This paper introduces an experimental paradigm based on probabilistic evidence of the interaction between construction decisions in a parsed corpus. The approach is demonstrated using ICE-GB, a one million-word corpus of English. It finds an interaction between attributive adjective phrases in noun phrases with a noun head, such that the probability of adding adjective phrases falls successively. The same pattern is much weaker in adverbs preceding a verb phrase, implying this decline is not a universal phenomenon. Noun phrase postmodifying clauses exhibit a similar initial fall in the probability of successive clauses modifying the same NP head, and embedding clauses modifying new NP heads. Successive postmodification shows a secondary phenomenon of an increase in additive probability in longer sequences, apparently due to ‘templating’ effects. The author argues that these results can only be explained as cognitive and communicative natural phenomena acting on and within recursive grammar rules.

中文翻译：

---

研究重复语言生产决策的加性概率

本文介绍了一种基于已解析语料库中构造决策之间相互作用的概率证据的实验范式。该方法使用 ICE-GB 进行了演示，ICE-GB 是一个 100 万字的英语语料库。它发现名词短语中的定语形容词短语与名词中心之间的相互作用，使得添加形容词短语的概率依次下降。动词短语前面的副词中的相同模式要弱得多，这意味着这种下降不是普遍现象。名词短语后修饰子句在连续子句修改同一 NP 中心词和嵌入子句修改新 NP 中心词的概率方面表现出类似的初始下降。连续的后修改显示出较长序列中附加概率增加的次要现象，这显然是由于“模板”效应。