This thesis discusses how compound structures affect their parts of speech. The formation of compounds is divided into adnominal-head structure, adverbial-head structure, coordinate structure, verb-object structure, verb-resultative structure, subject-predicate structure, serial verbs structure, noun-measurer structure noun-localizer structure, numeral-measurer structure, serial measurers structure, preposition-object structure, functional morpheme structure, clitic-structurer, reduplicative-morphemes structure, affixation structure, all together sixteen types, and the word formations of the most commonly used ten thousand words are also labeled in the thesis. Statistics show that compounds formed by a certain word forming rules are of specific parts of speech. For instance, 86 percent of compounds with the adnominal-head structure are nouns. 96 percent with the verb-resultative-structure areverbs.94 percent with the serial verb structure are verbs. 100 percent with the noun-measurer structure are nouns. 75 percent with the verb-object structure are verbs. 62 percent with the adverbial-head structure are verbs. However, some compounds formed by other word forming rules are of varied parts of speech. For example, 51 percent of compounds with the coordinate structure are verbs, 28 percent are nouns and 15 percent are adjectives, 44 percent of compounds with the Subject-predicate structure are verbs, 25 percent are nouns, 15 percent are distinctive words and 6 percent are adjectives. Therefore, there is no correlation between word formation of compounds and their parts of speech. The affection of compound structures on their parts of speech is mainly embodied by the correlation between the grammatical features of the head and the parts of speech of the compounds. The head-feature percolation theory can be applied to explaining this phenomenon, namely the grammatical features of head-element of compounds can be percolated to the whole compounds and thus determine their parts of speech. Statistics show that 83 percent of compounds accord with head-feature percolation rule. But 17 percent do not. The head-feature shift rule is used to explain the phenomenon. The theory indicates that the grammatical-features of head will change when it percolates upwards. As a result, parts of speech of the whole compounds are different from those of head. We call the compounds that do not accord with head-feature percolation rule as endocentric-words and those that accord with head-feature shift rule as exocentric-words. There are two ways to form an exocentric-word. The first is regular shift, namely in the process of the combination of two word formation element, head-feature shift is regular. The second is irregular head-feature shift, which is caused by the evolvement of pats of speech and metonymy word formation.

本文探讨合成词的内部结构对合成词的词性的影响。本文把汉语合成词的构造方式分为定中、联合、述宾、状中、述补、主谓、连谓、名量、方位、数量、复量、介宾、虚构、截取、重叠、附加共16种类型，并对最常用的1万词作了构词方式标注。统计表明，有一些构词方式构成的合成词的词性主要集中在某一特定词类上,如定中式构词主要形成名词(86%)，述补式构词主要形成动词(96%)，连谓式构词主要形成动词(94%)，名量式构词全部形成名词(100%)，述宾式构词主要形成动词(75%)，状中式构词主要形成动词(62%)；而另一些构词方式形成的合成词的词性则较为分散，如联合式构词中有51%形成动词，28%形成名词，15%形成形容词，主谓式构词可以形成动词(44%)、名词(25%)、区别词(15%)和形容词(6%)。合成词的内部构造对合成词的词性的影响更主要地体现在核心成分与合成词整体词性的相关性上。我们用核心属性渗透理论来说明这种现象，即合成词中的核心成分的语法属性会渗透到整个合成词上，从而决定了合成词的词性。统计表明，83%的合成词都符合核心属性渗透规则。但仍有17%的合成词不符合核心属性渗透规则，我们用核心属性转移来说明这种现象，即合成词的核心成分的语法属性往上渗透时发生改变，使合成词整体的词性与核心成分的词性不同。我们把符合核心属性渗透规则的合成词叫向心词，把符合核心属性转移的合成词叫离心词。离心词的形成有两种方式，第一种是规则性转移，即两个构词成分组合时，核心属性转移是一种规则；第二种是非规则核心属性转移主要由词类演变和转喻构词引起。