1.1 Persian word accent

Although Persian word prominence has been described as “word stress” (e.g., Ferguson 1957; Windfuhr 1979; Mahootian 1997), recent experimental evidence shows that it is realized only through fundamental frequency (F0) (Abolhasanizadeh et al. 2012). Thus, in technical terms, it is an accent, a tone in a specific location, like the lexical (prosodic) prominence in Japanese and unlike the English stress, which is signalled by phonetic features other than just F0 (Beckman 1986). Throughout this text, we will use the term accent for the Persian word prominence. Phonologically, the Persian accent consists of a H tone.¹

Persian morphological words are accented on the final syllable without exception. By morphological word (henceforth MWORD) we mean both morphologically simplex words (consisting of a simple stem) and complex words, i.e. derived words (consisting of a stem plus some derivational affix(es)) and compounds (consisting of more than one stem). The hallmark of MWORDs is that they are treated as single units by inflectional and syntactic rules. Thus, for example, Persian compounds are indivisible constructions whose components cannot be interrupted by inflectional markers or modal particles.² Note that our definition of MWORD constitutes only major word classes like noun and adjective. Some grammatical word classes will be briefly discussed in Subsection 1.2.

Three MWORDs are illustrated in (1), where (1a) contains a simple stem, (1b) a stem plus a derivational suffix, and (1c) a stem-stem compound. Each example is given in four lines. The first line gives the IPA transcription with morpheme boundaries. Suffix boundaries and clitic boundaries are shown by “–” and “=” respectively, whereas components of a compound are separated by “+”. The second line gives the prosodic representation. Phonological words (to be defined below) are marked by round brackets, while periods show syllable boundaries. Accent is marked with an acute over vowels. The third line gives the morpheme-by-morpheme gloss, when relevant. The fourth line gives the English translation.

The position of the accent in a Persian MWORD always coincides with its right edge, regardless of the phonological word structure, with which MWORDs are pervasively non-isomorphic (cf. Nespor & Vogel 1986). The most obvious cue to phonological wordhood in Persian is syllabification (cf. Bijankhan 2005; Hosseini 2014: 42). We therefore define the phonological word (henceforth PWORD) as the domain of obligatory syllabification, meaning the Maximum Onset Principle (Kahn, 1976) is applicable across its entire domain, with a syllable structure (C)V(C)(C).³ This formulation leads to the generalization that a Persian PWORD corresponds to a simple stem plus its bound morphemes (an extension of this generalization will be given in Subsection 1.2, where we introduce clitics). Thus, while suffixes form a single PWORD with their stem on the left, as illustrated in (1b), the constituents of a compound form separate PWORDs, as shown in (1c). However, in all of (1a, b, c), there is only one accented syllable, the final syllable of the MWORD.⁴⁵

The final accent pattern is not affected by the length and internal complexity of the MWORD (cf. Vahidian-Kamyar 2001). Thus, the reduplicated compound [doxtar+amme+moxtar+ammé] ‘cousin and stuff’, a single MWORD realized as four PWORDs, receives only one accent on its final syllable.⁶ Moreover, the alignment of accent to the right edge of MWORD is fixed, meaning that it is not possible to accent the first constituent of a compound, say, not even if it is narrowly focused (see Subsection 1.3 for more discussion on this).

Before moving on, a remark should be made about two types of syllabification in Persian. Obligatory syllabification takes place in the PWORD domain, as stated above. Optional resyllabification occurs (e.g., in fast speech) across PWORDs, to the extent that all PWORDs within a sentence-size utterance can be syllabified together (cf. Hosseini 2014: 43).⁷ Thus, the compound in (1c) may be pronounced as [dox.ta.ram.mé]. Variations in syllabification in no way affect the position of accent. That is, a compound is always accented on the final syllable, regardless of its syllabification.

