On the emergence of the French Accentual Phrase Текст научной статьи по специальности «Языкознание и литературоведение»

ФИЛОЛОГИЧЕСКИЕ НАУКИ

УДК 81

Sang-Cheol Ahn1, Philippe Martin2

On the Emergence of the French Accentual Phrase

'M.K. Ammosov North-Eastern Federal University, Yakutsk, Russia 2Universite Paris Diderot, Paris, France

Abstract. This paper reanalyzes the basic tonal patterns in French within the domain of Accentual Phrase (AP henceforth) and how the prosodic constraints for AP interact with those for the Intonational Phrase (IP). The level of Accentual Phrase (AP) and the higher level, the Intonational Phrase (IP) have been proposed in the earlier studies (Di Cristo & Hirst 1993, 1996, Delais-Roussarie 1996, Di Cristo 1999, Jun & Fougeron 2000, 2002, Post 2000, Rolland & Lrevenbruck 2002, Gussenhoven 2004, etc.). Employing the cyclic/stratal framework of Optimality Theory (Kiparsky 2000, Ahn 2008, etc.), we therefore suggest a path to overcome those shortcomings. In addition, we show that the possible indeterminacy problem occurring in the traditional autosegmental account can be easily solved by the interaction of constraints. Furthermore, by categorizing the phonetic factors as constraints, we can incorporate the phonetic details within our phonological categorization.

Keywords: Accentual Phrase, Intonational Phrase, French, prosody, Optimality Theory, constraints, basic tonal pattern, rhythm.

DOI 10.25587/SVFU.2019.71.31945

Ан Сан Чоль1, Филипп Мартин2

К образованию акцентной фразы в французском языке

1СВФУ им. М.К. Аммосова, г. Якутск, Россия 2Университет Париж Дидро, Париж, Франция

Аннотация. Эта статья повторно исследует основные тональные рисунки французского языка в области акцентной фразы (далее AP) и то, как просодические ограничения для AP взаимодействуют с интонационной фразой (IP). Акцентная фраза (AP) и интонационная фраза (IP) были изучены в более ранних исследованиях Di Cristo & Hirst 1993, 1996, Delais-Roussarie 1996, Di Cristo 1999, Jun & Fougeron 2000, 2002 , Post 2000, Rolland & Lrevenbruck 2002, Gussenhoven 2004 и др. Используя циклический/стратальный каркас теории оптимальности (Kiparsky 2000,

SANG-CHEOL AHN - PhD, Professor of M.K. Ammosov North-Eastern Federal University. АН Сан Чоль - доктор филологических наук, профессор СВФУ им. М.К. Аммосова. E-mail: scahn.mail@gmail.com

PHILIPPE MARTIN - PhD, Professor, Université Paris Diderot.

ФИЛИПП Мартин - доктор филологических наук, профессор, Университет Париж Дидро E-mail: philippe.martin@utoronto.ca

Ahn 2008 и т.д.), мы предлагаем собственный способ решения проблем, которые существуют в настоящее время. Кроме того, мы показываем, что возможная проблема неопределенности, возникающая в традиционной автосегментной фонологии, может быть легко решена с помощью взаимодействия тех же проблем. Более того, классифицируя фонетические факторы как ограничения, мы можем включить фонетические детали в нашу фонологическую категоризацию.

Ключевые слова: акцентная фраза, интонационная фраза, французский, просодия, теория оптимальности, ограничения, основной тональный рисунок, ритм.

Introduction: background

Since the early works on prosody by Pierrehumbert (1980), Selkirk (1984) and Nespor & Vogel (1986), numerous efforts have been made to achieve the prosodic grouping of speech. In the early 1990's, for example, Silverman et al. (1992) proposed a new transcription system, called the To-BI (Tones and Break Indices), to analyze large amounts of prosodically transcribed English speech. Since its proposal, the To-BI system has been developed and extended to other languages (Avranti & Baltazani 2005, Jun 2005, Jun & Fougeron 2002, Pierrehumbert 2000, etc.). There are, however, various problems with the system, such as the indeterminacy of categorizing the H and L tones, loose listing of the distributional patterns, etc. Nevertheless, no satisfactory analysis has been attempted to effectively explain how the variation or deviation can be derived from the basic pattern.

