Scholarlink

<p>Most accounts of English stress assume that it is determined by a combination of morphological and phonological factors. Two main factors have been proposed to determine the patterns of observed stress in English verbs with no productive morphology: whether the verb is prefixed, and the weight of the final syllable. This paper tests whether stress assignment to English verbs can be accounted for without recourse to abstract morphological properties and whether, instead, alleged morphological effects emerge as a function of the statistical patterning of stress among recurrent forms in the lexicon. The paper implements a discriminative learning network (NDL), a well-established computational learning model that is trained using error-driven learning (Rescorla &amp; Wagner, <CitationRef CitationID="CR47">1972</CitationRef>; Baayen et al., <CitationRef CitationID="CR6">2011</CitationRef>, see Arndt-Lappe et al., <CitationRef CitationID="CR1">2023</CitationRef> for application to stress). The model, which uses 1,948 verb types from Jones’ pronunciation dictionary (Roach et al., <CitationRef CitationID="CR48">2006</CitationRef>), works with phonotactic information only and does not have access to higher-level information about the morphological structure of words. The NDL model reaches the same level of accuracy as competing regression models that use the traditional predictors (syllable number, weight, morphological status of initial string). Closer inspection reveals that the NDL model brings about the alleged morphological and weight effects by a gradient association of phonotactics and stress. In the network, which has no explicit information about morphological structure, morphological effects emerge as a function of the recurrence of certain phonological strings. At a higher level, the results speak in favour of theories in which stress (and effects of weight and morphology) emerge from dynamically changing associations between representations in the lexicon.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Modeling the role of prefixation in determining stress assignment in English verbs

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月2日

查看全文
  • Ingo Plag,
  • Sabine Arndt-Lappe,
  • Aaron Seiler,
  • Quentin Dabouis

摘要

Most accounts of English stress assume that it is determined by a combination of morphological and phonological factors. Two main factors have been proposed to determine the patterns of observed stress in English verbs with no productive morphology: whether the verb is prefixed, and the weight of the final syllable. This paper tests whether stress assignment to English verbs can be accounted for without recourse to abstract morphological properties and whether, instead, alleged morphological effects emerge as a function of the statistical patterning of stress among recurrent forms in the lexicon. The paper implements a discriminative learning network (NDL), a well-established computational learning model that is trained using error-driven learning (Rescorla & Wagner, 1972; Baayen et al., 2011, see Arndt-Lappe et al., 2023 for application to stress). The model, which uses 1,948 verb types from Jones’ pronunciation dictionary (Roach et al., 2006), works with phonotactic information only and does not have access to higher-level information about the morphological structure of words. The NDL model reaches the same level of accuracy as competing regression models that use the traditional predictors (syllable number, weight, morphological status of initial string). Closer inspection reveals that the NDL model brings about the alleged morphological and weight effects by a gradient association of phonotactics and stress. In the network, which has no explicit information about morphological structure, morphological effects emerge as a function of the recurrence of certain phonological strings. At a higher level, the results speak in favour of theories in which stress (and effects of weight and morphology) emerge from dynamically changing associations between representations in the lexicon.