Kwak'wala is often transcribed with unaddressed assumptions about the basic word-unit. A syntactition will regard the grammatical-word as the word constituent, whereas a phonologist will see the phonological-word as most relevant. As a morphologist, I am caught between these two definitions of the 'word'. Since Kwak'wala has always been a predominantly oral language, I assume the spoken-word as most relevant 'word' constituent for Kwak'wala speakers, with phonological criteria for defining its boundaries. These boundaries can be isolated through three tests: 1) the stress domain, 2) occurrences of pauses, and 3) self-repair resets. These tests show that most Kwak'wala clitics are integrated with their phonological-word hosts. However the definite article clitic and the case marking clitics display more independence than all others by allowing pauses and self-repair to occur between them and their hosts. Still other clitics, when uttered in a successive string, also create the same independence from their hosts.