1.2 Minimal pairs

The location of accent has been generally recognized as highly contrastive in Persian, in that it contributes to distinguishing between morphosyntactically different utterances (Ferguson 1957; Lazard 1992; Vahidian-Kamyar 2001). An important source of these contrasts is a closed set of morphemes which are integrated into a single PWORD with a morphological host on their left. While these cliticizing morphemes are similar to suffixes in their lack of prosodic autonomy, they differ from suffixes in that they are syntactically independent units (cf. Ghomeshi 1996; Kahnemuyipour 2003). In fact, they have a variety of functions that represent positions in the syntactic phrase, like auxiliaries, pronouns, conjunctions and focus governing morphemes. We refer to them as clitics, because they fulfil classic criteria for clitichood, such as not being particular about the word class of their host and failing to exhibit arbitrary gaps in their combinations with their host (Zwicky & Pullum 1983).

Crucially, since clitics, unlike suffixes, fall outside the morphological domain of the host, they are not assigned accent, causing the accent on the MWORD-plus-clitic combinations to remain on the final syllable of the MWORD. This property of clitics makes them similar to a group of unaccented morphemes which are generally referred to as function words, like the prepositions [az] ‘from’ and [be] ‘to’. However, unlike clitics which are always cliticized leftward to a host, function words exhibit optionality in the direction of cliticization.⁸⁹

The functional load of accentual contrast associated with clitics is extremely high at the level of PWORD. Two types of PWORD-level minimal pairs can be distinguished in terms of their morphosyntactic structure. First, segmentally identical MWORDs and MWORD-plus-clitic sequences contrast for the location of accent. The minimal pair in (2a) illustrates the contrast between a simple MWORD and a simple MWORD-plus-clitic, whereas (2b) gives the contrast between a derived MWORD and a simple MWORD-plus-clitic combination. Most clitics are homophonous with suffixes.

Given that accentual contrasts of the sort just described arise from a difference in morphosyntactic structure, there are potentially countless numbers of minimal pairs of MWORD and MWORD-plus-clitic combinations in the language, especially in cases where suffixes are productive. An example of a highly productive suffix is the (colloquial) referential marker [-e] which is homophonous with the 3SG copula clitic [=e]. Thus, the referential nominal construction N+[-e], which is an MWORD, forms an accentual minimal pair with the predicate nominal construction N+[=e], which is an MWORD-plus-clitic, where the N slot can be filled by any noun/adjective of the language.

In the second type of minimal pairs, the two members are both MWORD-plus-clitic sequences. This situation arises in colloquial Persian only, since in such minimal pairs one member has undergone some morphophonological modification which only occurs in colloquial language. Specifically, some clitics have a particularly close relation to their MWORD host, as evidenced by the deletion of their initial vowel after an MWORD ending in vowel other than [e], thus resolving a potential vowel hiatus by the loss of a syllable. An example is the 1SG possessive clitic [=am] as in [labu]+[=am] → [labu=m] [labúm] ‘my cooked beets’. A MWORD-final [e] is deleted in preference to the initial vowel of the clitic, as shown in (3a), whereby the accent appears on the vowel of the clitic (Ferguson 1957: 128). These forms contrast with segmentally identical MWORD-plus-clitic forms in which the MWORD ends in a consonant.

The pair in (3b) illustrates the contrast between past and perfect verb forms. Since the perfect form consists of the past stem plus the participle suffix [-e] plus the personal clitic, all perfect forms minimally contrast with past forms which consist of past stem plus the personal clitic. Thus, these two verb systems are only distinct by the accent in colloquial Persian.¹⁰ In formal/literary Persian the perfect forms are written as their underlying form and are pronounced as such. That is, the formal/literary equivalent of [bordám] is [bordéam].

Accentual minimal pairs are also highly frequent at the clause level, for instance due to the status of compounds as single MWORDs. To illustrate, the NP-VP clausal structure (4a) contrasts with the structure (4b), where [fɒrsi+zabɒn] is a compound.¹¹ Persian is a null-subject language.

