultation of the FLaReNet European project. Recommendations cover a broad range of topics and activities, spanning over production and use of language resources, licensing, maintenance and preservation issues, infrastructures for language resources, resource identification and sharing, evaluation and validation, interoperability and policy issues. The intended recipients belong to a large set of players and stakeholders in LRT, ranging from individuals to research and education institutions, to policy-makers, funding agencies, SMEs and large companies, service and media providers. The main goal of these recommendations is to serve as an instrument to support stakeholders in planning for and addressing the urgencies of the LRT of the future.

Language Matrices and a Language Resource Impact Factor

We feel it is important to have a clear picture of what exists in terms of Language Resources and Evaluation (LRE) in order to be able to carry on research investigations in computational linguistics and develop language processing systems. The language coverage is especially important in order to provide technologies that can help multilingualism and protect endangered languages. It implies that one knows what is necessary and exists for some languages, detects the gaps for other languages, and finds a way to address them. In order to have access to that information, we based our study on the LRE Map, which was produced within the FLaReNet EC project. The LRE Map is built on data gathered at conferences directly from the authors, and therefore provides actual data obtained from the source, not an estimate of such data. At the time of this study, it covered 10 conferences from 2010 to 2012. We consider here Language Resources (LR) in the broad sense, including Data, Tools, Evaluation and Meta-Resources (standards, metadata, guidelines, etc.). We took into consideration the names, types, modalities and languages attached to each entry in the LRE Map. A huge amount of manual cleaning was necessary before being able to use the data. In order to check the availability of Language Resources for the various languages, we designed a software tool called “LRE Matrix” that automatically produces Language Matrices presenting the number of resources of various types that exist for various modalities for each language. We slightly modified the software code in order to also compute the number of times a Language Resource is mentioned, what we may call a “Language Resource Impact Factor” (LRIF). Given their quantitative, objective nature, our results are precious for comparing the situation of the various national and regional languages in Europe regarding the availability of Language Resources in a survey conducted within the META-NET network. We faced in our studies the need for a tedious normalization and cleaning process that showed the necessity to assign a Unique and Persistent Identifier to each Language Resource in order to identify it more easily and follow its use and changes over time, a process that requires an international coordination.

The French Language in the European Information Society

With 128 million “native and real speakers” worldwide and an estimate of close to 300 million persons speaking French overall, French appears only as the 16th most spoken native language, but as the 6th most spoken language in the world, after English, Chinese Mandarin, Spanish, Hindi and Russian.