Search

Abstract: A limited number of studies have investigated language in Huntington’s disease (HD). These have generally reported abnormalities in rule-governed (grammatical) aspects of language, in both syntax and morphology. Several studies of verbal inflectional morphology in English and French have reported evidence of over-active rule processing, such as over-suffixation errors (e.g., walkeded) and over-regularizations (e.g., digged). Here we extend the investigation to noun inflection in Hungarian, a Finno-Ugric agglutinative language with complex morphology, and to genetically proven pre-symptomatic Huntington’s disease (pre-HD). Although individuals with pre-HD have no clinical, motor or cognitive symptoms, the underlying pathology may already have begun, and thus sensitive behavioral measures might reveal already-present impairments. Indeed, in a Hungarian morphology production task, pre-HD patients made both over-suffixation and over-regularization errors. The findings suggest the generality of over-active rule processing in both HD and pre-HD, across languages from different families with different morphological systems, and for both verbal and noun inflection. Because the neuropathology in pre-HD appears to be largely restricted to the caudate nucleus and related structures, the findings further implicate these structures in language, and in rule-processing in particular. Finally, the need for effective treatments in HD, which will likely depend in part on the ability to sensitively measure early changes in the disease, suggests the possibility that inflectional morphology, and perhaps other language measures, may provide useful diagnostic, tracking, and therapeutic tools for assessing and treating early degeneration in pre-HD and HD. [Copyright &y& Elsevier]

Previous studies have found that priming a target like cerr-o (“hill”) with an unmasked stem-homograph (e.g., cerr-ar, “to close”) or with an unmasked allomorph of a stem-homograph (e.g., cierr-a, “he/she/it closes”) has an inhibitory effect on target recognition. We report evidence from stem-homograph priming studies that implicate both morphological parsing and lexical selection processes at the lemma level. We argue that stem-homograph inhibition is the product of lexical selection mechanisms that are engaged just when a fully ambiguous stimulus (here, an inflectional stem) must be assigned one of its viable interpretations in order to afford conscious perception of lexical identity. [Copyright &y& Elsevier]

Arabic is a highly inflected language and a morpho-syntactically complex language with many differences compared to several languages that are heavily studied. It may thus require good pre-processing as it presents significant challenges for Natural Language Processing (NLP), specifically for Machine Translation (MT). This paper aims to examine how Statistical Machine Translation (SMT) can be improved using rule-based pre-processing and language analysis. We describe a hybrid translation approach coupling an Arabic–French statistical machine translation system using the Moses decoder with additional morphological rules that reduce the morphology of the source language (Arabic) to a level that makes it closer to that of the target language (French). Moreover, we introduce additional swapping rules for a structural matching between the source language and the target language. Two structural changes involving the positions of the pronouns and verbs in both the source and target languages have been attempted. The results show an improvement in the quality of translation and a gain in terms of BLEU score after introducing a pre-processing scheme for Arabic and applying these rules based on morphological variations and verb re-ordering (VS into SV constructions) in the source language (Arabic) according to their positions in the target language (French). Furthermore, a learning curve shows the improvement in terms on BLEU score under scarce- and large-resources conditions. The proposed approach is completed without increasing the amount of training data or radically changing the algorithms that can affect the translation or training engines. [ABSTRACT FROM AUTHOR]

Abstract: Do the production and interpretation of patterns of plural forms in noun-noun compounds reveal the workings of innate constraints that govern morphological processing? The results of previous studies on compounding have been taken to support a number of important theoretical claims: first, that there are fundamental differences in the way that children and adults learn and process regular and irregular plurals, second, that these differences reflect formal constraints that govern the way the way regular and irregular plurals are processed in language, and third, that these constraints are unlikely to be the product of learning. In a series of seven experiments, we critically assess the evidence that is cited in support of these arguments. The results of our experiments provide little support for the idea that substantively different factors govern the patterns of acquisition, production and interpretation patterns of regular and irregular plural forms in compounds. Once frequency differences between regular and irregular plurals are accounted for, we find no evidence of any qualitative difference in the patterns of interpretation and production of regular and irregular plural nouns in compounds, in either adults or children. Accordingly, we suggest that the pattern of acquisition of both regular and irregular plurals in compounds is consistent with a simple account, in which children learn the conventions that govern plural compounding using evidence that is readily available in the distribution patterns of adult speech. [ABSTRACT FROM AUTHOR]

• Class-based language models are investigated for very large vocabulary speech recognition tasks in two morphologically rich languages, Finnish and Estonian. • We evaluate different morphological constraints in class training in detail, providing new insights to the class inference. Improved classifications were reached, especially with a smaller text corpus, bearing importance for both less-resourced tasks and in language model adaptation. • The class-based models are evaluated both in the first recognition pass and in a second rescoring pass using class-based NNLMs. We show that better accuracy can be reached compared to a subword-based unlimited vocabulary recognizer using a combination of class-based and subword based models. • The software for class inference and decoding is available as open source with a permissive licence to ease the reproducibility of the results. We study class-based n-gram and neural network language models for very large vocabulary speech recognition of two morphologically rich languages: Finnish and Estonian. Due to morphological processes such as derivation, inflection and compounding, the models need to be trained with vocabulary sizes of several millions of word types. Class-based language modelling is in this case a powerful approach to alleviate the data sparsity and reduce the computational load. For a very large vocabulary, bigram statistics may not be an optimal way to derive the classes. We thus study utilizing the output of a morphological analyzer to achieve efficient word classes. We show that efficient classes can be learned by refining the morphological classes to smaller equivalence classes using merging, splitting and exchange procedures with suitable constraints. This type of classification can improve the results, particularly when language model training data is not very large. We also extend the previous analyses by rescoring the hypotheses obtained from a very large vocabulary recognizer using class-based neural network language models. We show that despite the fixed vocabulary, carefully constructed classes for word-based language models can in some cases result in lower error rates than subword-based unlimited vocabulary language models. [ABSTRACT FROM AUTHOR]