There have been numerous studies on French intonational phonology (Martin, 1975, Hirst & Di Cristo1984, 1996, Post 2000, Rossi, 1999, Martin 2009, etc.). As most studies have not considered the basic tonal pattern, however, it is hard to show the basic rhythm or to constrain the variation patterns. On the other hand, after Post (2000), Jun & Fougeron (2000, 2002) and Rolland & Lrevenbruck (2002) propose the LHiLH* as the basic tonal pattern in French, based on the basic premises of the To-BI system. Although they list the realization patterns of the basic tonal pattern, they do not explain when or why such variations occur. More recently, Gussenhoven (2004) made an independent effort to account for the French intonational pattern within the framework of Optimality Theory (McCarthy & Prince 1995). Not assuming a basic tonal pattern, however, he relies on the categorical explanation on the variational pattern, rather than a detailed explanation on the cause of the distributional variation. As he states, however, that French prosodic (i.e., f) structure is variable (Gussenhoven 2004), there should be a way to explain when and how such variable structures occur.

Considering these problems, we will provide an Optimality-theoretic description on the variational patterns (McCarthy & Prince 1995), based on the stratal version of Optimality Theory (Kiparsky 2000, Ahn 2008, etc.). In elaborating our analysis, we will assume the two basic prosodic categories, Accentual Phrase (AP) and Intonational Phrase (IP). Based on the prosodic hierarchy of Selkirk (1984) and Nespor & Vogel (1986), Fougeron & Jun (1998) and Jun & Fougeron (2000, 2002) propose the level of Accentual Phrase (AP, henceforth) and the higher level, Intonational Phrase (IP, henceforth), as shown below.

(1)

IP

LHi

LH*

%

(IP: Intonation Phrase, AP: Accentual Phrase, W: phonological word, a: syllable, %: Intonation phrase boundary tone)

As shown in (1), an AP is marked by an initial LH rising tone and a final LH rising tone. With these basic concepts, we will explain how the basic AP tonal pattern can deviate and how the AP and IP domains interact with each other.1

1. Representation of the basic tonal pattern: LHLH

In order to categorize the tonal patterns in French, Rolland & Lrevenbruck (2002) and Jun & Fougeron (2002) propose the LHiLH* basic tonal pattern from which the five basic distributions are derived, as shown in (2). According to these studies, an LH is different from an initial LHi and from a final LH*. They also claim that the initial H, i.e., Hi, is a phrase accent, while the final H (i.e., H*) is a pitch accent (and a tone boundary) as Hi occurs in an AP-initial position, while H* occurs in an AP-final position.

(2) a. LHiLH*

b. LH*, LLH*, LHiH*, HiLH*, LHL*

(the underlined examples are less common in short sentences)

This list shows the major variation patterns in French but no explanation is provided on why and when such variations occur. For LH* (or LLH*), for example, they do not explain why the H is assigned as a pitch accent tone, rather than a phrase tone. It is also difficult to say why the final pitch accent tone disappears in the LHL* pattern. Moreover, the representation of L* shows another inconsistency problem as it violates the general H* assignment to an AP-final position.

In order to elaborate on our analysis, we first argue that the distinction between Hi and H* is not necessary since there is no phonological distinction between the initial and the final H tones and their locations can be easily predicted by a couple of Alignment constraints. 2 Jun & Fougeron (2002) also mention a constraint "Preserve Peripheral" to mark the final H tone. But they still rely on the complex notional variety, i.e., Hi, H, H* for H tones, while using the L, L* and L% for L tones. That is, their framework would need at least three distinctions for H and L tones. And this notional complexity gets worse in other languages, e.g., T, +H, L+, La, Ha for Korean (Jun 2005).3

In our current account, therefore, we propose the canonical AP tonal pattern simply as LHLH which remains stable through the various stages of realization.

(3) Canonical AP tonal pattern: LHLH

Positing the simple LHLH pattern, we thus do not need any complication in the tonal representation. Moreover, we can show that those variations are the consequences of the interaction of necessary constraints in the AP (or the IP) domain. Below, employing the

1 The prosodic grouping can be further supported by the earlier works on French (Di Cristo & Hirst (1993, 1996) and Delais-Roussarie (1996). Especially, Post (2000) proposes that a Rhythmic Unit (RU) consists of more than one Tonal Units (TU). Therefore, Post's TUs and RUs correspond to our APs and IPs, respectively.

{[ ]TU [ ]TU [ ]TU}RU

As shown in Di Cristo & Hirst (1993, 1996), however, each sequence of LH tones forms one Tone Unit regardless of word boundaries and the grouping of TUs is independent of their tonal shape. Therefore, there is no limit on the number of tone units within an AP (or Rhythmic Unit) (Delais-Roussarie 1996, Delais-Roussarie & Post 2008, Post 2010).