Significantly, despite this extremely high functional load of accent location, Persian listeners have been found to be stress deaf, meaning that they perform poorly on reproducing the order of series of non-word stimuli that minimally differ in the position of phonetic prominence (e.g., [númi] and [numí]) (Rahmani et al. 2015). This is surprising if stress deafness is not expected to be found in languages with contrastive stress/accent, as suggested by Dupoux et al. (2001). However, as argued by Rahmani et al. (2015), if stress deafness in the short-term recall task relies on the listener’s ability to store word prosody in their lexicon, the poor performance of Persian listeners can be explained by the assumption that Persian accent distinctions arise post-lexically, i.e. after words are inserted into phrase structure. Importantly, our data, as well as those of Rahmani et al. (2015), suggest that whereas these generalizations about accent assignment can be transparently described in terms of morphosyntactic constituency, they are not obviously governed by the prosodic phrasing.¹²¹³

In our experiment, we will use both types of minimal pairs that are formed at the level of PWORD, i.e. MWORD vs. MWORD-plus-clitic and MWORD-plus-clitic vs. MWORD-plus-clitic minimal pairs, exemplified above in (2) and (3) respectively. We will now move on to discuss two contexts that have been claimed to delete accent, the focus and presupposition constructions.

1.3 Focus

As in many languages, the variation in the focus of the sentence has intonational effects in Persian. Any MWORD in the sentence can be intonationally marked for focus, unless the MWORD is structurally unaccented within the sentential construction.¹⁴ Variation in the focus of the sentence is illustrated for the subject-object-verb (SOV) structure in (5), which uses a question/answer paradigm. The size of the focus constituent is indicated by square brackets. The response in (5a) illustrates focus for the whole sentence. The responses in (5b) to (5d) illustrate focus on the subject, the object and the verb, respectively. In the response given in (5e), the subject and the object are both focused. The response in (5a) may be referred to as “neutral focus” in the sense that no constituent in the sentence is informationally more prominent than any other. All other responses, i.e. (5b) to (5e), can be seen as instances of “non-neutral focus”.¹⁵

Figure 1 gives the F0 contour of the responses in (5a–5e), pronounced by a male speaker. As is obvious, post-focal constituents have a highly reduced F0 range. This has been widely observed in previous literature on Persian prosody (e.g., Mahjani 2003; Sadat-Tehrani 2007; Scarborough 2007), and been systematically studied by Taheri & Xu (2012), who described Persian as a Post Focus Compression (PFC) language. There is also evidence that the presence of PFC is the most robust perceptual cue for focus (Taheri et al. 2014), a point that will be discussed more in Section 4.2.

Figure 1

F0 contours, plotted below the associated waveforms, for the SOV structure [ɒlmɒn junɒno bord] ‘Germany beat Greece’, spoken by a male speaker in response to What happened? (panel (a)), Who beat Greece? (panel (b)), Who did Germany beat? (panel (c)), What did Germany do to Greece? (panel (d)), and Who beat whom? (panel (e)).

The F0-lowering effect on the post-focal constituents is so significant that many researchers on the topic have claimed that the post-focal materials lose their accent to the extent that minimal accent pairs become homophonous (e.g., Eslami 2000; Vahidian-Kamyar 2001), although these researchers reported no experimental data. The idea that constituents become deaccented after the focus, however, has been challenged by the experimental evidence in Abolhasanizadeh et al. (2012). We will turn to this issue in Subsection 1.5.

Another point evident from Figure 1 is that pre-focal MWORDs retain their accent, even if they are discourse-given, i.e. mentioned in the context question. Thus, responses in panel (a), where the whole sentence is focused, and in panel (d), where only the final MWORD is focused, have similar F0 shapes, although the MWORDs [ɒlmɒn] ‘Germany’ and [junɒn] ‘Greece’ are both given in the immediate context related to the latter. Moreover, when there are two focused MWORDs in the sentence (as in the response in panel (e)), the F0-compression starts after the second MWORD. The F0 shape of the response in panel (e) is thus comparable to that of in panel (c), where only the second MWORD is focused.

