Henning Reetz

Distinctive features: Phonological underspecification in representation and processing

Abstract Phonological variation of any sort (determined by speech styles, phrasing, or morphophonological rules) affecting the shapes of words and morphemes are a matter of concern for theories of speech perception and language comprehension. To come to grips with parsing the speech stream, accessing the lexicon and ultimately recognizing words, both representational as well as processing issues must be considered. The central questions in the research presented here are: W hat is represented in the mental lexicon? H ow is it represented? H ow is the speech signal parsed and information mapped onto the mental lexicon? In this paper we will address four issues within the framework of our Featurally Underspecified Lexicon model (FUL): (a) our assumptions concerning distinctive feature organization defined by phonological, perceptual and acoustic constraints; (b) specification of features in the mental lexicon (based on universal and language specific requirements); (c) extracting distinctive features from the signal; (d) mapping features from the signal to the lexicon. We claim that phonological features are extracted from the variable acoustic signal based on broad acoustic properties. A three-way matching algorithm maps these features onto highly abstract phonological mental representations. We provide evidence from synchronic phonological analyses, language change, psycholinguistic and neurolinguistic data.

When BEAT becomes HOUSE: Factors of word final /t/-deletion in German

The deletion and reduction of alveolar /t/ is a phenomenon that has been given considerable attention in the research on speech production and perception. Data have mainly be drawn from spoken language corpora, where a tight control over contributing factors of /t/-deletion is hardly possible. Here, we present a new way of creating a spoken language corpus adhering to some crucial factors we wanted to hold constant for the investigation of word-final /t/-deletion in German. German is especially interesting with regard to /t/-deletion due to its rich suffixal morphology, attributing morphological status to word-final /t/ in many paradigms. We focused on verb inflection and employed a verb form production task for creating a concise corpus of naturally spoken language in which we could control for factors previously established to affect /t/-deletion. We then determined the best estimators for /t/-productions (i.e. canonical, deleted, or reduced) in our corpus. The influence of extra-linguistic factors was comparable to previous studies. We suggest that our method of constructing a natural language corpus with carefully selected characteristics is a viable way for the examination of deletions and reductions during speech production. Furthermore, we found that the best predictor for non-canonical productions and deletions was the following phonological context.

Phonological and morphological constraints on German /t/-deletions

Abstract In running speech, deviations from canonical pronunciations are omnipresent. In extreme cases, segments such as /t/ are deleted altogether. On the other hand, /t/ may have morphological meaning, for instance, as marker of past tense in deal-t. Is it thus less likely that /t/ is deleted in deal t than in monomorphemic words, such as pain t ? Previous research suggests that morphological constraints on /t/-deletions indeed exist in English. However, in languages like German with richer morphology than English, the probability that /t/ with morphological information is deleted seems to be higher, particularly in contexts where /t/-deletion can allow for cluster simplification. Would such phonological effects override morphological constraints on /t/-deletion? To this end, a novel inflectional spoken verb form corpus was constructed in order to analyze the role of phonological and morphological influences on /t/-deletions. Final /t/ was part of suffixes in 2nd and 3rd person singular present tense verb forms (e.g., mach-s t ; mach- t ; ‘make’). Statistical analyses on /t/-deletions revealed that phonological context was highly predictive of /t/-deletions, particularly in cases where cluster simplifications were possible. This was true even in the 3rd person verb forms, where /t/ is morphologically more meaningful than in the 2nd person verb forms, and despite the fact that overall, /t/ was deleted less often in the 3rd than in the 2nd person. Altogether, this suggests that both phonology and morphology may constraint (or predict) /t/-deletions in German, but phonology can override morphological constraints in certain situations.

Infant cry reliability: Acoustic homogeneity of spontaneous cries and pain-induced cries

Infant cries can indicate certain developmental disorders and therefore may be suited for early diagnosis. An open research question is which type of crying (spontaneous, pain-induced) is best suited for infant cry analysis. For estimating the degree of consistency among single cries in an episode of crying, healthy infants were recorded and allocated to the four groups spontaneous cries, spontaneous non-distressed cries, pain-induced cries and pain-induced cries without the first cry after pain stimulus. 19 acoustic parameters were computed and statistically analyzed on their reliability with Krippendorffs Alpha. Krippendorffs Alpha values between 0.184 and 0.779 were reached over all groups. No significant differences between the cry groups were found. However, the non-distressed cries reached the highest alpha values in 16 out of 19 acoustic parameters by trend. The results show that the single cries within an infants episode of crying are not very reliable in general. For the cry types, the non-distressed cry is the one with the best reliability making it the favorite for infant cry analysis.