2 Within the (early) framework of Optimality Theory (McCarthy & Prince 1993), Delais-Roussarie (1996) also employs alignment constraints for a different purposed, i.e., proper phonological phrasing.

3 This approach is akin to those works in early generative phonology appealing to various boundary symbols.

framework of Optimality Theory, we will show why and how this basic tonal pattern is realized in various surface forms.

2. Constraints for tonal association

Optimality Theory (OT henceforth, McCarthy & Prince 1995) is a model of constraints and constraint interactions, while the standard generative theory consists of rules and derivations. The OT grammar contains two major components: GEN(erator) which maps the input onto an infinite set of candidate output forms, and EvAL(uator) which evaluates the candidate output forms by a set of ranked constraints and selects the optimal output among the possible candidates. Therefore, Gen produces a set of logically possible candidates of the input and Eval selects the optimal analysis of the input. Then the main analytical proposal of OT is that constraints are ranked in a hierarchy of relevance. Lower-ranked constraints can be violated in an optimal output to respect higher-ranked constraints. An optimal output can thus violate certain low-ranked constraints minimally. (For simpler optimality description, however, we use the traditional violations tableau format, rather than the comparative or combination tableau formats (McCarthy & Prince 2005).)

/Input/ Constraint 1 Constraint 2 Constraint 3 Constraint 4

a. Candidate 1 *! *

b. Candidate 2 *

c. f Candidate 3 * *

In sum, as shown in the above tableau, the constraints are intrinsically conflicting and Candidate 3 is selected as optimal since it minimally violates the lower-ranked constraints. The outline of the classic parallel OT can be schematized as follows (McCarthy 2006). Classic OT has thus two levels of representation, Input and Output. Gen generates the output candidates and, comparing the output candidates, EvAL selects the optimal output in a given (language-particular) constraint hierarchy. That is, classic OT is nonderivational, mapping inputs directly to outputs without intermediate stages.4

In order to analyze the tonal variation, we need various types of constraints which interact each other within the AP and IP. First of all, we posit the following faithfulness constraints.

(6) Max-T An input tone may have an output correspondent. (No tone deletion) Dep-T An output tone may have an input correspondent. (No tone insertion) We also need the following alignment constraints for the tone distribution.5

(7) Align-R Align the final H to the right edge of an AP.6 Align-L Align the initial L to the left edge of an AP.

Employing these two alignment constraints, the relevant tones in the basic LHLH pattern

4 Various approaches (Kiparsky 2000, Rubach 2000, 2003, Ahn 2008, etc.), however, pointed out the problems occurring in the parallelism of "classic" Optimality Theory. Considering that there are at least two different prosodic levels in French, AP and IP, we will employ the concept of "serialism" and the cyclic application of constraints in Optimality Theory. Here we argue for positing a re-ranking mechanism in each evaluation cycle, i.e., different constraint ranking at the AP and IP levels

5 Delais-Roussarie (1996) employs similar alignment constraints, i.e., Align Xhead and Align Xmax. Unlike our alignment constraints, however, Delais-Roussarie's alignment constraints are used for phonological phrasing by dividing a sentence/utterance into a certain number of phonological phrases

6 AP right boundary must be aligned on the last syllable of some lexical entry in French. (No category restriction in English; for example, it can be N, V, Adj, Adv, Pro, Conj, etc.)

can be easily assigned to the left and right edges. As a result, there is no need to group the first [LH] or the second [LH] together.

There are, of course, other prosodie as well as morphological constraints. For example, we need the following constraints which generate eurhythmy on the prosodic level.

(8) Rhythm constraints

* Contour A contour tone may not be allowed.

*[LL An initial L may not be branched.

*Lapse Three or more consecutive L tones may not be allowed.

*HH A branching H may not be allowed. (No clash)

First of all, *Contour discourages any syllable linked to both H and L tones within an AP. *[LL indicates that the first AP may not allow the spreading of the first L tone (in the basic LHLH pattern). It is thus similar to those constraints found in other languages like English, discouraging a long initial iambic beat. *Lapse discourages a sequence of three or more L tones. As will be shown later, however, that this constraint is easily violable and there seems to be no restriction on the number of L tones when forced for eurhythmy in a long AP, i.e., HLLLLLLH in [l'hippopotame sud-africain]. Finally, *HH is another eurhythmy constraint discouraging any tone clash within a word.

We also need a couple of morpho-phonological constraints to indicate that French requires certain morphological information available even on the prosodic level, i.e., AP. This morphological information is crucial when we distinguish content words from function words as they behave differently.

