本文將先說明平埔族群語言的各種文獻資料，從十七世紀荷蘭傳教士所做的記錄，到十九世紀歐美人士的零星資料，直到日治時期的有計畫的調查，都為我們留下了珍貴的語言資料。可惜有清一代二百多年間，並沒留下多少語言資料：一則中國傳統文人並不重視田野調查，二則他們也沒有適當可用的記錄工具。根據我們所能掌握的各種平埔族群語言資料，所有的平埔族群語言可以分為這三個支群：(一) 西部平埔族群：Taokas、Babuza、Papora、Hoanya、Thao，(二) 北部Kavalan、Basay 和南部Siraya 平埔族群，(三) 巴宰語（Pazih）。西部平埔族群，有二種共同的音變並不見於其他任何台灣南島語言：(1) *n 跟*ŋ 合併，(2) *s 跟*t 合併。其中Taokas、Babuza、Papora 跟Hoanya 又有另外二種共同的音變：(3) *k 消失，(4) 語詞尾的*-y消失。根據各種音韻演變規則，我們繪出這五種語言的親疏遠近樹圖。北部的Kavalan和Basay，南部的Siraya 語群，事實上也跟東部的Amis 都有一種共同的音變：(1) *j跟*n 合併，並不見於任何其他南島語言。其中Basay 跟Kavalan 又有二種共同的音變：(2) *j、*n 又進一步跟*N 合併，(3) *k 分化為k 和q，因此這兩種語言的關係最近。對於南部平埔族群的三種語言（Siraya, Taivuan, Makatau），我們也比較它們之間的異同和親疏遠近關係，並且發現荷蘭時期的聖經翻譯都是以Taivuan 語為主，而不是Siraya語。

Many Formosan languages were spoken in the Taiwan plains; most are now extinct. For data on them, we must rely on the writings of 17^(th) century Dutch missionaries, of early 20^(th) century Japanese field workers, and of a few Chinese or other Western observers from the 17^(th) through 19^(th) centuries. For Siraya in the southwestern plains and Favorlang on the central west coast, there is a fair amount of documentation. For all the other western plain languages, however, we have less than 400 lexical items. We still have informants for only three plain languages: Pazih, Thao, and Kavalan. For these, language data and linguistic information are much more abundant and reliable. All the plain languages can be classified into three main subgroups based on phonological innovations: (1) Western Plains: Taokas, Papora, Babuza, Hoanya, and Thao; (2) the so-called ＂East Formosan＂: Basay, Kavalan, Siraya, and Amis; and (3) Pazih. The five western plain languages share exclusively two phonological innovations: (1) merger of Proto-Austronesian *n and *ng, and (2) merger of *s and *t. The western plain languages, except for Thao, lost word-final *-s. Thao is, therefore, somewhat less closely related to the other four languages. Proto-Austronesian *j and *n have merged in the four East Formosan languages, a merger not found anywhere else. Basay and Kavalan share two innovations, indicating a closer relationship: (1) merger of *j, *n and *N, and (2) split of *k into k and q (*_a), so they are more closely related to each other. It is interesting to note not only that geographically adjacent languages may be closely related, as is the case with the five western plain languages, but also that geographically distant languages can likewise be closely related, as in the case of the four so-called ＂East Formosan＂ languages. In Li 2009, I examine the language data of the southwestern plains in greater detail and sort out the linguistic differences among Siraya, Taivuan, and Makatau. As it turns out, the Dutch biblical translations are based on Taivuan rather than Siraya.