The minimal constituent for intonational marking of focus is the MWORD as noted before. Strikingly, a PWORD that is contained inside a MWORD cannot be intonationally marked for focus. For instance, the compound [fɒrsi+zabɒ́n] ‘speaker of Persian’, a single MWORD expressed as two PWORDs, is ambiguous between focus on the first member, on the second member, and on the whole construction, as shown in (6). However, these three focus domains may be distinguished by phonetic cues other than F0, as suggested by informal observation.¹⁶

The fact that Persian does not allow accentuation of the first constituent of a compound is not surprising given the status of accent as a morphosyntactic boundary marker, as shown for the noun+noun sequence [fɒrsi+zabɒn] (Persian+language) in (4). Thus, while [fɒrsí zabɒn ast] is an appropriate answer to the question Is Iranian a language?, it would be an inappropriate answer to Is she a speaker of French? That is, the phonological structure [fɒrsí zabɒn ast] can only be interpreted as [Persian]_FOC is a language (cf. 4a), not as She is a speaker of [Persian]_FOC.

In addition to the possible phonetic effects of focus size, there have been two further distinctions that have been considered as sources of systematic phonetic variation. First, it has been noted that focused constituents have a relatively high F0 (e.g., Sadat-Tehrani 2007; Taheri & Xu 2012). In particular, Scarborough (2007) suggested that the accent for focus is part of the phonological representation, resulting from a H^ (superhigh) tone, as opposed to the neutral H tone. The evidence that we present in the current study suggests that there is no phonological difference between the accent for focus and the neutral accent, and that F0-raising for focus can be attributed to paralinguistic speaker strategies. Second, focus type has been claimed to have an effect. The question/answer pairs in (5) illustrate presentational focus, a term that has been used for information provided by the speaker in reply to the hearer’s request, either overt or implied. Previous studies about Persian focus have also been concerned with other types of focus, in particular contrastive focus, which relates the focus constituent to a restricted set of alternatives (cf. Gussenhoven 2008). In her syntactic approach to Persian focus, Karimi (2005: 132) claimed that contrastive focus and presentational focus show different prosodic realizations, although she is not clear as to exactly what the difference is.¹⁷ We are not aware of any experimental study that systematically compares intonational effects of presentational focus and contrastive focus in Persian, but the existing data do not indicate any obvious intonational differences between these focus types (Vahidian-Kamyar 2001; Mahjani 2003; Scarborough 2007).

1.4 Factive verbs and presupposed dependent clauses

Persian has certain constructions that are semantically related to focus and which have similar intonational effects. Scarborough (2007) has described them as “focus-like” constructions which are triggered by specific syntactic contexts or are determined by the lexical semantics of particular items. She notes that the prosody of these constructions resembles the prosody of focused constituents, as they are “marked by a high pitch accent and followed by deaccenting” (Scarborough 2007: 28). Examples are wh-questions, negation constructions and clausal complements of certain predicates such as [dunestan] (realized as [dɒnestan] in formal styles) ‘to know’. Here we limit our discussion to the latter.

Scarborough observed two intonation patterns in the pronunciation of sentences containing complement clauses, exemplified in (7), where complement clauses are in brackets. Persian clausal complements, which appear in post-verbal position, may be introduced by the morpheme [ke].¹⁸

The author argues that while (7b) follows the neutral intonation of Persian declaratives in the sense that all MWORDs are accented, in (7a) the verb form [midune] ‘she knows’ is pronounced as if it was focused, since the MWORDs in the complement clause are deaccented, even though speakers produced this intonation in a contextless reading. Scarborough suggested that this “focus-like” pronunciation might be due to an “obligatory semantic focus” as triggered by the lexical semantics of a particular verb class, of which the verb [dunestan] ‘to know’ is an example. She provided no further explanation of this observation. We also refer the reader to Sadat-Tehrani (2007) on these effects for verbs meaning ‘to forget’, the only other reference to this phenomenon in the literature as far as we are aware.

The semantic contrast between the main verbs in (7a) and (7b) may remind us of a traditional distinction between two classes of clause-taking verbs, factives (e.g., know, realize, regret) and non-factives (e.g., say, believe, think). Factive verbs presuppose the truth of their complement clause while non-factive verbs do not (Kiparsky & Kiparsky 1970). For example, in she knows that the cat is in the garden, the speaker presupposes that the cat is in the garden, while in she says that the cat is in the garden, there is no such requirement (Crystal 2008: 184).