(9) Morphological constraints

*Wc-L A L tone may not be assigned to a (mono-syllabic) content word. *Wf-H A H tone may not be assigned to a function word.

These constraints show the basic guideline of tone assignment, i.e., H tone linking to content words, while L tones to function words. 3. Patterns of tonal distribution Tone mapping in Accentual phrases

The most basic and straightforward AP patterns in French are those which consist of four to seven syllables, as a sequence of a maximum of seven syllables must have at least one syllable stressed (Wioland 1985, Delais-Roussarie 1996).7 For this, we first analyze the "typical" example such as a short AP which consists of a noun phrase.

(10) [L H L H]%

Le bon garçon 'the good boy' [Le bon garcon]

The last syllable is associated with the last H, while the first syllable with the initial L. And the remaining syllables are associated with the remaining H and L.

In the data shown in Jun & Fougeron (2002), however, the basic tonal pattern is often altered depending on a given context, as shown in the following example illustrating a high tone clash in the first AP.

7 An AP duration is constrained between about 250 ms and 1,200 ms (Universal).

(11)

[Le coléreux garçon] [ment à sa mère] 'The choleric boy lies to his mother.'

(12) [L H L H] [L H L H] L%

I f\ /1 I I /1 V

[Le coléreux garçon] [ment à sa mère]

For the first AP [Le coléreux garçon], an autosegmental description may show the following procedures. First, we assign the last H to -çon (by Align-R) and the first L to le- (by Align-L). Then, the second and the penultimate syllables get H and L tones, respectively (by Max-T). Finally, the remaining syllables, -lé- and -reux are linked to H and L respectively (by Max). As there are six syllables which are more than the number of the four basic tones LHLH, an indeterminacy problem may occur in the tone association of the medial syllables.

(13) Default pronunciation

LH LH Le coléreux garçon *[LL *HH Align-R Align-L *Lapse

a. L H L H M 1 1 Le coléreux garçon *! (Le, co-)

b. L H L H 1/11 A Le coléreux garçon **!

c. L H L H e 1 L^ 1 Le coléreux garçon *

d. L H L H ®l N /11 Le coléreux garçon *

Here all the four candidates obey both alignment constraints. The first candidate showing a long iambic beat is eliminated as it fatally violates the top ranking constraint *[LL. The second candidate violates *HH twice as both syllables of the last word are linked to the final H tone. The third candidate, however, does not violate *HH, although it minimally violates *Lapse. On the contrary, the last candidate shows violation of the *HH constraint. We therefore take the third candidate as the optimal output. Nevertheless, the selected output in (13) is not the one shown in (11) (Jun & Fougeron 2002) since, by positing the adjective after the noun, this AP represents focusing. Therefore, we need an additional constraint required for the focused AP, and it dominates *Lapse. As a consequence, we can correctly select the "focused" optimal output as follows.8

8 Gussenhoven (2004: 266, 272) also states that NoClash may be violated in disyllabic fs, provided a particular intonation contour is used, e.g., très vite 'very fast'

56 -

(14) Focus H A focused multi-syllabic content word may have a branching H.

(15)

LH LH Le coléreux garçon Focus H *[LL *HH Align-L, R *Lapse

a. L H L H M II Le coléreux garçon *! * (Le, co-)

b. L H L H Ml /1 Le coléreux garçon **!

c. L H L H 1 1 1 Le coléreux garçon *! *

d. L H L H f 1 N ^ 1 Le coléreux garçon *

The second AP structure [ment à sa mère]% shows the typical LHLH tonal display, so we can follow the same constraint application procedures as shown above. As the AP occurs IP-finally, however, we need to separate the two prosodic levels. Therefore, at the AP level, Align-R assigns H to mère and L to sa, while the initial content word is linked to the initial H tone, showing a HLL pattern. Thus, the IP-boundary tone should override the earlier AP-final H tone at the IP level.

[Level 1: AP representation]

The candidate evaluation begins at the first prosodic domain, i.e., AP level, in which *Wc-L and *Wf-H play key roles as shown below.

