Efficient communication in word formation: How syntactic and lexical surprisal jointly shape English conversion over the past centuryEfficient communication in word formation: How syntactic and lexical surprisal jointly shape English conversion over the past century

Abstract: Conversion degree, defined as the proportion of converted-form use out of total lemma use, has notably increased in English over the past century. For instance, track had a conversion degree of only 5% in the 1920s (95% noun, 5% verb) but rose to 35% by the 2010s (65% noun, 35% verb). This historical trend presents a communicative paradox: it favors speaker economy (articulatory ease via form reuse) seemingly at the expense of listener economy (increased ambiguity). The present study investigates this paradox through the framework of efficient communication, which predicts that ambiguous forms like conversion should be licensed in predictable contexts. We tested this prediction by analyzing 639 noun-to-verb and 443 verb-to-noun conversion words from the 1920s to the 2010s. Generalized linear mixed models were fitted with surprisal estimates from three architectures: local 5-grams, syntactic POS-grams, and broad-context GPT-2. Results indicate that the historical rise in conversion degree is accompanied by increasingly predictable contexts, suggesting that contextual predictability compensates for the absence of explicit morphological marking. We further demonstrate that syntactic and lexical surprisal exert independent inhibitory effects on conversion degree. Crucially, their positive interaction indicates that the constraining effect of lexical surprisal is stronger when syntactic surprisal is low and attenuated when syntactic surprisal is high, consistent with a resource competition account. These findings portray the evolution of English conversion not as random drift, but as a dynamic optimization reflecting a negotiation between communicative efficiency and cognitive resource limitations, offering a novel information-theoretic perspective on language change.

摘要： 转类度指转类形式的使用量占词元总使用量的比例。过去一百年间，英语的转类度显著上升。例如，track 在 1920 年代的转类度仅为 5%（名词 95%，动词 5%），到 2010 年代已升至 35%（名词 65%，动词 35%）。这一历时趋势构成了一个交际悖论：它有利于说话者经济性（复用同一形式，发音更省力），却似乎以牺牲听话者经济性（歧义增加）为代价。本研究在高效交际框架下考察这一悖论。该框架预测，转类这类有歧义的形式应当出现在可预测的语境中。为检验这一预测，我们分析了 1920 年代至 2010 年代的 639 个名转动词和 443 个动转名词，并用三种架构估算的惊异度拟合广义线性混合模型：局部 5-gram、句法词性 n 元（POS-gram）和宽语境 GPT-2。结果表明，转类度的历时上升伴随着语境可预测性的不断提高，说明语境可预测性弥补了显性形态标记的缺失。我们还发现，句法惊异度和词汇惊异度各自独立地抑制转类度。尤为关键的是，二者存在正向交互作用：句法惊异度低时，词汇惊异度的制约作用更强；句法惊异度高时，这一作用减弱，这与资源竞争解释相一致。上述发现表明，英语转类的演变并非随机漂移，而是一种动态优化，体现了交际效率与认知资源限制之间的权衡，为语言变化研究提供了新的信息论视角。