We will refer to deaccented clausal complements of factive verbs as presupposed clauses. In (7a), where the embedded clause is deaccented, the ripeness of the pineapple is presupposed, while no such presupposition is present in the accented clausal complement of (7b). While we do not deny that in an out-of-the-blue context clausal complements of factive verbs are more likely to be deaccented as suggested by Scarborough’s data, they can be accented under specific discourse conditions, specifically when they contain non-presupposed information.¹⁹ Example (8) may be spoken in reply to What is the latest news from the World Cup? In this context, the complement clause does not convey presupposed information in the sense discussed above.

Crucially, the main verb [midunam] does not merely express factivity, but is used to indicate that the information in the complement clause is only a partial answer to the question, a meaning which has been categorized as non-exhaustive focus by Elordieta & Irurtzun (2010). We may paraphrase (8) as follows:

“I am not in a position to give a full answer to your question. But here is a relevant fact which I do know: Germany beat Greece.”

A syntactic correlate of the distinction between the non-exhaustive complement clause (8) and the presupposed one in (7a) is that only the non-exhaustive one allows preposing, as illustrated in (9).²⁰ When this happens, the (normally factive) verb loses its accent.

These observations are of practical relevance to our experiment. The fact that clausal complements of factive verbs can be accented as well as deaccented depending on the meaning makes them suitable as stimuli in our experiment.

1.5 The problem

Abolhasanizadeh et al. (2012), using both production and perception data, have shown that while the F0 range of their target minimal pairs is considerably reduced after the focus, the tonal structure remains intact, as reflected in above-chance identification of the post-focal items by the listeners. Thus, the accentual difference between the MWORD [tɒbéʃ] and the MWORD-plus-clitic [tɒ́b=eʃ] was preserved in the post-focal context in (10).

Our interest in the present study is to reconsider this issue. We believe that the validity of the data used by Abolhasanizadeh et al. (2012) is questionable. In fact, the status of focus is unclear in their production design, which in turn yielded the stimuli for their perception experiment. They used a reading task in which the intended focused elements were printed in bold letters. Since there was no question/answer paradigm, it is unclear how each speaker interpreted information status of the bold elements in the text. It is possible that at least some speakers pronounced the bold items merely in a more emphatic way, i.e. with more acoustic energy. This is important because emphasis, unlike focus, does not create post-focal deaccenting in Persian (we will provide more information on the distinction between focus and emphasis in Subsection 4.3). This consideration motivated us to replicate Abolhasanizadeh’s study with a more realistic elicitation task and a larger corpus. The specific objective of our study was thus to determine whether Persian words undergo deaccentuation. We investigated this question through the use of two experiments, a production experiment and a perception experiment.

In order to place the investigation in a wider perspective, we decided to include different putative deaccenting contexts in our corpus. As target items, we used two minimal pairs with different morphological structures. The target minimal pairs were each examined in the context of the two types of information structural configurations, namely contrastive focus and factive verb construction described above in Subsections 1.3 and 1.4.

2.1 Materials

We built up a corpus of sentences featuring two minimal pairs as target items, given in Table 1. The first pair is a contrast between a simple noun and a noun-plus-clitic, while the second pair illustrates a contrast between two verb-plus-clitic sequences.²¹ Target items that are initially accented are referred to as initial accent condition, while those with final accent are referred to as final accent condition.

We constructed four carrier sentence structures, each corresponding to one of the possible combinations of the two minimal pairs ([tɒ́beʃ/tɒbéʃ], [dídan/didán]) and the two information structural configurations (contrastive focus, factive verb construction). The carrier sentences were designed in such a way that they provide the minimal context required to cover all experimental conditions in which they were used. Table 2 gives the structure of the corpus. There were three experimental conditions for focus, (1) neutral (out-of-the-blue pronunciation), (2) focal (the target item is contrastively focused), and (3) post-focal (the MWORD preceding the target item is contrastively focused). For factive verb construction, there were two experimental conditions, (1) non-presupposed (the target item is part of a non-presupposed complement clause) and (2) presupposed (the target item is part of a presupposed complement clause).

