Most languages are primarily spoken, with only a minority of languages
or speaker communities developing a writing system. Despite the
primacy of the spoken mode, most work on grammar has relied on the
study of written representations, with typology being no
exception. The information on grammatical structures needed for a
typological study is usually extracted from transcribed examples in
reference grammars or taken from a typological database such as WALS
or Grambank, which are based on written resources of languages
themselves. Even typological or cross linguistic studies that use
corpus data often have to rely on written records, given that most
cross-linguistic corpus collections (such as the Universal Dependency
treebanks) are based on compilations of written data. As a
consequence, studies rarely include phonological, phonetic and
prosodic information, even though we know that e.g. phonetic reduction
or prosodic patterns are important factors for language change and
grammaticalization. Yet, quantitative typology as a field has largely
ignored this dimension when analyzing and explaining grammatical
structures and their cross-linguistic distributions. In this talk, I
will show three typological case studies to illustrate the importance
of the spoken signal when anayzing grammar. I use cross-linguistic
spontaneous speech data from the DoReCo collection (Seifart, Paschen &
Stave 2024) to examine (i) frequency effects in verbal person marking,
(ii) the association of different types of syntactic boundaries and
silent pauses, and (iii) to test for potential phonetic differences
between affixes and clitics.