A classification model for infant cries with hearing impairment and unilateral cleft lip and palate.

Objective: The aim of the study was to examine the acoustic features of infant cries in order to find differences between infants with normal development, hearing impairment (HI) and unilateral cleft lip and palate (UCLP). The study focused on two issues: (1) are differences in acoustic cry parameters specific of a certain pathological development, and (2) do these differences allow for a reliable classification of infant cries? Patients and Methods: In total, 128 spontaneous cries of infants up to 12 months of age were recorded. The mean values of acoustic cry parameters were statistically analyzed by one-way analysis of variance and Sheffé post hoc tests. A C5.0 decision tree was trained to classify infant cries by their acoustic parameters. Results: Significant differences in cry duration, fundamental frequency, formants 2 and 4, intensity, jitter, shimmer and harmonics-to-noise ratio were found between the groups. The C5.0 decision tree was able to predict an infant’s membership to the healthy, UCLP or HI groups with a probability of 89.2%. Conclusion: We conclude that C5.0 decision trees are suited to reliably classify healthy infants as well as infants with HI and UCLP by acoustic parameters.

Comparison of Supervised-learning Models for Infant Cry Classification / Vergleich von Klassifikationsmodellen zur Säuglingsschreianalyse

Abstract Cries of infants can be seen as an indicator for several developmental diseases. Different types of classification algorithms have been used in the past to classify infant cries of healthy infants and those with developmental diseases. To determine the ability of classification models to discriminate between healthy infant cries and various cries of infants suffering from several diseases, a literature search for infant cry classification models was performed; 9 classification models were identified that have been used for infant cry classification in the past. These classification models, as well as 3 new approaches were applied to a reference dataset containing cries of healthy infants and cries of infants suffering from laryngomalacia, cleft lip and palate, hearing impairment, asphyxia and brain damage. Classification models were evaluated according to a rating schema, considering the aspects accuracy, degree of overfitting and conformability. Results indicate that many models have issues with accuracy and conformability. However, some of the models, like C5.0 decision trees and J48 classification trees provide promising results in infant cry classification for diagnostic purpose.

Do listeners recover "deleted" final /t/ in German?

Reduction and deletion processes occur regularly in conversational speech. A segment that is affected by such reduction and deletion processes in many Germanic languages (e.g., Dutch, English, German) is /t/. There are similarities concerning the factors that influence the likelihood of final /t/ to get deleted, such as segmental context. However, speakers of different languages differ with respect to the acoustic cues they leave in the speech signal when they delete final /t/. German speakers usually lengthen a preceding /s/ when they delete final /t/. This article investigates to what extent German listeners are able to reconstruct /t/ when they are presented with fragments of words where final /t/ has been deleted. It aims also at investigating whether the strategies that are used by German depend on the length of /s/, and therefore whether listeners are using language-specific cues. Results of a forced-choice segment detection task suggest that listeners are able to reconstruct deleted final /t/ in about 45% of the times. The length of /s/ plays some role in the reconstruction, however, it does not explain the behavior of German listeners completely.

