This study examined the acquisition of numeral classifiers in 120 monolingual Japanese children. Previous research has argued that the complex semantic system underlying classifiers is late acquired. Thus, we set out to determine the age at which Japanese children are able to extend the semantic properties of classifiers to novel items/situations. Participants completed a comprehension task with a mouse-tracking extension and a production task with nonce and familiar items. While the comprehension results showed ceiling effects on familiar and nonce items, age significantly modulated a difference in accuracy between familiar and nonce items in the production task. The findings suggest that the acquisition of the underlying semantic system is acquired much earlier than previously argued. Previously attested issues with Japanese classifier production in young(er) children are more likely to reflect accessing difficulties than indexing the underlying grammatical competence of the classifier system.

In this work we investigate the internal syntax and semantics of quantifier phrases (QP) involving cardinal numerals. Concentrating on a set of previously documented puzzles concerning Case and number agreement within the numeral phrase in Russian, we argue that these agreement patterns follow naturally if one recognizes three structural layers in a numeral-based QP: the countability layer, the number layer and the quantificational layer. Our central theoretical claim is that the countability layer is implemented as a (pseudo-)classifier structure whose morphological manifestation obeys a principle of syntactic ‘visibility’. Our specific claim for Russian is that, diachronically, this countability layer has emerged as a result of the loss of the dual number in the course of transition between Old and Modern Russian. We strengthen our conclusions with psycholinguistic evidence from a sentence completion study that tests Russian speakers’ sensitivity to the countability layer.

Predicting the distribution of endangered species from habitat data is frequently perceived to be a useful technique. Models that predict the presence or absence of a species are normally judged by the number of prediction errors. These may be of two types: false positives and false negatives. Many of the prediction errors can be traced to ecological processes such as unsaturated habitat and species interactions. Consequently, if prediction errors are not placed in an ecological context the results of the model may be misleading. The simplest, and most widely used, measure of prediction accuracy is the number of correctly classified cases. There are other measures of prediction success that may be more appropriate. Strategies for assessing the causes and costs of these errors are discussed. A range of techniques for measuring error in presence/absence models, including some that are seldom used by ecologists (e.g. ROC plots and cost matrices), are described. A new approach to estimating prediction error, which is based on the spatial characteristics of the errors, is proposed. Thirteen recommendations are made to enable the objective selection of an error assessment technique for ecological presence/absence models.

The paper proposes and evaluates a binary network architecture and complementary training algorithm designed for pattern classification, with applications in a variety of engineering problems. The advantages of the system are that it can always converge on zero error for a set of unambiguous training patterns (if required), converges rapidly, and circumvents the issue of how many hidden neurons to incorporate in a network. The main benefit of the proposed system over Hamming networks, its main counterpart, is that it can group patterns into different classes along any boundary. The system is shown to outperform 1) the Hamming network for a character recognition problem where the images are subject to both position change and noise; and 2) a radial-Gaussian network in a truck-type classification problem. A variant of the system, where hidden neuron thresholds are set to zero, is shown to further improve performance if a comprehensive set of noise-free input patterns are available for training.