This paper studies contrastive divergence (CD) learning algorithm and proposes a new algorithm for training restricted Boltzmann machines (RBMs). We derive that CD is a biased estimator of the log-likelihood gradient method and make an analysis of the bias. Meanwhile, we propose a new learning algorithm called average contrastive divergence (ACD) for training RBMs. It is an improved CD algorith...

A perception experiment with native German listeners provided evidence for the relevance of the tonal onglide in nuclear accents--the pitch movement leading towards the target on the accented syllable. Listeners were able to distinguish between two pragmatic meanings of a short phrase (given/non-contrastive and new/contrastive) using the tonal onglide as the sole acoustic cue. On the basis of t...

Diana Santos § Abstract In the present paper, the concept of vagueness will be taken up in detail, in order to make it more precise and to pin down its relevance for translation. In addition to view (some) translation choices as contrastive vagueness, I will argue that grammar too has to be contrasted, introducing the concepts of grammatical vagueness and contrastive grammatical vagueness. The ...

We present a method for producing a bird’s-eye view of statements that are expressed on Web pages on a given topic. This method aggregates statements that are relevant to the topic, and shows contradictory and contrastive relations among them. This view of contradictions and contrasts helps users acquire a top-down understanding of the topic. To realize this, we extract such statements and rela...

This paper focuses on the encoding of contrast in European Portuguese (EP), specifically by analysing contrastive parallelism structures, which seem to be crucial in the construction of the type of discourse that will be analysed: the argumentative discourse. Hence, my main goal is discussing how contrast is prosodically encoded in these structures and to relate the results with previous ones f...

This paper addresses a contrastive study between Neural Networks and Rough Sets on data classification. The experiments were carried out using the Iris database, of public domain, to evaluate the classification. The confusion matrix method was used to evaluate the performance of these classifiers. With these contrastive experiments, we investigated the capacity of each classifier for applicatio...

Unlike most other modern Bantu languages, Kinande has faithfully preserved the Proto-Bantu contrast between [-ATR] and [+ATR] high vowels (Hyman, 2002). As a result, while other Bantu languages typically have only two contrastive high vowels (i.e. high back and high front), Kinande has four: [-ATR] [i u] and [+ATR] i u ]. 1 These four high vowels, together with mid [e o] and low [a], are laid o...

Phonemic analysis, the process of identifying the contrastive sounds in a language, involves finding allophones; phonetic variants of those contrastive sounds. An algorithm for finding allophones (developed by Peperkamp et al.) is evaluated on consonants in the TIMIT acoustic phonetic transcripts. A novel phonetic filter based on the active articulator is introduced and has a higher recall than...

Orthographic similarities across languages provide a strong signal for probabilistic decipherment, especially for closely related language pairs. The existing decipherment models, however, are not wellsuited for exploiting these orthographic similarities. We propose a log-linear model with latent variables that incorporates orthographic similarity features. Maximum likelihood training is comput...