Transcript Prosodic marking of appositive relative clause types in

Prosodic marking of appositive
relative clause types in spoken
discourse: pragmatic and phonetic
analyses of a British English corpus
Cyril Auran & Rudy Loock
Laboratoire Savoirs, Textes,
Langage Université Lille 3 - CNRS
UMR 8163
Global project and methodology


Aim: relate discourse functions/structure and prosody
Study of Appositive Relative Clauses in English (ARCs, see 1)
(1) The students, who like linguistics, also like translation.

2-step methodology:



Discourse annotation: discourse function, information status of ARC and
MC, syntactic characteristics
Prosodic annotation: semi-automatic analysis of the corresponding
recordings using original scripts with Praat (cf. Boersma & Weenink
2006)
Corpora:



Aix-MARSEC (cf. Auran, Bouzon & Hirst 2004)
IViE (cf. Grabe & Post 2002)
ICE-GB (cf. Greenbaum 1996)
ARCs and their functions in discourse

Starting point:
Loock’s (2003, 2005, 2007) taxonomy of
ARCs and their discourse functions:
ARCs and their functions in discourse
ARCs and their functions in discourse

Examples :
(2) he was convinced # the battle # for the hearts # and minds of
the people # was being won # especially # among the
Ovambo # who form the majority # of SWAPO's support
(3) the first book he took from the library was Darwin's # Origin of
Species # which inspired him with the dream of becoming a
geologist
(4) Israelis # have sympathy and liking for Americans # which is
just as well # since the country is swarming # with transatlantic
visitors
ARCs and their functions in discourse

Differences in the hierarchisation of the informational contents
(ARC vs. MC):


Relevance/subjectivity: MC = foreground vs. ARC = background
Continuative: informational contents on the same level (narrative
dynamism traditionally restricted to independent clauses; cf.
Depraetere 1996)
=> Are continuatives independent clauses?
(cf. Ross 1967, Emonds 1979, McCawley 1982, Fabb 1990 among
others, who express this idea for ARCs as a whole.)
=> Prosodic investigation: are ARCs realized with the intonation
contour of independent clauses or parentheticals?
Prosodic analysis

Prosody as a macro-system:




tonal aspects (tone and intonation, in
relation with speech melody)
temporal aspects (unit durations and
speech rate)
intensity (one of the major correlates of
loudness)
voice quality (in relation with spectral
characteristics of the speech signal)
Prosodic analysis

Representation levels (Hirst et al. 2000):



acoustic level: physical characteristics of the
speech signal (F0, raw durations, dB)
phonetic level: retains only linguistically
significant elements (low-level physical
constraints factored out: MoMel modelling, ztransformed durations)
phonological levels (surface and deep): discrete
and abstract coding
Prosodic analysis
Prosodic analysis

2 prosodic dimensions:
Data extraction and analysis

Number of items per ARC type:






Relevance
37
Subjectivity
8
Continuative
1
Relevance/Subjectivity
Ambiguous continuative
Unidentified
4
2
2
Data extraction and analysis

Discourse parameters (5):





ARC type
Position (initial/medial/final)
Information status of antecedent
Information status of ARC
Phrastic status of antecedent
Data extraction and analysis
Prosodic parameters (48):


Tonal domain (32): ARC mean F0 (Htz + semitones or ST), ARC minimum F0 (Htz
+ ST), ARC maximum F0 (Htz + ST), ARC register span (Htz + ST), ARC onset (Htz
+ ST), ARC offset (Htz + ST), previous IU mean F0 (Htz + ST), previous IU
minimum F0 (Htz + ST), previous IU maximum F0 (Htz + ST), previous IU register
span (Htz + ST), previous IU offset (Htz + ST), next IU mean F0 (Htz + ST), next IU
minimum F0 (Htz + ST), next IU maximum F0 (Htz + ST), next IU register span (Htz
+ ST), next IU onset (Htz + ST), difference between previous IU offset and ARC
onset (ST), difference between ARC offset and next IU onset (ST)

Temporal domain (10): ARC duration (raw and normalised), previous IU duration
(raw and normalised), next IU duration (raw and normalised), difference between
previous IU normalised duration and ARC normalised duration, difference between
ARC normalised duration and next IU normalised duration, silence duration before
ARC, silence duration after ARC

Intensity domain (6): mean of ARC global intensity, standard deviation of ARC
global intensity, mean of previous IU global intensity, standard deviation of previous
IU global intensity, mean of next IU global intensity, standard deviation of next IU
global intensity
Results

All ARC types:
Results

Relevance vs.
Subjectivity:
Results
Results
Discussion

All ARCS  both typical and atypical
characteristics:



Register and intensity levels: lower than those
of surrounding units  typical of prosodic
parentheticals
Register and intensity spans + speech rate 
classical IUs realizing independent clauses
to be linked with the possibility for ARCs to:


have the syntactic behaviour and the semantic
interpretation of independent clauses
convey independent speech acts (cf. Emonds 1979,
McCawley 1982 among others)
Discussion

Relevance vs. Subjectivity ARCs:


Discourse discontinuity marking through high onset values for
both types
Subjectivity ARCs display even stronger discontinuity ↔
more important rupture with the discourse topic (cf. shift between
the referential and interpretative levels; see frame 1)



More peripheral information conveyed by subjectivity ARCs (nontopical comment or judgement)
Lower intensity level values for Subjectivity ARC: strategy
used by the speaker to induce the perception of intermediate
levels between otherwise discrete categories such as
continuity/discontinuity, subjectivity/objectivity, etc. (conflicting
prosodic characteristics for subjective episodes; cf. Di Cristo et
al. 2004)
Speech rate differences need further investigation (the great
majority of subjectivity ARCs qualifies sentential antecedents (cf.
Loock 2007): parameters difficult to separate)
Conclusion




Apparent correlation between discourse functions and
prosody
Some prosodic characteristics atypical of appositions in
general
Differences among ARC types: Subjectivity ARCs display
prosodic rupture cues, on a par with the peripheral
information which they convey
Further investigation:


The respective roles of the syntactic status of the antecedent
(nominal vs. sentential) and of the ARC type, particularly with
relation to speech rate, need to be closely analysed.
Prosodic characteristics of Continuative ARCs: typical of
subordinate or main clauses?
Problem: availability of acoustically exploitable
unscripted/spontaneous data