Within the carrier sentences, the target items had a cliticized form, i.e. X=e for the structures involving the minimal pair [tɒ́beʃ/tɒbéʃ], and X=eʃ for the structures with the minimal pair [dídan/didán], which makes them all part of trisyllabic PWORDs that contrast in having the accent on the antepenultimate syllable (for initial accent target items) or on the penultimate syllable (for final accent target items). The motivation to use a PWORD-final unaccented syllable in all cases was to avoid local phrase-final effects on the realization of the target syllables.

To elicit the desired experimental conditions, carrier sentences were preceded by a context sentence, with which they formed mini-dialogues. This resulted in 12 mini-dialogues for focus and eight mini-dialogues for factive construction (see Table 3 for examples; all mini-dialogues are given in Appendix 1). Example mini-dialogues, which were different from the test materials, were provided to the speakers to practice each condition before the actual recording. There were three example dialogues for focus and two example dialogues for factive construction, which were recorded by two male speakers different from the test speakers.

2.2 Speakers

Eight speakers took part in the experiment, four male and four female, aged from 27 to 37. They were native speakers of Standard Persian, all with university education. Informed consent was obtained from each participant.

2.3 Procedure and recording

The dialogues were presented to the speakers in two experimental blocks. One block consisted of materials related to contrastive focus, while the materials related to factive verb construction were put in the other block. The order of blocks and of test dialogues within each block was randomized per speaker in order to control for presentation order. Each block was presented to the speakers in the form of a booklet with one dialogue per page in Persian orthography. MWORD-plus-possessive clitics can be written either as one orthographic word (the traditional standard) or as separate orthographic words. We wrote [tɒ́beʃ] as two orthographic words to distinguish it from [tɒbéʃ]. Since there is no standard way of distinguishing between [dídan] and [didán] in Persian orthography, these items were written in their formal/literary form, which corresponds to [didand] and [dideand], respectively. The person/number marker [=an] (3PL) is realized as [=and] in formal and literary Persian.

Before the recording of each block, speakers were given an opportunity to try out the example dialogues by listening to the sound files played from a laptop and pronouncing the target sentence, for as long as they wished. In general, the speakers found the task clear and easy and quickly moved on from the trial.²² During the recording, the experimenter pronounced the context sentences and the speakers read aloud the response sentence from the booklet, without being able to see the experimenter’s face.²³ The procedure was performed twice to obtain two recordings from each condition per participant.

Unlike previous studies which used bold letters for focused items (e.g., Abolhasanizadeh et al. 2012), we did not highlight focused items in any way, assuming that the appropriate context questions should suffice to elicit the intended focus structure. Highlighting the intended focus elements typographically might moreover have caused subjects to pronounce these items in particularly emphatic ways. We return to this point in Section 4.

The speakers were recorded individually in the studio of the Linguistics Department of the University of Tehran using a Shure SM58 vocal cardioids microphone (44.1 kHz, mono channel, 16-bit). For analysis, one recording was randomly selected from the utterances of each sentence by each speaker which resulted in 20 response utterances per speaker overall.

2.4 Results

We report general observations on the effects of information structure on the realization of accentual contrasts based on visual inspection of time-normalized averaged F0 curves pooled over all eight speakers, followed by statistical evaluation with analysis of variance for repeated measures (rmANOVA). A Praat script was used to carry out the measurements and to correct pitch errors resulting from creaky voice or octave jumps (Boersma & Weenink 2014).

F0 measurements for initial accent and final accent target items will be reported separately for each of the four parts of the corpus given in Table 2. Figures 2 and 3 show averaged contours on normalized time scales for the minimal pairs [tɒ́beʃ/tɒbéʃ] and [dídan/didán] in the three focus conditions, while Figures 4 and 5 show the two factive conditions. Time-normalized F0 measurements were collected at 12 equidistant points per syllable.

