Refine
Year of publication
Document Type
- Conference Proceeding (38) (remove)
Has Fulltext
- yes (38)
Keywords
- Computerlinguistik (6)
- Deutsch (6)
- Mehrsprachigkeit (5)
- Artikulatorische Phonetik (4)
- Automatische Sprachanalyse (4)
- Korpus <Linguistik> (4)
- Historische Phonetik (3)
- Kempelen, Wolfgang von (3)
- Maschinelles Lernen (3)
- Natürliche Sprache (3)
Publicationstate
- Postprint (38) (remove)
Reviewstate
- Peer-Review (22)
- (Verlags)-Lektorat (4)
- Peer-review (3)
Publisher
This paper is concerned with a novel methodology for generating phonetic questions used in tree-based state tying for speech recognition. In order to implement a speech recognition system, language-dependent knowledge which goes beyond annotated material is usually required. The approach presented here generates phonetic questions for decision trees are based on a feature table that summarizes the articulatory characteristics of each sound. On the one hand, this method allows better language-specific triphone models to be defined given only a feature-table as linguistic input. On the other hand, the feature-table approach facilitates efficient definition of triphone models for other languages since again only a feature table for this language is required. The approach is exemplified with speech recognition systems for English and Thai.
MRI data of German vowels and consonants was acquired for 9 speakers. In this paper tongue contours for the vowels were analyzed using the three-mode factor analysis technique PARAFAC. After some difficulties, probably related to what constitutes an adequate speaker sample for this three-mode technique to work, a stable two-factor solution was extracted that explained about 90% of the variance. Factor 1 roughly captured the dimension low back to high front; Factor 2 that from mid front to high back. These factors are compared with earlier models based on PARAFAC. These analyses were based on midsagittal contours; the paper concludes by illustrating from coronal and axial sections how non-midline information could be incorporated into this approach.
Distributional models of word use constitute an indispensable tool in corpus based lexicological research for discovering paradigmatic relations and syntagmatic patterns (Belica et al. 2010). Recently, word embeddings (Mikolov et al. 2013) have revived the field by allowing to construct and analyze distributional models on very large corpora. This is accomplished by reducing the very high dimensionality of word cooccurrence contexts, the size of the vocabulary, to few dimensions, such as 100-200. However, word use and meaning can vary widely along dimensions such as domain, register, and time, and word embeddings tend to represent only the most prevalent meaning. In this paper we thus construct domain specific word embeddings to allow for systematically analyzing variations in word use. Moreover, we also demonstrate how to reconstruct domain specific co-occurrence contexts from the dense word embeddings.
HMMs are the dominating technique used in speech recognition today since they perform well in overall phone recognition. In this paper, we show the comparison of HMM methods and machine learning techniques, such as neural networks, decision trees and ensemble classifiers with boosting and bagging in the task of articulatory-acoustic feature classification. The experimental results show that HMM methods work well for the classification of such features as vocalic. However, decision tree and bagging outperform HMMs for the fricative classification task since the data skewness is much higher than for the feature vocalic classification task. This demonstrates that HMMs do not perform as well as decision trees and bagging in highly skewed data settings.
Precise multimodal studies require precise synchronisation between audio and video signals. However, raw audio and audio from video recordings can be out of sync for several reasons. In order to re-synchronise them, a dynamic programming (DP) approach is presented here. Traditionally, DP is performed on the rectangular distance matrix comparing each value in signal A with each value in signal B. Previous work limited the search space using for example the Sakoe Chiba Band (Sakoe and Chiba, 1978). However, the overall space of the distance matrix remains identical. Here, a tunnel matrix and its according DP-algorithm are presented. The matrix contains merely the computed distance of two signals to a pre-specified bandwidth and the computational cost is equally reduced. An example implementation demonstrates the functionality on artificial data and on data from real audio and video recordings.
Beyond the stars: exploiting free-text user reviews to improve the accuracy of movie recommendations
(2009)
In this paper we show that the extraction of opinions from free-text reviews can improve the accuracy of movie recommendations. We present three approaches to extract movie aspects as opinion targets and use them as features for the collaborative filtering. Each of these approaches requires different amounts of manual interaction. We collected a data set of reviews with corresponding ordinal (star) ratings of several thousand movies to evaluate the different features for the collaborative filtering. We employ a state-of-the-art collaborative filtering engine for the recommendations during our evaluation and compare the performance with and without using the features representing user preferences mined from the free-text reviews provided by the users. The opinion mining based features perform significantly better than the baseline, which is based on star ratings and genre information only.
In this paper the authors briefly outline editing functions which use methods from computational linguistics and take the structures of natural languages into consideration. Such functions could reduce errors and better support writers in realizing their communicative goals. However, linguistic methods have limits, and there are various aspects software developers have to take into account to avoid creating a solution looking for a problem: Language-aware functions could be powerful tools for writers, but writers must not be forced to adapt to their tools.
In the context of the HyTex project, our goal is to convert a corpus into a hypertext, basing conversion strategies on annotations which explicitly mark up the text-grammatical structures and relations between text segments. Domain-specific knowledge is represented in the form of a knowledge net, using topic maps. We use XML as an interchange format. In this paper, we focus on a declarative rule language designed to express conversion strategies in terms of text-grammatical structures and hypertext results. The strategies can be formulated in a concise formal syntax which is independend of the markup, and which can be transformed automatically into executable program code.
Wolfgang von Kempelen's book "The Mechanism of Human Speech" from 1791 is a famous milestone in the history of speech communication research. It has an enormous relevance for the phonetic sciences and it marks an important turning point for the development of the (mechanical) speech synthesis. So far no English version of this work was available, which excludes many interested researchers. Access to the original versions in German and French is restricted for various reasons. For example the blackletter script of the German version is troublesome for most of today's readers. We report here on a new edition of Kempelen's book which unites a better readable German version and its English translation. It will now also be in a searchable electronic format and has been enriched with many commentaries, which aid in the understanding of details of the late 18th century that are little known or unknown to many researchers today.