本文探討以日語為母語之幼兒在語言習得過程中所犯的語誤，研究焦點放在兩歲幼兒在詞序上所產生的語誤。研究的結果指出，幼兒在二詞階段 (two-word stage) 與電報式言語階段 (telegraphic speech) 時，即已具備母語中詞組及構詞的相關知識；然而其對詞組性質及詞組互動的認識，則尚在成長建構之中。透過對幼兒的詞序語誤進行分析，本文也修正了近五十年來學界對兩歲幼兒語言習得的認知與研究。

In this paper, we will revisit the grammar of very early utterances that young children produce at around age 2 under the linguistic theory of the past 50 years. We will argue that very young children in the stage of two words and the telegraphic speech, who already know the head-complement relationship, the basic word order, the morphological properties regarding whether or not the language is agglutinative, have a specific system ofφ-feature agreement then, and the detailed mechanism of the operation of Merge and Labeling proposed in the recent Minimalist thesis can be considered to be still under construction. Children are free to project unattested hypotheses based on their innately endowed knowledge, and in the course of language acquisition, there are intriguing cases where English-acquiring children fail to match their hypothesis to the input of the target language, thereby producing ‘erroneous’ scrambled sentences. We argue in this paper that the scrambled sentences that very young English-acquiring children produce (e.g., Powers 2000) can naturally be explained by assuming the child-specific system of φ-feature agreement, or the system whose spirit is consistent with AGR/TNS Omission Model (ATOM) (Schütze and Wexler 1996) for the child grammar, and Saito (2016) for the adult grammar.

非限定主句；類非限定主句；語言習得之二詞階段；微言主義；變序；普遍語法

Root Infinitives；Root Infinitive analogues；two-word utterances；Minimalist program；scrambling；Universal Grammar