本論文基於五個賽德克語方言（德固達雅、都達、陶賽、德鹿谷、太魯閣）的語料，以比較方法重建原始賽德克語的音韻、形態句法及詞彙系統，並提出賽德克語的方言分群。

原始賽德克語共可重建出 18 個輔音（*p, *t, *k, *q, *b, *d, *c, *s, *x, *h, *g, *m, *n, *ŋ, *l, *r, *w, *y），4 個單元音（*a, *i, *u, *ə），以及 3 個雙元音（*ay, *aw, *uy）。而原始賽德克語在詞尾位置有較嚴格的限制，這亦造成賽德克語相當豐富的形態音韻變化。

形態句法方面，原始賽德克語區分主格 *ka 及屬格 *na 兩種名詞組標記，但在德固達雅以外的方言皆丟失屬格標記 *na。原始賽德克語共有五套人稱代詞，加上一套主格＋屬格的融合附著式人稱代詞。筆者亦重建了原始賽德克語的語態系統（即「焦點系統」），包含四個語態及三套不同詞綴。

原始賽德克語的一些詞彙重建，須仰賴如原始南島語或原始泰雅語的支持。而本文另外一個重要的發現是「特殊雙式詞現象」，即形式相近但不完全相同的近義詞或同義詞。這些「雙式詞」與泰雅語的性別語域系統高度相關。

筆者將賽德克語分為兩個主要分支，其中德固達雅方言獨立一支，其餘方言則屬都達－德鹿谷（太魯閣）群。都達－德鹿谷（太魯閣）群共享變化有：中央元音 *ə 的音變、一些零星音變、屬格標記 *na 的丟失、詞彙詞形或語意創新。都達－德鹿谷（太魯閣）群下則又分為都達群及德鹿谷（太魯閣）群，前者主要音韻證據為 *g、*w 的合流，後者則為 *c、*s 的合流以及隨後的 *t、*d 顎化。

This thesis reconstructs the phonology, morphosyntax, and lexicon of Proto-Seediq by applying the Comparative Method to data from five Seediq dialects (Tgdaya, Central Toda, Eastern Toda, Central Truku, and Eastern Truku) and proposes a refined subgrouping hypothesis for these varieties. The reconstructed Proto-Seediq phonemic inventory comprises 18 consonants (*p, *t, *k, *q, *b, *d, *c, *s, *x, *h, *g, *m, *n, *ŋ, *l, *r, *w, *y), 4 monophthongs (*a, *i, *u, *ə), and 3 diphthongs (*ay, *aw, *uy). Proto-Seediq had strict phonotactic constraints at word-final positions, which led to a rich array of morphophonological alternations.

In the morphosyntactic domain, Proto-Seediq distinguishes between a nominative marker *ka and a genitive marker *na. The language originally had five sets of personal pronouns plus an additional clitic set that combined the nominative and genitive functions. Moreover, A detailed reconstruction of the Proto-Seediq voice system reveals four voices and three distinct sets of voice affixes.

Lexical reconstruction identifies several items that require support from Proto-Austronesian or Proto-Atayal. A significant discovery is the occurrence of "special doublets"—synonymous forms with similar phonological shapes—that appear to be associated with the Atayal Gender Register System.

The Seediq dialects are classified into two main branches. The Tgdaya dialect constitutes an independent branch, while the remaining dialects form a Toda-Truku subgroup characterized by shared innovations involving historical *ə, sporadic sound changes, the loss of the genitive marker, and specific lexical shifts. Within Toda-Truku, the two Toda dialects merge *g and *w, while the two Truku dialects merge *c and *s and palatalize *t and *d.