Figure 2

Mean F0 contours for [un tɒbeʃe] on a normalized time scale for initial accent and final accent items separately, with target items in neutral pronunciation (panel (a)), in focal pronunciation (panel (b)), and post-focal pronunciation (panel (c)), pooled over 8 speakers. The values are given in semitones (reference: 100 Hz).

Figure 3

Mean F0 contours for [unɒ didaneʃ] on a normalized time scale for initial accent and final accent items separately, with target items in neutral pronunciation (panel (a)), in focal pronunciation (panel (b)), and post-focal pronunciation (panel (c)), pooled over 8 speakers. The values are given in semitones (reference: 100 Hz).

Figure 4

Mean F0 contours for [midunam tɒbeʃe] on a normalized time scale for initial accent and final accent items separately, with target items in non-presupposed pronunciation of the embedded clause (panel (a)), and presupposed pronunciation of the embedded clause (panel (b)), pooled over 8 speakers. The values are given in semitones (reference: 100 Hz).

Figure 5

Mean F0 contours for [midunam didaneʃ] on a normalized time scale for initial accent and final accent items separately, with target items in non-presupposed pronunciation of the embedded clause (panel (a)), and presupposed pronunciation of the embedded clause (panel (b)), pooled over 8 speakers. The values are given in semitones (reference: 100 Hz).

We turn to the results for focus first. Panels (a) in Figures 2 and 3 present the contours in the neutral condition for [tɒ́beʃ/tɒbéʃ] and [dídan/didán], respectively, which show the effect of the position of the accent. The initial syllable is 3.83 ST (SD=1.95) ([tɒ́beʃ/tɒbéʃ]) and 3.09 ST (SD=1.68) ([dídan/didán]) higher when accented than its unaccented counterpart. For the final syllable, the difference is 2.41 ST (SD=1.24) and 2.80 ST (SD=2.31) for [tɒ́beʃ/tɒbéʃ] and [dídan/didán], respectively. Panels (b) present the contours in the focal condition, which show a comparable pattern. The initial syllable here is 4.56 ST (SD=1.66) ([tɒ́beʃ/tɒbéʃ]) and 4.21 ST (SD=1.36) ([dídan/didán]) higher, and the final syllable is 3.37 ST (SD=1.71) ([tɒ́beʃ/tɒbéʃ]) and 2.74 ST (SD=1.04) ([dídan/didán]) higher when accented than when unaccented. Panels (c) present the contours for the post-focal condition. A different pattern is observed in the low F0 plateau for both minimal pairs, which would appear to show that the post-focal items are deaccented.²⁴

Turning to the effect of factive verb construction, we observe a similar pattern of deaccentuation for both minimal pairs. While a considerable F0 difference exists between the members of each minimal pair in the non-presupposed condition (panels (a) in Figures 4 and 5), there is a low plateau on the target items in the presupposed condition (panels (b) in Figures 4 and 5). In the non-presupposed condition, the F0 difference between the initial accented and the initial unaccented syllables is 4.51 ST (SD=1.80) ([tɒ́beʃ/tɒbéʃ]) and 3.60 ST (SD=1.79) ([dídan/didán]), whereas for the final syllables the difference is 2.67 ST (SD=1.29) ([tɒ́beʃ/tɒbéʃ]) and 3.16 ST (SD=0.93) ([dídan/didán]).²⁵ Thus, the presupposed condition appears to have the same deaccenting effect as the post-focal condition of contrastive focus.

In order to find statistical evidence for these results, we conducted eight repeated measures ANOVAs. Each of these included data from either the first or the second syllable of the members of a minimal pair from either the focus or the factive part of the experiment, i.e. 2 minimal pairs × 2 syllable positions × 2 information structure constructions. The independent variables were information structure (neutral, focal, post-focal for contrastive focus; non-presupposed, presupposed for factive verb construction) and accent position (initial accent, final accent). The dependent variable was the mean over the six F0 values extracted from the second half of each syllable. This limitation was applied so as to minimize carryover effects of preceding syllables. The analyses are reported in Tables 4, 5, 6, 7. Figures 6, 7, 8, 9 show the corresponding mean values.

