Focus, phrase length, and the distribution of phrase-initial rises in French
James German 1, Mariapaola D'Imperio2
1 Laboratoire Parole et Langage (UMR 6057 CNRS), Aix-en-Provence, France
2 Aix-Marseille I & Laboratoire Parole et Langage (UMR 6057 CNRS), Aix-en-Provence, France
This study addresses the relationship between information
structure and intonation in French. More specifically, it tests
whether phrase-initial rises (LHi) are associated with the left
edge of contrastively focused constituents in wh-
interrogatives. Since LHi distribution has also been correlated
with length, the study further examines the relative
contribution of constraints operating at two distinct levels:
information structure and phonological structure. The results
show that each set of constraints makes an independent
contribution to the occurrence of LHi, but with no interaction.
In other words, LHi is more likely to occur at the left edge of a
contrastive focus domain, and more likely to occur in longer
phrases, though phrase length does not influence the extent to
which LHi marks focus. The findings of this study represent
the first quantitative assessment of focus realization in French
in a non-corrective context, and establish a previously
undocumented link between LHi and discourse-level meaning.
Index Terms: French, intonation, focus, information structure
In French, utterances are parsed into a series of phrasal units,
which minimally consist of a prosodic word (typically a
content word plus zero or more associated functional items)
and define the domain of stress and tone assignment. It is
generally agreed that these units are obligatorily marked by a
prominent F0 rise near the end of the phrase, and optionally
include an additional F0 rise near the beginning of the phrase.
In this paper, we analyze the structure of these phrasal
units within the autosegmental framework of  and . In
that framework, the structure of accentual phrases (APs) and
their constituent early and late rises can be described by a
sequence of high and low tones. The late rise, or final accent,
consists of a low plus a high tone (denoted by LH*) and is
associated to a metrically strong syllable that is typically the
last full syllable of the phrase (excluding word-final schwa).
The early rise, or initial rise, also involves a sequence of low
and high tones (denoted by LHi) but is doubly associated to a
syllable near the beginning of the phrase (typically either the
first or second syllable of the first content word occurring in
the phrase) as well as to the left edge of the AP . Since
LH* is obligatory, a minimal AP consists of the LH*
sequence, while a maximal AP would consist of the sequence
LHiLH* (see Figure 1). Other well-formed and attested
sequences include LLH*, LHiH* and HiLH*.
It is generally accepted for many languages that
utterances may be partitioned into information that is part of a
background, in the sense of being shared by interlocutors, and
information that is part of a focus, in the sense of being
presented by the speaker as either novel or contrastive. In
French, the extent to which such a partition is reflected in the
prosodic signal has not been fully resolved. Previous work in
this regard emphasizes the role of contrast in emphatic or
corrective contexts. Typically, the focused element is said to
be marked by a single rising-falling contour, which is distinct
in various respects from either the early or late rise, and which
may be realized on either final or non-final syllables in the
phrase. For , for example, this is the accent d'insistance.
For  and , it is the focus accent, or Hf, which they show
experimentally to consist of a rise that is both higher and
aligned later than a typical AP-final accent in a broad focus
context. The region following the contrastive element, or the
post-focal region, is often characterized by the absence of
prominent pitch movements [1, 4, 5, 6]. The pre-focal region,
by comparison, retains the early and late rises associated with
rhythmically regular phrasing and in this sense is not
markedly different from the broad focus pattern, though there
is some evidence that this region may be associated with a
compressed pitch range [1, 5, 7] and a reduced number of
phrase boundaries .  explores the marking of focus in
question-answer contexts. While no quantitative findings are
presented, the study suggests an overall tendency for a focus
to be phrased as a single unit. The pre-focal region also shows
this tendency, though a syntactic subject is generally phrased
separately regardless of its relationship to the focus.
Figure 1: F0 trace, spectrogram and intonation labels for one
speaker's production of a target sentence (PP-focus long).
It is generally agreed that the transition from the focal
region to the post-focal region is readily identified in French
on the basis of post-focal deaccenting (for polar interrogatives,
a high plateau). The extent to which the left edge of a focus
region is prosodically marked, however, remains an open
question. In general, the pre-focal and focal regions are
regularly phrased into APs (subject to conditions of rhythm
and syntax). Therefore, the transition between them cannot be
reliably identified on the basis of LH* alone. Given the
tendency for foci to be phrased as a unit , however, an
extended stretch of unaccented material might in principle
serve to mark a kind of minimum boundary for the leftward
extension of a focused region. Assuming probabilistic
constraints on the maximum size of a phrase, however, such
outputs are predicted to be dispreferred for larger foci except
in very fast speech. It remains to be tested, therefore, whether
other intonational features might serve to mark this boundary.
