Gerrit Bloothooft

Corpus-based methods in language and speech processing

Introduction. 1. Corpus-Based Statistical Methods in Speech and Language Processing H. Ney. 2. Hidden Markov Models in Speech and Language Processing K. Knill, S. Young. 3. Spoken Language Dialogue Systems E. Giachin, S. McGlashan. 4. Part-of-Speech Tagging and Partial Parsing S. Abney. 5. Data-Oriented Language Processing R. Bod, R. Scha. 6. Statistical Language Modeling Using Leaving-One-Out H. Ney, S. Martin, F. Wessel. Author Information. Bibliography.

The Family Name as Socio-Cultural Feature and Genetic Metaphor: From Concepts to Methods

Abstract A recent workshop entitled “The Family Name as Socio-Cultural Feature and Genetic Metaphor: From Concepts to Methods” was held in Paris in December 2010, sponsored by the French National Centre for Scientific Research (CNRS) and by the journal Human Biology. This workshop was intended to foster a debate on questions related to the family names and to compare different multidisciplinary approaches involving geneticists, historians, geographers, sociologists and social anthropologists. This collective paper presents a collection of selected communications.

A breakpoint analysis procedure based on temporal decomposition

Temporal decomposition (TD), which is an analysis procedure based on a linear model of the effects of coarticulation, yields a linear approximation of a time sequence of speech parameters in terms of a series of time-overlapping interpolation functions and an associated series of data vectors. The number and positions of these interpolation functions show a high correspondence with phonetic events present in the speech signal. A new and more robust interpolation scheme for TD that gives it a geometric interpretation as a breakpoint analysis procedure in a multidimensional parameter space is described, where breakpoints are connected by straight line segments. The interpolation scheme can be viewed as a generalization of segmentation, allowing for a gradual transition from one segment towards the next. >

Name Clustering on the Basis of Parental Preferences

Abstract Parents do not choose first names for their children at random. Using two large datasets, for the UK and the Netherlands, covering the names of children born in the same family over a period of two decades, this paper seeks to identify clusters of names entirely inferred from common parental naming preferences. These name groups can be considered as coherent sets of names that have a high probability to be found in the same family. Operational measures for the statistical association between names and clusters are developed, as well as a two-stage clustering technique. The name groups are subsequently merged into a limited set of grand clusters. The results show that clusters emerge with cultural, linguistic, or ethnic parental backgrounds, but also along characteristics inherent in names, such as clusters of names after flowers and gems for girls, abbreviated names for boys, or names ending in –y or -ie.

Assessment of Systems for Nominal Retrieval and Historical Record Linkage

Problems in retrieval of names form large data bases and in nominal record linkage are discussed with respect to computational solutions. The quest for robust methods that can handle the typical variability of historical nominal information is discussed, with some emphasis on probabilistic methods. It is argued that comparison and assessment of different systems used on the same data could enhance our understanding of methodological issues.

Learning Name Variants from Inexact High-Confidence Matches

Name variants which differ more than a few characters can seriously hamper record linkage. A method is described by which variants of first names and surnames can be learned automatically from records that contain more information than needed for a true link decision. Post-processing and limited manual intervention (active learning) is unavoidable, however, to differentiate errors in the original and the digitised data from variants. The method is demonstrated on the basis of an analysis of 14.8 million records from the Dutch vital registration.

Evaluation of the Bayesian method to derive migration patterns from changes in surname distributions over time.

Abstract Known migration in The Netherlands between the periods 1950– 1969 and 2007, for 4.5 million individuals, was used to estimate the origin of migration by means of a Bayesian method on the basis of surname distributions in these two periods. Results of the method depend on the geographic specificity of the surnames and tend to be positioned between population density and actual probability of migration origin. An optimum in the correlation between estimated and actual percentages of origin of migration, and their differentiation as expressed by the correlation between the estimated and actual entropy across 40 distinguished areas, was found after a few iterations. The optimal correlation was 0.806 (Spearman), which shows that the Bayesian method provides a reasonable proxy of the rank order of a migrants origin.

Perception of synthetic emotion expressions in speech: Categorical and dimensional annotations

In this paper, both categorical and dimensional annotations have been made of neutral and emotional speech synthesis (anger, fear, sad, happy and relaxed). With various prosodic emotion manipulation techniques we found emotion classification rates of 40%, which is significantly above chance level (17%). The classification rates are higher for sentences that have a semantics matching the synthetic emotion. By manipulating the pitch and duration, differences in arousal were perceived whereas differences in valence were hardly perceived. Of the investigated emotion manipulation methods, EmoFilt and EmoSpeak performed very similar, except for the emotion fear. Copy synthesis did not perform well, probably caused by suboptimal alignments and the use of multiple speakers.

Multiple First Names in the Netherlands (1760–2014)

Although in Europe the custom to choose more than one name for a child first arose in the early Renaissance in Italy, its popularity reached the Netherlands only in the eighteenth century. On the basis of a high-coverage sample of the birth and name information for 26 million individuals from 1760 until 2014, the preference for multiple first names in the Netherlands has been studied, both annually and geographically. With the exception of recent years, religion has played a dominant role in name choices, including the number of names. Protestants only started to adopt double names in the nineteenth century, while Catholics increasingly chose three names, many including Maria (but only during a period of 50 years in the mid-twentieth century for boys). A tax proposal in 1915 on the number of names not only demonstrates the effect of financial cost on naming, but also the consistency of parents’ name choices concerning their children. Whereas during previous centuries the changes were gradual, preferences varied rapidly over the last 50 years.

Lexical Modeling for Proper name Recognition in Autonomata Too

The research in Autonomata Too aimed at the development of new pronunciation modeling techniques that can bring the speech recognition component of a Dutch/Flemish POI (Points of Interest) information providing business service to the required level of accuracy. The automatic recognition of spoken POI is extremely difficult because of the existence of multiple pronunciations that are frequently used for the same POI and because of the presence of important cross-lingual effects one has to account for. In fact, the ASR (Automatic Speech Recognition) engine must be able to cope with pronunciations of (partly) foreign POI names spoken by native speakers and pronunciations of native POI names uttered by non-native speakers. In order to deal adequately with such pronunciations, one must model them at the level of the acoustic models as well as at the level of the recognition lexicon. This paper describes a novel lexical modeling approach that was developed and tested in the Autonomata Too project. The new method employs a G2P-P2P (grapheme-to-phoneme, phoneme-to-phoneme) tandem to generate suitable lexical pronunciation variants. It was shown to yield a significant improvement over a baseline system already embedding state-of-the-art acoustic and lexical models.