Front & Back Matter

Each paper needs a structured abstract of 200 words in English. It should be structured as follows: – Objective – Patients and Methods – Results – Conclusion Footnotes: Avoid footnotes. Tables and illustrations: Tables and illustrations (both numbered in Arabic numerals) should be sent in separate files. Tables require a heading and figures a legend, also in a separate file. Due to technical reasons, figures with a screen background should not be submitted. When possible, group several illustrations in one block for reproduction (max. size 180 ue033 223 mm). Black and white halftone and color illustrations must have a final resolution of 300 dpi after scaling, line drawings one of 800–1,200 dpi. Color illustrations Online edition: Color illustrations are reproduced free of charge. In the print version, the illustrations are reproduced in black and white. Please avoid referring to the colors in the text and figure legends. Print edition: Up to 6 color illustrations per page can be integrated within the text at CHF 800.– per page. References: In the text identify references by Arabic numerals [in square brackets]. Material submitted for publication but not yet accepted should be noted as ‘unpublished data’ and not be included in the reference list. The list of references should include only those publications which are cited in the text. Do not alphabetize; number references in the order in which they are first mentioned in the text. The surnames of the authors followed by initials should be given. There should be no punctuation other than a comma to separate the authors. Preferably, please cite all authors. Abbreviate journal names according to the Index Medicus system. Also see International Committee of Medical Journal Editors: Uniform requirements for manuscripts submitted to biomedical journals (www.icmje.org). Examples (a) Papers published in periodicals: Chatel J-M, Bernard H, Orson FM: Isolation and characterization of two complete Ara h 2 isoforms cDNA. Int Arch Allergy Immunol 2003;131:14–18. (b) Papers published only with DOI numbers: Theoharides TC, Boucher W, Spear K: Serum interleukin-6 reflects disease severity and osteoporosis in mastocytosis patients. Int Arch Allergy Immunol DOI: 10.1159/000063858. (c) Monographs: Matthews DE, Farewell VT: Using and Understanding Medical Statistics, ed 3, revised. Basel, Karger, 1996. (d) Edited books: Hone SW, Smith RJH: Understanding inner ear physiology at the molecular level; in Cremers Cor WRJ, Smith RJH (eds): Genetic Hearing Impairment. Adv Otorhinolaryngol. Basel, Karger, 2002, vol 61, pp 1–10. Reference Management Software: Use of EndNote is recommended for easy management and formatting of citations and reference lists. Supplementary Material Supplementary material is restricted to additional data that are not necessary for the scientific integrity and conclusions of the paper. Please note that all supplementary files will undergo editorial review and should be submitted together with the original manuscript. The Editors reserve the right to limit the scope and length of the supplementary material. Supplementary material must meet production quality standards for Web publication without the need for any modification or editing. In general, supplementary files should not exceed 10 MB in size. All figures and tables should have titles and legends and all files should be supplied separately and named clearly. Acceptable files and formats are: Word or PDF files, Excel spreadsheets (only if the data cannot be converted properly to a PDF file), and video files (.mov, .avi, .mpeg). Digital Object Identifier (DOI) S. Karger Publishers supports DOIs as unique identifiers for articles. A DOI number will be printed on the title page of each article. DOIs can be useful in the future for identifying and citing articles published online without volume or issue information. More information can be found at www.doi.org Author’s ChoiceTM Karger’s Author’s ChoiceTM service broadens the reach of your article and gives all users worldwide free and full access for reading, downloading and printing at www. karger.com. The option is available for a one-time fee of CHF 3000.–, which is a permissible cost in grant allocation. More information can be found at www.karger.com/ authors_choice. NIH-Funded Research The U.S. National Institutes of Health (NIH) mandates under the NIH Public Access Policy that final, peer-reviewed manuscripts appear in its digital database within 12 months of the official publication date. As a service to authors, Karger submits the final version of your article on your behalf to PubMed Central. For those selecting our premium Author’s ChoiceTM service, we will send your article immediately upon publishing, accelerating the accessibility of your work without the usual embargo. More details on NIH’s Public Access Policy is available at http://publicaccess.nih.gov/policy.htm Self-Archiving Karger permits authors to archive their pre-prints (i.e. pre-refereeing) or post-prints (i.e. final draft post-refereeing) on their personal or institution’s servers, provided the following conditions are met: Articles may not be used for commercial purposes, must be linked to the publisher’s version, and must acknowledge the publisher’s copyright. Authors selecting Karger’s Author’s ChoiceTM feature, however, are also permitted to archive the final, published version of their article, which includes copyediting and design improvements as well as citation links. Page Charges There are no page charges for papers of 5 or fewer printed pages (including tables, illustrations and references). Each additional complete or partial page is charged to the author at CHF 650.–. The allotted size of a paper is equal to approx. 16 manuscript pages (double-spaced, A4, including tables, illustrations and references). Proofs Unless indicated otherwise, proofs are sent to the corresponding author and should be returned with the least possible delay. Alterations other than the correction of printer’s errors are charged to the author. Reprints Order forms and a price list are sent with the proofs. Orders submitted after the issue is printed are subject to considerably higher prices. Guidelines for Authors