Lexical splits: a novel perspective on the structure of words

许多语言通过词形的变化来表达语法意义的变化，这一过程被称为屈折。通常情况下，这是完全透明的：在像“talk”这样的动词后面加-s是第三人称单数现在时（s/he talks），加-ed是过去时（s/he talks）。但有时这个过程更复杂，比如“go”，它的过去式“went”似乎是一个完全不同的词。我们在这里说这个词被“分裂”成两个不同的词干。但分裂不仅仅包括不同的形式。例如，俄语动词在形式所代表的意义范畴中表现出现在-过去的分裂。现在时标记人称，而过去时标记性别，因此动词在使用人称和性别特征上是分开的。通过对分裂采取更抽象的看法，我们开始注意到在孤立地观察定义分裂的个别现象时不明显的模式。这种研究形态和形态句法不规则性的方法为许多不同的不规则性实例带来了希望，这些不规则性是由几种基本模式的组合产生的。我们假设一个工作类型学，它确定了四个二元标准，根据这些标准可以区分分裂：结构——分裂是否涉及单词的形式（例如go~went）或表达的意义类别（例如person vs . gender）？排列——这种分裂是否与已经在语言的句法、语义或音韵中发挥作用的区别相对应，或者它是任意的？3. 规律——分裂是影响同一类的所有单词，还是只影响几个？相关性——分裂对单词本身以外的其他东西有影响吗（例如，它如何与其他单词交互）？组合这些参数产生16个逻辑上可能的分裂类型。我们的初步研究表明，每一个都得到了证实，但分布是严重倾斜的，这表明在可能的分割配置上存在一些固有的限制。采用我们的方法的理论含义是令人兴奋的：语法结构的不同模型表明，不同类型的分裂应该占主导地位，并且只有某些分裂的组合（多重分裂）应该是可能的；我们的类型学框架允许我们以经验的方式测试一些关于屈折结构组织的广泛假设。此外，我们可以预测，由声音变化引起的分裂与由语法或语义变化引起的分裂会导致非常不同的分裂类型。我们提出的彻底的类型学研究旨在支持或反驳这些假设。在更普遍的层面上，我们对分裂的调查提供了一个新的视角：语法结构的描述通常是分开的，所以我们说语法、形态学和形态语法（两者的交集）。但是我们的方法跨越了这些区别，并提供了在更基本的层面上对共同的语言结构进行分类和分析的方法。鉴于不同语言之间的分裂的多样性，本研究将解决以下问题，以阐明是什么构成了一个可能的词：-世界语言中分裂类型的类型学，遗传和地理分布是什么？分裂类型和它们的表现方式之间有关联吗？- 16种类型的词汇分裂中哪一种可以同时发生？它们如何相互作用的限制是什么？（也就是说，多重分裂可能有哪些类型？）——多重分裂是如何产生的，它们随着时间的推移有多稳定？

Many languages express changes in grammatical meaning through changes in word form, a process known as inflection. Often this is perfectly transparent: adding -s to a verb like 'talk' gives the 3rd person singular present tense (s/he talks), and adding -ed gives the past tense (s/he talked). But sometimes the process is more involved, as with 'go', whose past tense 'went' seems to be a different word entirely. We say here that the word is "split" between two distinct stems. But splits can involve much more than just different forms. Russian verbs, for instance, display a present-past split in the categories of meaning that the forms represent. The present tense marks person, while the past tense marks gender, so that verbs are split in their use of the person and gender features.By taking a more abstract view of splits we start to notice patterns that are not apparent when looking at the individual phenomena which define splits in isolation. This way of approaching morphological and morphosyntactic irregularity holds open the promise that many diverse instances of irregularity result from combinations of a few basic patterns. We assume a working typology which identifies four binary criteria according to which splits may be distinguished:1. Structure - does the split involve the forms of the word (e.g. go~went) or the categories of meaning expressed (e.g. person versus gender)?2. Arrangement - does the split correspond to distinctions that already play a role in the syntax, semantics or phonology of the language, or is it arbitrary? 3. Regularity - does the split affect all words of the same class, or just a few?4. Relevance - does the split have an effect on anything other than the word itself (e.g. how it interacts with other words)?Combining these parameters yields sixteen logically-possible split types. Our pilot study shows that each is attested, but the distribution is dramatically skewed, suggesting there are some inherent constraints on possible configurations of splits.The theoretical implications of adopting our approach are exciting: different models of the architecture of grammar suggest that different types of split should predominate, and that only certain combinations of splits (multiple splits) should be possible; our typological framework allows us to test some of these wide-ranging hypotheses about the organization of inflectional structure in an empirical manner. Further, we would predict that splits which developed out of sound changes result in very different split types to those splits which were induced by changes in syntax or semantics. The thorough typological study we propose is designed to support or to disprove such hypotheses. At a more general level, our investigation of splits offers a novel perspective: accounts of grammatical structure have typically been compartmentalised, so we speak of syntax, of morphology, and of morphosyntax (the intersection of the two). But our approach cross-cuts these distinctions, and provides the means for classifying and analysing shared linguistic structures that operate at a more fundamental level.In light of the diversity of splits across languages, this research will address the following questions to elucidate what constitutes a possible word: - What is the typological, genetic and geographic distribution of split types across the world's languages?- Are there correlations between split types and the way they are manifested?- Which of the sixteen types of lexical splits can co-occur and what are the restrictions on how they interact with each other? (That is, what types of multiple splits can there be?)- How do multiple splits arise and how stable are they over time?