(16)

LH LH ment à sa mère *Wf-H *Wc-L Align-L Align-R Max-T

a. [L H L H] 1 t II J ment à sa mère *! *

b. [L H L H] N 1 1 ment à sa mère *! ** * *

c. [L H L H] 1 1 ment à sa mère *! * *

d. [L H L H] F 1 /1 | ment à sa mère * *

*Wc-L indicates that an L tone may not be assigned to a (mono-syllabic) content word, while *Wf-H disallows association of a function word with a high tone.9 The first candidate shows a complete match of a tone and a syllable, but it violates the high-ranking constraints *Wf-H and *Wc-L. The second candidate is worse in that it violates the top constraint as well. The third candidate obeys the top constraint but violates *Wf-H and Align-L. Therefore, in spite of the

9 As shown in Jun & Fougeron (2002), however, this constraint can be violated if there are a series of clitic words preceding a content word or if there is a function word carrying an H tone.

violations of the lower ranking constraints, the last candidate is chosen as the optimal output since the violations are "minimal".

[Level 2: IP tone assignment]

At the AP level, we get the HLH pattern but this pattern is not what we may observe at the IP level. As we move to the higher IP level, therefore, we demote the earlier top-ranked Align-R to a lower position so that the AP final H tone can be overridden by the IP boundary tone. For the realization of the IP boundary tone, we need a new constraint *Contour which requires that the IP boundary tone be realized as a single tone, rather than a contour one. (Here we consider only those candidates with the final IP tone association. That is, the IP tone association is mandatory at the IP level.)

(17)

[L H L H] L% lt ,/1 1, ment à sa mère *Contour *Wf-H *Wc-L Align-L Align-R Max-T

a. [L H L H] L% 1 1 1/ ment à sa mère * * (L%) * *

b. [L H L H] L% M\>1 1 ment à sa mère * (à) * (ment) *

c. [L H L H] L% 1 A'V ment à sa mère * (sa) * (L%) * * *

d. [LH L H] L% f 1 ,/i y ment à sa mère * (L%) * * **

The new top-ranking constraint *Contour prohibits both H and L tones from being linked to mère. Instead, we need to replace the AP-final H tone with the IP boundary L tone. As the earlier top-ranked Align-R is demoted, the final H tone may not surface on this level. (Here, we omit the fatal violation mark as all the constraints seem to become violable from this prosodic level.)

As shown so far, the AP constraints interact with the IP constraints at the IP (or higher) level, so we employed a new optimality-theoretic analysis for these processes. Just like in the other cases, the prosodic variation caused by the interaction between these two types of constraints can be easily accounted for within a cyclic/stratal Optimality Theory (Kiparsky 2000,

T) (b)

These figures shows F0 tracks for a sentence containing multiple clitics, "il faut que je le lui donne", in which an H tone appears on the clitic "le" in (a), and on"lui" in (b). (The figures showing F0 tracks are taken from Jun & Fougeron (2002: 163).)

Rubach 2000, Ahn 2008, etc.). We then posited a re-ranking mechanism in each prosodic level, in which violation of certain higher-ranked AP constraints may not be fatal in the IP level.

Tone association in short APs

The evaluation tableau in (18) shows that, if the AP-final word is multi-syllabic, the final H tone is preserved by simply being moved to the immediately preceding syllable as in [des bananes\lp (in "Marion mangera des bananes'") showing a LHL pattern.

According to Jun & Fougeron (2002), the surface tonal pattern is LH* (or LLH*) if the number of syllables is less than 4. This account, however, does not say which L of the basic LHLH pattern shows up on the surface, whereas the final H is marked as the AP-final H tone.

150-1-i—J

0 Time (s) 1 509

[Marion] [mangera] [des bananes]. 'Marion will eat some bananas.'

In our account, however, we claim that the surface L tone is the result of the association of the first L, rather than the second L. To this end, we first show the following autosegmental representation of the tonal associations.

(19) [LHLH\ [L H LH\ [L HL H\ L%

\\l I II w

[Marion] [mangera] [des bananes].

As the tone mapping is not based on a one-to-one association, we may face some indeterminacy problems within a traditional autosegmental account. As will be shown below, however, the tonal realization patterns can be easily explained within OT in which we employ several basic constraints and their interactions.

The second AP Marion shows the LH (or LLH) distribution. Within a conventional autosegmental account, however, there may occur an indeterminacy problem since any L or H tone can be associated with a target syllable, as long as the association lines do not cross.

(20) ? L HLH ? LHLH ? LHLH ? LHLH

M K/ I // /f I

Marion Marion Marion Marion

In (20), we need a principled way to secure the prominence of the AP-final H tone. Moreover, we have to decide on only one correct pattern among all of the four possible L mapping patterns. Within a traditional rule-based autosegmental account, these problems cannot easily be solved unless we rely on arbitrary mapping rule formulation.