Widespread agreement on the prosodic hierarchy of
French is limited to the distinction between a single lower
level unit and a single higher level unit (e.g., the accentual
hal-00541970, version 1 - 1 Dec 2010
Author manuscript, published in "International Conference on Speech Prosody, Chicago : United States (2010)"
phrase and intonational phrase, respectively). A number of
recent studies, however, have raised the possibility that an
intermediate level of phrasing exists for French, which may be
recruited for demarcative functions at the level of syntactic or
pragmatic constituency.  and , for example, provide
evidence for an intermediate level of phrasing at the boundary
between a syntactic subject and a VP, whose domain is
marked by systematic tonal scaling relationships between the
F0 peaks of its constituent AP-final accents. Moreover, 
show that the distribution of initial rises (initial accent or IA in
that work) is correlated with syntactic constituency for
contrasts involving attachment ambiguities. Thus, for the
sequence les bagatelles et les balivernes saugrenues ('the
crazy trifles and nonsense') the adjective saugrenues ('crazy')
may be interpreted as modifying only the second NP (1a), or
as modifying both the first and second NPs (1b).
(1) a. [les bagatelles] [et les balivernes saugrenues]
b. [les bagatelles et les balivernes] [saugrenues]
Higher rates of initial rises were observed on the adjective
when taking wide scope (1b), suggesting that this feature is
recruited to mark the stronger syntactic boundary occurring
between the second NP and the adjective in that case.
Correspondingly, more initial rises were observed at the left
edge of the second NP when the strong syntactic boundary
occurs at the NP1-NP2 juncture (1a). The authors suggest that
this correlation may reflect a more general tendency for initial
rises to mark an intermediate level of phrasing, which may in
turn be sensitive to constituency in other domains.
If 's suggestion is correct, then there is reason to
suspect that the distribution of initial rises may be linked to
information structural boundaries. Indeed,  found
evidence that initial rises are more likely to occur in APs
whose left edge coincides with the left edge of a focus region.
The current study addresses this issue directly through a
controlled production experiment involving wh-interrogatives
with focus domains of varying sizes. In contrast to earlier
studies, the materials used in this study were not designed to
evoke corrective or emphatic meaning. Thus, the feature
being targeted in our study is not the accent d'insistance of 
or Hf of  and , which generally occurs alone within an
AP, but the initial rise (LHi), whose distribution is limited to
phrases which also bear a final accent (LH*).
We hence assume a notion of contrastive focus in which
a set of two or more closely related utterances are partitioned
into information that is either shared or not shared between
them. We further assume that such interpretations are likely
to occur in contexts in which sets of utterances related in this
way are uttered in close sequence. Contrastive focus is
typically discussed in connection with declarative utterances,
though theories of focus interpretation typically extend readily
to other types of utterances including interrogatives [13, 14].
For the current study, it is sufficient to assume that a series of
question-answer pairs like those in (2) is predicted to induce a
focus partitioning as indicated by F-marking.
(2) A: Who wrote [the music]F for South Pacific?
A: Who wrote [the lyrics]F for South Pacific?
We now turn to the distributional properties of initial
rises. The initial rise has often been characterized as optional,
since APs are frequently observed without it. Known
predictors of its occurrence, however, include speaking rate,
number of syllables in the phrase [1, 2, 15], and
morphological status (e.g., content versus function word).
 found a simultaneous association to syntactic
constituency and number of syllables in the phrase, which
they take as evidence that the demarcative function of initial
rises is first and foremost a prosodic one. Building on their
model, then, our study seeks to characterize the distribution of
initial rises along two dimensions. In addition to testing for an
association with the left edge of focus, we simultaneously
varied the length, in number of syllables, of the word units
being phrased into APs.  show that in English,
probabilistic constraints on the accentability of certain part-of-
speech categories lead to a reduced tendency for nuclear
accent placement to encode focus. If it is the case that phrase
length represents a similarly probabilistic constraint on the
realization of initial rises, such that phrases with fewer
syllables are less likely to include them, then one conceivable
outcome is that phrase length and focus will interact as
predictors of initial rise. When a phrase is very small, for
example, then initial rise may be so dispreferred that focus
will have little effect on the likelihood of its occurrence. In
the extreme case, initial rises on one-syllable phrases may not
be well-formed. By comparison, longer phrases may actually
require initial rises for rhythmic or structural reasons , and
thus the occurrence of initial rise may be insensitive to focus
in a different way.
