Combining Gender and Classifiers in Natural Language

在许多语言中，名词被系统地分成几类。在性别系统中，如在意大利语中，这是基于性别的：所有名词都被视为阳性或阴性-即使是那些意义与生物性别无关的名词。具有分类器系统的语言采取了完全不同的方法。在这里，分类是基于名词固有的细粒度意义区别，包括形状，功能，安排，地点或时间间隔。这样一种语言是Kilivila（特罗布里恩群岛），它至少有177个不同的类别。在大多数情况下，一种语言将只有一个系统或另一个，但在一些有趣的情况下，我们发现两个系统在一起。这种根本不同的范畴系统如何在一种语言中相互作用，这是一个尚未得到认真考虑的问题，但它潜在地揭示了语义、形态和认知范畴之间的相互作用。我们提出的研究问题是：（1）当两个范畴化系统共存时，它们是否在形态（形式）或句法（句子结构）层面上相互作用？(2)在这些语言中，哪些原则决定了名词如何被归类到它们各自的类别中？(3)两个范畴化系统的存在如何影响语法系统的其余部分？驱动这些问题的是一个更大的问题：混合语言的性别和分类系统与单系统语言相似，还是它们有自己的特殊属性？初步调查表明，双重分类系统确实超过了其部分的总和。典型的性别系统和典型的分类系统出现在非常不同的语言类型中：性别是屈折语言的特征，而分类器是缺乏复杂屈折形态的语言的特征。自然，当这两种倾向同时出现在一种语言中时，这两种相反的倾向之间就会发生冲突，从而导致与预期的偏差。因此，任何具有双重范畴的语言在某种意义上都是非规范的。这些语言分布有限，主要分布在大洋洲和美洲。我们将集中讨论以下语言群体：巴布亚语（特别是米安语和蒂多雷语）、维托托语（米拉尼亚语）、玛雅语（阿卡泰克语）、阿拉瓦坎语（塔利亚纳语）、图卡诺语（雷图阿拉语）和澳大利亚语（恩甘吉提梅里语）。第一个研究问题需要对这些关键语言进行深入分析，辅之以基于世界语言结构地图集的更广泛样本。第二个研究问题涉及类别本身的性质和相互作用。例如，《勉》中的名词同时按性（区分阳性、阴性、中性1类和中性2类）和量词（区分束类、长类和覆盖类）分类。这两个系统部分重叠，但我们仍然缺乏一种精确的方法来描述每个系统的相对贡献以及它们相互作用的方式。这就是我们提出的严格计算模型可以使这个复杂的因素网络变得清晰的地方。该方法还允许我们基于特定的规则系统来评估不同候选分析的相对成功程度。第三个研究问题探讨了语法的其他部分如何处理双重范畴化，特别是在参考追踪中。例如，在一种只有性别的语言中，一个代词指的是以前使用过的名词，它的性别会与之一致。但在一种具有双重分类的语言中，一系列额外的选择开始发挥作用。为了揭示所涉及的多个因素，我们将进行语料库研究，将我们的目标语言之一与代表整个类型范围的其他三种语言进行比较：一种只有性别，一种只有分类器，一种没有。

In many languages, nouns are systematically categorized into groups. In a gender system, as in Italian, this is based on sex: all nouns are treated as either masculine or feminine -- even those nouns whose meaning has nothing to do with biological sex. Quite a different approach is taken by languages with a classifier system. Here categorization is based on fine-grained meaning distinctions inherent to the noun, involving shape, function, arrangement, place or time interval. Such a language is Kilivila (Trobriand Islands), which has at least 177 distinct categories. For the most part a language will have only one system or the other, but in a few interesting cases we find both systems together. How such fundamentally different systems of categorization interact within a single language is a question which has not yet been seriously considered, but it potentially uncovers a great deal about the interaction of semantics, morphology, and cognitive categories in general. The research questions we pose are:(1) Where two categorization systems co-occur, do they interact with each other at the level of morphology (form) or syntax (sentence structure)? (2) Which principles determine how nouns are grouped into their respective categories in such languages?(3) How is the rest of the grammatical system affected by the presence of two categorization systems? Driving each of these questions is a larger one: are the gender and classifier systems of mixed languages similar to those found in one-system languages, or do they have special properties of their own?Preliminary investigations suggest that dual categorization systems are indeed more than the sum of their parts. Canonical gender systems and canonical classifier systems occur in very different types of languages: gender is characteristic of inflecting languages, while classifiers are characteristic of languages that lack complex inflectional morphology. Naturally, where the two co-occur in a single language, there is a conflict between these opposing tendencies, leading to a deviation from expectations. Thus any language with dual categorization is in some sense non-canonical. Such languages are limited in their distribution, largely in Oceania and the Americas. We will concentrate on the following language groupings: Papuan, (specifically Mian and Tidore), Witotoan (Miraña), Mayan (Akatek), Arawakan (Tariana), Tucanoan (Retuarã), and Australian (Ngan'gityemerri). The first research question requires in-depth analyses of these key languages, complemented by a broader sample based on the World Atlas of Language Structures. The second research question addresses the nature and interaction of the categories themselves. For example, in Mian nouns are categorized simultaneously by gender (distinguishing masculine, feminine, neuter type 1 and neuter type 2) and classifiers (distinguishing such categories as bundles, long things and covering things). These two systems partly overlap, but we still lack a precise way of characterizing the relative contribution of each system and the way in which they interact. This is where the rigorous computational model that we propose can bring clarity to this complex network of factors. This method further allows us to evaluate the relative degree of success of different candidate analyses based on particular rule systems. The third research question explores the way dual categorization is treated by the rest of the grammar, in particular in reference tracking. For example, in a language with gender alone, a pronoun referring back to a noun previously used will agree with it in gender. But in a language with dual categorization, a whole range of additional choices come into play. To uncover the multiple factors involved, we will conduct a corpus study comparing one of our target languages to three other languages representing the full typological range: one with gender only, one with classifiers only, and one with neither.