A Review of A Grammar of Speech 39
apparent in pre-planned discourse such as news-reading, bible-reading and
anecdotes. He states, however, that it is not impossible to fi nd phonological
paragraphing in more spontaneous forms of discourse.
Wichmann (2000: 108) produces corpus evidence^18 which both favours
and disfavours the theory of supradeclination: the decline of pitch across
paratones in English. While the pitch level of the initial onset of the fi rst
sentence tended to be highest and the pitch level of the initial onset of
the fi nal sentence tended to be the lowest, the pitch levels of the initial
onsets of the intervening sentences did not exhibit a gradual decline.
She attributes this lack of supradeclination within paratones to both the
information structure inside sentences and the rhetorical relations between
sentences (ibid. 118). She states:
A shift from a ‘new’ topic to ‘additional but related’ information seems to
generate a step down in pitch, while a shift from ‘background’ information
(e.g. elaboration or explanation) to ‘new’ or ‘additional’ information
prompts a step up. Only a shift between sentences of equal rhetorical
value (‘new’ – ‘new’ or ‘addition’ – ‘addition’) does not appear to have a
systematic effect on scaling.
Wichmann’s view is close to that of Brazil in that she recognizes three
communicatively signifi cant values of key and attributes values similar to
those proposed by Brazil.
In an investigation of the phonetic clues hearers use in order to identify
the ‘spoken equivalent of sentences’ Nakajima and Allen (1993) measured
the height of F0 peaks between the utterance units within speaker turns in
a simulated conversation and found that that a high F0 reset signalled topic
shift, a mid F0 reset signalled topic continuation and a low F0 reset
signalled elaboration.
Scholars (e.g. Liberman 1975, Pierrehumbert 1980) who work within the
autosegmental/metrical tradition do not recognize phonological units
such as paratones. Instead they notate speech as a linear string of high and
low tones on stressed syllables. The phonetic scaling of an individual
high (H) or low (L) tone depends upon a variety of local factors such
as emphasis and information structure. H tones signal that the items
made salient are to be treated as new to the discourse (Pierrehumbert and
Hirschberg 1990: 289). L tones mark items made salient which are not
intended to alter the hearer’s existing beliefs (ibid. 291). Within an utter-
ance H tones at the beginning are higher than H tones at the end. Finality
is signalled by a fi nal prominent syllable which attracts the lowest pitch