In summary, our study seeks to address the following
questions. First, to what extent is the left edge of a contrastive
focus region systematically marked in French, and what role
does initial rise play in this encoding? Second, what are the
factors affecting the distribution of initial rises? Are they truly
'optional' as is often assumed, or can the variability in their
distribution be explained by constituency at other levels of
description? Finally, to what extent do constraints from
distinct levels of the grammar (i.e., prosody and information
structure) interact? Do they make equal contributions to the
likelihood that initial rise will occur, or do constraints from
one level interfere with the ability of initial rise to encode
constituency at another level?
The materials in our study consisted of 24 pairs of subject-
extracted wh-interrogative sentences beginning with the word
qui 'who'. Crucially, these included a direct object consisting
of a noun phrase followed by a prepositional phrase. The
noun phrase and the prepositional phrase each consisted of a
single functional item (either an article, a preposition or a
prepositional contraction) followed by a noun. Each target
sentence then occurred in one of two versions, according to
the constituent length of the second noun (underlined): either
short (2 syllables) as in (3a) or long (4 syllables) as in (3b).
Target sentences in a pair were identical in all other respects.
(3) a. Qui a commandé le merlan aux navets ce soir?
'Who ordered the whiting with turnips this evening?'
b. Qui a commandé le merlan aux macadamias ce soir?
'Who ordered the whiting with macadamias this evening?'
Target sentences were produced as the second of a series
of three information-seeking questions. Thus, as outlined in
Section 1, specific patterns of focus were induced by
manipulating the scope of correspondence between the target
sentence and the questions occurring before and after it. In
the DO-focus condition (4), for example, the domain of the
hal-00541970, version 1 - 1 Dec 2010
focus is predicted to be the entire direct object (DO), since the
surrounding context suggests that the DO is the relevant point
of contrast. In the PP-focus condition (5), the domain of focus
is limited to the prepositional phrase. Thus, each item set
occurred in four conditions: DO-Focus/short, DO-Focus/long,
PP-Focus/short and PP-Focus/long.
(4) DO-Focus (short)
i. Qui a commandé l'entrecôte ce soir?
'Who ordered the rib steak this evening?'
ii. Qui a commandé [le merlan aux navets]F ce soir?
'Who ordered the whiting with turnips this evening?'
iii. Qui a commandé les gambas ce soir?
'Who ordered the shrimp this evening?'
(5) PP-Focus (short)
i. Qui a commandé le merlan à la sauce citron ce soir?
'Who ordered the whiting with lemon sauce this evening?'
ii. Qui a commandé le merlan [aux navets]F ce soir?
'Who ordered the whiting with turnips this evening?'
iii. Qui a commandé le merlan aux câpres ce soir?
'Who ordered the whiting with capers this evening?'
The resulting 96 experimental items were divided among four
versions balanced for condition. 10 filler items, consisting of
a variety of question types (including polar, object-extracted
wh-, etc.) were included as distractors.
Participants were seated in a sound-attenuated booth with a
conversational partner (native speaker of French). They were
asked to pose each question to their partner "as naturally as
possible" and then record the response on a sheet of paper.
Preceding each item, a short context was read aloud by the
participant. This served both to fix the interpretation of the
lexical items comprising the targets and to help the
participants situate themselves in the speaking role.
Recorded target sentences were aligned at the level of
words, syllables and phonemes using EasyAlign . Target
sentences were labeled by the authors for the presence and
syllable association of the following features:
• AP-final accent on the lexical verb (V)
• Initial rise on the first NP (I1)
• AP-final accent on the first NP (F1)
• Initial rise on the second NP, within the PP (I2)
• AP-final accent on the PP (F2)
There is generally good agreement regarding the
identification of AP-final accents in French. It is widely
assumed, for example, that they are characterized by a
prominent F0 rise that reaches a peak at or near the end of the
final full syllable of a content word. Typically, syllables
bearing a final accent also have longer rhyme durations .