On the contrary, however, such an indeterminacy problem does not occur as we observe that the second L tone association is sanctioned by the constraint ranking. Even if there are three syllables within an AP, there is not much complication in tone association as the necessary constraint interaction allows us to select the optimal output.

(21)

L H LH Marion *[LL Align-L Align-R Max-T

a. LHLH Marion * *

b. LH L H \\l Marion * *

c. L HLH N / Marion *!

d. LH LH f \ || Marion *

The comparison between the candidates (c) and (d) reveals that the L realization in the second syllable occurs with the association of the second L, rather than the first L.

The final vowel in [mangera\ gets the final H tone, while the initial L is associated with the first

syllable by Align-L.

(22)

L H LH mangera *Contour Align-L Align-R Max-T

a. LH L H mangera *! * **

b. LH LH 1 N mangera *! * **

c. LH LH f mangera **

d. LH L H / 1 mangera * **

For this case, Jun & Fougeron (2002) stated that the first L, not the second L, is linked to the first syllable and our account confirms their claim in a logical way.

At the small AP level, the single word bananes in [des bananes] has a LH pattern, which is reversed to the HL pattern as we impose the boundary L% tone. In other words, we need two levels of evaluation tableaux as the IP tone assignment applies to the upper level, i.e., IP. (We omit irrelevant constraints here for convenience.)

[Level 1: AP representation]

LH LH des bananes *Wf-H *Wc-L Align-L Align-R Max-T

a. [L H LH] des bananes * *

b. [LH L H] des bananes * * *

c. [L HL H] N 1 des bananes **

d. [L HL H] f des bananes *

Both (c) and (d) show an LLH pattern in actual pronunciation, so it might be difficult to find out how we get the second L tone in a non constraint-based account. In our current account, however, the last candidate wins over the third one as it violates Max-T less seriously.

[Level 2: IP tone assignment]

Here we notice that the earlier top-ranked Align-R is demoted to a lower position since the final AP tone can no longer be maintained due to the assignment of the IP boundary tone. However, the AP-final H still surfaces as it is now linked to the preceding syllable. We can thus observe the phonological prominence of the AP-final H tone of a multisyllabic word.

Observe that the multisyllabic final word bananes should have an H tone for eurhythmy. The realization of the H tone is a consequence of the linking of the phonologically prominent AP-final tone, rather than the somewhat unstable initial H tone. In order to secure the stability of the AP-final tone in a multi-syllabic IP final word, we may rely on the *Wc-L in that the content word should have an H tone. In this case, however, there occurs an indeterminacy problem with which of the two H tones is associated with the content word. Considering that the final H is retained in most variation types, i.e., LH*, LLH*, LHiH*, HiLH* (except the rare LHL* type), we can posit the following constraint.10 (We omit irrelevant constraints for an easier illustration.) This mechanism complies with the "Preserve Peripheral" constraint found in Jun & Fougeron (2002) by which the final H tone is preserved.

(24) Max-H] An AP-final H tone is preserved.

[LH L H] L% l l k' des bananes *C0nt0ur Max-H] Align-L Align-R Max-T

a. [LH L H] L% 1 W des bananes *! *

b. [LHL H] L% 1 \v des bananes *! * **

c. [LH L H] L% 1 * 1 i ' 1 \i--k des bananes *! * **

d. [LHL H] L% f I YV des bananes * **

The first candidate violates *Contour since the final syllable of the AP is linked to both the AP-final H and the IP boundary tone. Then the second and the third candidates are eliminated by violating the new Max-H] constraint. Note that, without this constraint, the initial syllable in bananes can be linked to the initial H, rather than the final one, in order

10 Conceptually, it is similar to the Max Feature constraint in Lombardi (1998) as the AP-final H tone has be to realized somewhere on the surface.

to obey Align-L. The evaluation procedure also verifies that the surface H tone comes from the final H, rather than the first one (Jun & Fougeron 2002).

4. Complication: Phonetic implementation

As can be observed in a rather long AP, phonetic interpolation occurs in a sequence of the same tones or is triggered by a following tone. We underline the relevant tones in the following examples for easier demonstration.

(26)

• • • • •

[Le coléreux garçon] 'the choleric boy' L H HL L H

(27)

• •

o

• • • . .

[L'hippopotame sud-africain] 'South-African hippopotamus' H L L H_L L L H

The second H in (26) is a consequence of the rising pitch representing a focused adjective being located before the noun. But we can also observe phonetic implementation showing interpolation between the H and L tones in that the final L tone in coléreux is higher than the L of the following noun garçon. In (27), on the other hand, the last H of L'hippopotame is lower than the first H and falls before the following L which is also higher than the next L. These adjustments of the tonal contours are a natural consequence for smoother articulation but, due to the gradient nature of the contours, they could be subject to the controversy of ambiguity or artificial categorization in a purely phonetic account. In the current account, however, we state that the prosodic interpolation can be represented in a later upper prosodic level by the following constraint.

