補語連詞(Complementizer)泛指連接子句補述語之詞(見英語中的that)，據類型語言學前人研究，表述說話之動詞於多種語言中均有語法化成補語連詞之現象。本文以橫向斷代方式分析廣東話「話」字，提出前述現象亦見於廣東話。「話」常見於各種表述言語或思想的謂語之後，一向被認為連動句式中的第二動詞，然而，本文以為「話」不一定用作動詞，於不同情況下，「話」之語意各異，其次語類制約亦有所不同，譬如：話1，意「說」[__ (PP) CP]/ [__ PP NP]；話2，意「責怪/指摘」[__ NP CP]；而話3，則為補語連詞，用作連接子句補述語[__ IP]。根據其表現之形式句法特徵，如：時體結構、論元選擇關係，本文論證了話3於作補語連詞時，形式句法上有別於說話動詞話1與及物動詞話2。 Complementizers are generally known as function words that introduce a clausal complement, like that in English, for instance (Radford 1997). In many languages, complementizers are re-analyzed from verba dicendi, or verbs of ‘saying’ (Lord 1976; Frajzyngier 1991; Hopper and Traugott 1993; Lord 1993). This paper argues for the existence of a complementizer re-analyzed from a verb of ‘saying’ in Cantonese by providing a synchronic analysis of waa6 . Waa6 has often been assumed to be a lexical verb in serial verb construction because of its following a ‘saying’ predicate or a cognitive predicate. However, this paper argues that waa6 is not always a verb, postulating that waa6 may have different meanings and subcategorizations in different situations, including waa61 meaning ‘say’ [__ (PP) CP] or [__ PP NP], the transitive verb waa62 meaning ‘blame/condemn’ [__ NP CP], and the complementizer waa63 selecting a clause [__ IP ]. This proposal is supported by different tests, such as aspect marking and argument selection, confirming that the complementizer waa63 formally exhibits different properties from that of the verbs waa61 and waa62.