Identification of final rises was therefore based primarily on a
combination of pitch track inspection and coder assessment of
overall prominence. Only prominent rises in word-final
positions were counted as final accents, allowing for the
possibility that F0 peaks may occur just after the end of the
There is less agreement regarding the identification of
initial rises. Since initial rises are typically not associated
with rhyme lengthening, overall prominence is not a reliable
measure. Furthermore, the pitch movements associated with
initial rises may be less marked than for final accents, and
may therefore be difficult to distinguish from perturbances in
the rises or falls of adjacent final accents. We therefore took a
conservative approach, counting as initial rises only those F0
movements that bore clear evidence of an independent high
tonal event. In general, this means that qualifying features
included a rise from an identifiable low target, followed by a
trough preceding an ascent to the peak of a following final
accent (i.e., the characteristic LHiLH* pattern of  and ).
Rises preceding a final accent were counted as initial rises if
they minimally reached a peak or plateau early in the
associated syllable and showed evidence of a subsequent fall
that could not be explained by microprosodic perturbances.
Rises following a final accent of the previous AP were
counted as initial rises if the peak of the preceding final accent
was reached within the AP-final syllable, and either (i) a
second peak was reached or (ii) there was a stable plateau
extending from the peak of a preceding final accent more than
50% of the way into the following syllable. Initial rises are
typically associated with the first or second syllable of the first
content word within an AP , though it has been informally
noted that they may also occur on a function word preceding
the first content word. Initial rises meeting the above criteria
were observed in all three positions (i.e., function word, first
syllable, and second syllable of the content word).
Two male and two female speakers participated in this
task either voluntarily or for pay. All were first language
native speakers of a continental variety of French.
Speakers produced a wide range of patterns in the target
region, though a final accent occurred on V in approximately
90% of all productions, and a final accent occurred on F2 in
95% of productions. Thus, most of the variability concerned
the status of I1, F1 and I2. Overall, initial rises occurred at I1
in 21% of all productions, and 75% of these occurred on the
first syllable of the noun, while 25% occurred on the function
word. Final accents occurred on F1 in 84% of productions.
Finally, initial rises occurred on I2 in 39% of all productions.
Of these, 38% occurred on the first syllable of the noun, 51%
occurred on the second syllable of the noun, and 11%
occurred on the function word.
Figure 2: Percentage of targets bearing initial rise at I1
by Focus and Constituent Length.
Initial rises were more frequent at I1 when this coincided
with a focus left edge (DO-Foc: 32%, PP-Foc: 18%) and when
the second noun was short (2-syllable: 37%, 4-syllable: 15%),
though as Figure 2 suggests, the effect of Focus appears to be
larger when the second noun is long. A mixed effects linear
regression analysis (treating Focus and Length as fixed effects
and items as random effects) found Length, but not Focus, to
be significantly correlated with the likelihood of initial rise at
I1 (p<0.05) with no interaction.
hal-00541970, version 1 - 1 Dec 2010
At I2, initial rises were more frequent when a focus edge
occurred there, but in comparison to I1, they were also more
frequent when the second noun was long (2-syllable: 30%, 4-
syllable: 83%). A second regression analysis found both
effects to be significant at the p<0.05 and p<0.1 levels,
respectively. As with I1, there was no significant interaction
of Focus and Length, suggesting that each factor makes an
independent contribution to the likelihood of an initial rise.
PPfocus(narrow) DOfocus(wi de)
Figure 3: Percentage of targets bearing initial rise at I2
by Focus and Constituent Length.
The above findings establish a new link between initial rises
in French and the marking of information structure.
Specifically, they suggest that initial rises may serve to mark
the left edge of contrastive focus regions, given that this
transition is otherwise obscured by the occurrence of
rhythmically regular phrasing into APs. Initial rises were not
reliably linked in our study to focus edge marking at the verb-
DO boundary, though the trend in the data suggests that this
association may be supported by further data collection.
Alternatively, this could be due to other factors. First of all,
with respect to the presence of a focus edge, the experimental
manipulation differed for this position (I1) as compared to the
DO-internal position (I2). At I2, the experiment manipulated
whether the relevant position was at a focus left edge or
internal to the focus, while at I1, the experiment manipulated
whether that location was at a focus left edge or external to it
(i.e., in the pre-focal domain). Second, the two positions
differed in terms of the syntactic boundary that was present. It
remains to be tested, for example, whether the juncture
between a verb and a direct object enforces special constraints
on prosodic structure that may impact the amount of
variability that is permitted with regard to early rises.
An additional issue concerns the detection of initial rises.