(28) Phonetic implementation: interpolation

Raise the F0 of the first part of a multi-linking L, if preceded by an H.

Lower the F0 of the H located between two L tones.

By interpolation, we can adjust the actual height of the tones, depending on the locations. Therefore, we raise the L pitch of the last syllable in coléreux in (37), while lowering the second H (38), to make a smoother pitch curve.

Another case of interpolation can also be observed in the following example, showing that the intonation of the first prosodic word is falling and contrasting with the final rising one.11

11 The following pitch display, however, shows that the pitch contour of the first prosodic words may depend on the syllable type since the monosyllabic content word rat shows a rising pattern. That is, le rat marron forms one AP(PW), with a secondary stress on rat and a primary stress on marron, while the secondary stress is always rising.

r 150 it mar on

le 100 x 7 avait A voulu mang- r le long v m jlot

50

40

20 ^^ÉhM

Secondary stress and the initial LH

The first AP I'elephanteau marron is divided in two prosodic words, both ending with an H, the first one falling and contrasting with the final rising one.

Furthermore, in the sequence of the L tones, we keep lower the F0 of the L tones until we reach a morpho-syntactic boundary. So, in the case shown in (37), the pitch continues to be lowered until it reaches the end of a noun phrase. In other words, we would need further morphological information when more morpho-syntactic structures are incorporated within an AP.

(30) L downstep

A long sequence of L tones continues to be lowered until it reaches the end of a morpho-syntactic boundary.

(31)

Si [l'hippopotame sud-africain] [prenait des bananes] L H L L L L L L L L L L L H

-

I'l-li Si r ppo \ ,po su 1 Pre

tame 1 d v V> a 111 I nai \ cam ' \ W ^V des ba /

»

Jr=Ç -1— ^^

5. Further remarks

So far we have discussed how the various types of tonal deviations can be accounted for within the framework of the stratal version of Optimality Theory. We have shown that those variations are the result of the sequences of the constraint interactions and their ranking relationships. Moreover, by maintaining the LHLH tone at every AP level, we can generalize the basic tonal pattern in French from a universal point of view. Furthermore, the gradient nature of phonetic implementation can be interpreted with respect to interpolation filling the

gap between two different tones. In the additional discussion to follow, we reexamine the prosodic hierarchy proposed in Jun & Fougeron (2002) and propose an alternative model compatible with the so-called Strict Layer Hypothesis in Selkirk (1986).

As proposed in Selkirk (1986) and Nepor & Vogel (1986), the main claim of prosodic phonology is a hierarchy consisting of prosodic units. According to "Strict Layer Hypothesis" (Selkirk 1984), a prosodic category of one level is exhaustively parsed into constituents of the next-lower level. And, those next-lower level constituents are all of the same type and prosodic structure is not recursive. Based on this, Nespor & Vogel (1986) propose a hierarchy of seven prosodic constituents with the requirement that each prosodic unit belong to a constituent of the next higher rank in the hierarchy. We can thus posit the following figure of prosodic hierarchy for French.

(32) The prosodic hierarchy Utterance (U) Intonational Phrase (IP) Phonological Phrase (f) (Clitic group) Phonological Word («) Foot (F) Syllable (a) Mora

Note, however, that Jun & Fougeron (2002)'s model of prosodic hierarchy deviates from the Strict Layer Hypothesis in that the tones are associated with the AP level, as well as with the syllables, and they could bypass the level of prosodic words. (Furthermore, the IP boundary tone is adjacent to the AP tones.)

(33)

LHi LH* %

In order to solve the conflict with the Strict Layer Hypothesis, we posit the tonal tier immediately below the AP tier since an AP has its own basic tonal pattern, LHLH, which is composed of two LH tones to be aligned from R^ L and L ^ R respectively. Moreover, each

LH can be linked to one or more prosodic words and each prosodic word consists of one or more syllables. (We do not discuss the clitics here for the simplicity of the argument.)

