How do speakers choose a word for production? One general idea is that they accumulate evidence until one word emerges as an acceptable option. According to this noncompetitive approach, the speed of lexical selection should depend on how strongly the strongest word is activated, independent of any alternatives. In this case, the selection process may be modeled as activation toward a simple threshold, whereby the first candidate to reach that threshold will be selected via a winner-take-all mechanism (e.g., Oppenheim, Dell, & Schwartz, 2010, Simulation 6). A competitive extension takes this idea further, suggesting that speakers accumulate evidence until one word emerges as clearly better than any alternative, for instance by surpassing a relative threshold (Roelofs, 1992, 2018). Although this competitive extension attracted unquestioning support for several decades, even serving as the basis of one of the most prominent theories of language production (Levelt, Roelofs, & Meyer, 1999), its necessity has more recently become the subject of robust debate on both empirical and computational grounds. As a step toward resolving this debate, Nozari and Hepner (2018) suggest that their hypothesized conflict monitoring mechanism (Nozari, Dell, & Schwartz, 2011) could provide a basis for assessing and possibly resolving task-incompatible conflict in lexical selection, essentially by scaling a relative threshold according to some function of baseline conflict and task demands. Perhaps the clearest evidence that strongly activated alternatives can delay lexical selection comes from picture-word interference (Glaser & Glaser, 1989; Schriefers, Meyer, & Levelt, 1990; Starreveld & Heij, 1995), a paradigm in which participants are directed to name pictures using pre-specified names (e.g., “dog”) while suppressing responses to other stimuli (e.g., the visually superimposed name of another item in the response set, “cat”). When the picture is semantically related to the distractor, correct productions of its intended name are typically slower than when it is not. Delays are typically assumed to reflect competition during lexical selection, where activation from the distractor somehow combines with activation from the normal retrieval process, making it harder for the target’s activation to surpass the distractor’s. However, because the interference is less consistent than one might expect (e.g., Miozzo & Caramazza, 2003), and because the experimental paradigm is rather complex and contrived, alternative explanations have proliferated (e.g., Dell’Acqua, Job, Peressotti, & Pascali, 2007; Dhooge & Hartsuiker, 2010 et passim; Mahon, Costa, Peterson, Vargas, & Caramazza, 2007 et passim; Navarrete & Mahon, 2013 et passim) including the idea that such behavioral results may simply reflect ad hoc monitoring processes. With such mechanisms in dispute, converging evidence from simpler paradigms, with less obvious manipulations, becomes more important. Cumulative semantic interference is a behavioral effect where naming a picture of a dog as “dog” makes speakers persistently slower and more error-prone when subsequently attempting to name a picture of a cat as “cat”. This interference occurs even in simple picture naming, so it is tempting to conclude that it provides important converging evidence for the competitive extension (e.g., Howard, Nickels, Coltheart, & ColeVirtue, 2006). However, after demonstrating that a simple model of lexical retrieval and incremental

中文翻译：

---

按需词汇竞争

演讲者如何选择一个词进行生产？一个普遍的想法是他们积累证据，直到出现一个词作为可接受的选择。根据这种非竞争性方法，词汇选择的速度应该取决于最强词被激活的强度，与任何替代词无关。在这种情况下，可以将选择过程建模为朝着简单阈值的激活，从而将通过赢家通吃机制选择第一个达到该阈值的候选人（例如，Oppenheim、Dell 和 Schwartz，2010 年，模拟 6 ）。竞争性扩展进一步扩展了这一想法，表明说话者会积累证据，直到出现一个词明显优于任何替代词，例如超过相对阈值 (Roelofs, 1992, 2018)。尽管这种竞争性扩展几十年来吸引了无可置疑的支持，甚至作为最突出的语言生成理论之一的基础（Levelt，Roelofs，& Meyer，1999），但它的必要性最近已成为关于这两种理论的激烈辩论的主题。经验和计算基础。作为解决这场争论的一个步骤，Nozari 和 Hepner（2018 年）建议他们假设的冲突监控机制（Nozari、Dell 和 Schwartz，2011 年）可以为评估和可能解决词汇选择中的任务不兼容冲突提供基础，主要是通过根据基线冲突和任务需求的某些功能来缩放相对阈值。也许最明显的证据表明强烈激活的替代品可以延迟词汇选择来自图片词干扰（Glaser & 格拉泽，1989 年；Schriefers, Meyer, & Levelt, 1990; Starreveld & Heij, 1995)，一种范例，其中参与者被引导使用预先指定的名称（例如，“狗”）命名图片，同时抑制对其他刺激的反应（例如，反应集中另一个项目的视觉叠加名称， “猫”）。当图片在语义上与干扰项相关时，其预期名称的正确产生通常比没有时更慢。通常假设延迟反映了词汇选择期间的竞争，其中干扰者的激活以某种方式与正常检索过程的激活相结合，使目标的激活更难超过干扰者的激活。然而，由于干扰不如人们预期的那么一致（例如，Miozzo & Caramazza，2003），并且由于实验范式相当复杂和人为，替代解释激增（例如，Dell'Acqua, Job, Peressotti, & Pascali, 2007; Dhooge & Hartsuiker, 2010 et passim; Mahon, Costa, Peterson, Vargas, & Caramazza, 2007 et passim; Navarrete & Mahon, 2013 et passim) 包括这种行为结果可能只是反映临时监控过程的想法。由于这些机制存在争议，从更简单的范式中汇聚证据，并进行不太明显的操纵，变得更加重要。累积语义干扰是一种行为效应，其中将狗的图片命名为“狗”会使说话者在随后尝试将猫的图片命名为“猫”时持续变慢且更容易出错。即使在简单的图片命名中也会出现这种干扰，因此很容易得出结论，它为竞争性扩展提供了重要的收敛证据（例如，Howard、Nickels、Coltheart 和 ColeVirtue，2006 年）。然而，在证明了一个简单的词法检索和增量模型之后