In this study, a conservative method was used that relied
primarily on inspection of the F0 contour. As noted earlier,
however, the effect of initial rises on the F0 contour may be
concealed by the effects of adjacent rises. As a result, some
initial rises may have gone undetected. Indeed, this is
consistent with the authors' impressions: there were many
cases where other acoustic properties of the syllable suggested
the presence of an initial rise, but the associated position could
not be labeled as such since it did not meet the established
criteria. An additional correlate of initial rise appears to be
lengthening of the onset of the associated syllable. Thus a
more sensitive coding system would take into account the
effect of initial rise in multiple acoustic dimensions.
By another measure, the reliability of our coding system
is supported by the fact that a significant correlation was
found between the prosodic features being measured, and a
known predictor of initial rise (i.e., phrase length). In other
words, there is good evidence in our data that the prosodic
features our study links with focus left edges correspond to the
same category treated elsewhere as initial rises.
Our study addresses the association between initial rises
and focus in production, but additional research is needed to
determine whether this association is utilized in perception.
Finally, while it has been suggested that the demarcative
function for initial rise is mediated by an intermediate level of
phrasing, this study does not address this issue directly.
Additional research is needed to determine whether
independent evidence for intermediate phrase constituency
can be linked to the type of alignment with information
structure discussed here.
 Jun, S.-A., and Fougeron, C., "A phonological model of French
intonation. In A. Botinis [Ed], Intonation: Analysis, modeling
and technology. Kluwer, 2000.
 Jun, S.-A., and Fougeron, C., "Realizations of accentual phrase
in French". Probus 14: 147-172, 2002.
 Welby, P., "The realization of early and late rises in French
intonation: A production study", Proceedings of Speech Prosody
2002, 695-698, 2002.
 Di Cristo, A., "Intonation in French". In D. Hirst, & A. Di Cristo
[Eds], Intonation systems: A survey of twenty languages, 195–
218. Cambridge, 1996.
 Touati, P. Structures Prosodiques du Suédois et du Français.
Lund University Press, 1987.
 Clech-Darbon, A., Rebuschi, G. and Rialland, A., "Are There
Cleft Sentences in French?" In G. Rebuschi & L. Tuller [Eds],
The Grammar of Focus, 83–118. Benjamins, 1998.
 Dohen, M. and Loevenbruck, H., "Pre-focal rephrasing, focal
enhancement and post-focal deaccentuation in French",
Proceedings of the 8th International Conference on Spoken
Language Processing, 785-788, 2004.
 Féry, C., "Intonation of focus in French". In C. Féry & W.
Sternefeld [Eds], Audiatur Vox Sapientes: A Festschrift for
Arnim von Stechow, 153-181. Akademi Verlag, 2001.
 Michelas, A. and D’Imperio, M., “Durational cues and prosodic
phrasing in French: evidence for the intermediate phrase”,
Proceedings of Speech Prosody 2010, this volume.
 D’Imperio, M. and Michelas, A. “Mapping syntax onto prosodic
structure: the intermediate phrase in French”. Proceedings of
Speech prosody 2010, this volume.
 Astésano, C., Bard, E. & Turk, A., "Structural influences on
initial accent placement in French". Language and Speech, 50(3),
 German, J., and D'Imperio, M., "Information structure and the
alignment of phrasal features in French qu-interrogatives".
Poster presented at Interfaces Discours et Prosodie, Paris, 2009.
 Rooth, M. "A theory of focus interpretation". Natural Language
Semantics, 1: 75-116, 1992.
 Schwarzschild, R. "Givenness, AvoidF and other constraints on
the placement of accent". Natural Language Semantics 7: 141-
 Welby, P., "French intonational structure: Evidence from tonal
alignment". Journal of Phonetics, 34: 343-371, 2006.
 German, J., Pierrehumbert, J. and Kaufmann, S., "Evidence for
phonological constraints on nuclear accent placement".
Language 82(1), 2006.
 Shattuck-Hufnagel, S. and Turk, A. "A prosody tutorial for
investigators of auditory sentence processing." Journal of
Psycholinguistic Research 25(2), 193-247, 1996.
 Goldman, Jean-Phillippe. EasyAlign v26.08.07. Downloaded
4/16/2009: <http://latlcui.unige.ch/phonetique/>, 2009.
 Pasdeloup, V., “Modèle de règles rythmiques du français
appliqué à la synthèse de parole”. Doctoral thesis, Université de
hal-00541970, version 1 - 1 Dec 2010