(34)

This figure shows that the basic LHLH tones are the property of an AP, which in turn dominate the prosodie words governing syllables. In this proposal of the prosodie hierarchy, therefore, there does not occur any logical mismatch or indeterminacy. Within the OT analysis, information on the content word and the function word can then be incorporated as constraints at the AP level. Then, the AP tone assignment can be overridden by the IP tone assignment since with the variable French prosodic unit (i.e., f), the IP boundary tone is determined by a sentence type, speech style, emphasis, etc. (Gussenhoven 2004).

R e f e r e n c e s

1. Ahn, Sang-Cheol. 2008. Schwa insertion in English. English Language and Literature 54:6, 1007-1026.

2. Delais-Roussarie, Elizabeth. 1996. Phonological phrasing and accentuation in French. In Marina, Nespor and Norval Smith (eds.) Dam Phonology: HIL Phonology Papers II, 1-38. The Hague: Holland Academic Graphics.

3. Dell, François. 1984. L'accentuation dans les phrases en français. In François Dell, Daniel Hirst and Jean-Roger Vergnaud (eds.) La forme sonore du language : la nature des représentations en phonologie. Paris : Herman.

4. Di Cristo, Albert. 1999. Intonation in French. In Daniel Hirst and Abert Di Cristo (eds.) Intonation Systems: A Survey of Twenty Languages. Pp. 195-218. Cambridge: Cambridge University Press.

5. Gussenhoven, Carlos. 2004. The Phonology of Tone and Intonation. Cambridge University Press.

6. Hirst, Daniel J. and Albert Di Cristo. 1984. French intonation: a parametric approach. Die Neueren Sprachen 83:5, 554-569.

7. Jun, Sun-Ah. 1998. The Accentual Phrase in Korean prosodic hierarchy. Phonology 15.2, 189-226.

8. Jun, Sun-Ah and Cécile Fougeron. 2002. Realization of accentual phrase in French intonation. Probus 14, 147-172.

9. Kiparsky, Paul. 2000. Opacity and cyclicity. The Linguistic Review 17, 351-367.

10. Lombardi, Linda. 1991. Laryngeal Features and Laryngeal Neutralization. Ph.D. dissertation, University of Massachusetts at Amherst. Published in 1994 by Garland Press, New York.

11. Lombardi, Linda. 1998. Evidence for Max Feature constraint from Japanese. University of Maryland Working Papers in Linguistics 7, 41-62.

12. Martin, Philippe. 1975. Analyse phonologique de la phrase française, Linguistics, 146 (Fév. 1975), 35-68.

13. Martin, Philippe. 2009. Intonation du français. Paris : Armand Colin.

14. McCarthy, John. 2006. Derivations as candidates in Optimality Theory. Talk given at the Phonology Forum 2006. The Phonological Society of Japan, Waseda University.

15. McCarthy, John and Alan Prince. 1993. Prosodic Morphology I: Constraint Interaction and

Satisfaction. Ms. University of Massachusetts and Brandeis University.

16. Nespor, Marina and Irene Vogel. 1986. Prosodie Phonology. Dordrecht: Foris.

17. Pierrehumbert, Janet. 2000. Tonal elements and their alignment. In M. Horne (ed.) Prosody: Theory and Experiment. Studies Presented to Gösta Bruce. Pp. 11-26. Dordrecht: Kluwer Academic Publishers.

18. Post, Brechtje. 2000. Tonal and Phrasal Structures in French Intonation. Doctoral dissertation, University of Nijmegen. The Hague: Holland Academic Graphics.

19. Post, Brechtje. 2010. An Autosegmental-Metrical grammar of French intonation. Invited contribution to appear in Journal of French Language Studies, special issue.

20. Rolland, Guillaume and Hélène Lœvenbruck. 2002. Characteristics of the Accentual Phrase in French: an acoustic, articulatory and perceptual study. ISCA archive http://www.isca-speech-org/archive, Speech Prosody 2002, Aix-en-Provence, France.

21. Rossi, Mario. 1999. L'intonation, le système du français : description et modélisation. Paris: Ophrys.

22. Rubach, Jerzy. 2000. Backness switch in Russian. Phonology 17, 39-64.

23. Selkirk, Elisabeth, 1986. On derived domains in sentence phonology. Phonology Yearbook 3, 371-405.

24. Silverman, Kim, Mary Beckman, John Pitrelli, Mari Ostendorf, Colin Wightman, Patti Price, Janet Pierrehumbert and Julia Hirschberg. 1992. ToBI: A standard for labeling English prosody. Proceedings of the 1992 International Conference on Spoken Language Processing. Banff, Canada. Vol. 92:2, 867-870.

25. Wioland, François. 1985. Les structures rythmiques du français, Slatkine-Champion, Paris.