Figure 6

Mean F0 values of the 2^nd half of the syllable [tɒ] (panel (a)) and 2^nd half of the syllable [be] (panel(b)) for initial accent and final accent items within the carrier structure [tɒbeʃe] in neutral, focal and post-focal conditions of contrastive focus, pooled over 8 speakers. The values are given in semitones (reference: 100 Hz). Error bars represent standard error of the mean.

Figure 7

Mean F0 values of the 2^nd half of the syllable [di] (panel (a)) and 2^nd half of the syllable [da] (panel(b)) for initial accent and final accent items within the carrier structure [didaneʃ] in neutral, focal and post-focal conditions of contrastive focus, pooled over 8 speakers. The values are given in semitones (reference: 100 Hz). Error bars represent standard error of the mean.

Figure 8

Mean F0 values of the 2^nd half of the syllable [tɒ] (panel (a)) and 2^nd half of the syllable [be] (panel (b)) for initial accent and final accent items within the carrier structure [tɒbeʃe] in non-presupposed and presupposed conditions of factive verb construction, pooled over 8 speakers. The values are given in semitones (reference: 100 Hz). Error bars represent standard error of the mean.

Figure 9

Mean F0 values of the 2^nd half of the syllable [di] (panel (a)) and 2^nd half of the syllable [da] (panel (b)) for initial accent and final accent items within the carrier structure [didaneʃ] in non-presupposed and presupposed conditions of factive verb construction, pooled over 8 speakers. The values are given in semitones (reference: 100 Hz). Error bars represent standard error of the mean.

In all analyses, the main effects of accent position and information structure were found to be significant (with an alpha level set to 0.05) with large effect sizes. Overall, first, accented syllables of target items have significantly higher F0 than unaccented ones, and, second, F0 is significantly lower in post-focal and presupposed target items than in the other conditions. Moreover, all ANOVAs revealed significant interactions of information structure and accent position with relatively large effect sizes, due to the fact that the F0 difference between accented and unaccented syllables is not significant in post-focal and presupposed conditions. These results confirm the expectation expressed above.

3.1 Materials

All response sentences analyzed in Experiment I were included in the corpus of Experiment II. The utterances were divided into two blocks, one of which contained sentences related to contrastive focus, and the other contained sentences related to factive verb construction. This resulted in 160 stimuli. The design is given in Table 8.

3.2 Listeners

A total of 21 listeners, different from the speakers in Experiment I, were recruited, 11 male and 10 female, aged from 17 to 45. They were native speakers of standard Persian, all of whom had obtained a university degree. None of them reported hearing problems and informed consent was obtained from each of them.

3.3 Procedure

The experimental task was presented with a Praat Multiple Forced Choice interface on a laptop (Boersma & Weenink 2014). The stimuli were played through headphones. Listeners were told that they were going to hear a series of sentences, each containing one of the four items [tɒ́beʃ] ‘his swing’, [tɒbéʃ] ‘proper name’, [dídan] ‘they saw’ and [didán] ‘they have seen’, and that they should select which of these items each sentence contained. As in Abolhasanizadeh et al. (2012), these items appeared on the screen in Persian orthography (in the same forms as described in Subsection 2.3) in four clickable boxes.

Within each block, the order of the stimuli was randomized per listener, while the order of blocks was counterbalanced across the listeners. Before the test, listeners were given eight trial stimuli to familiarize themselves with the task. These were selected from the neutral and non-presupposed conditions, in which the target items were accented, so as to facilitate their recognition. All participants indicated that they thought the task was clear to them. They were told that during the experiment they could listen to each stimulus as often as they wished, but that once they had made their choice, it could not be changed.

Participants were tested in different places in Tehran: ten in their home, four in libraries and eight in the studio of the Linguistics Department of the University of Tehran.

3.4 Results

In our analysis, a response was considered correct if the listener identified the item that the speaker in the production experiment was supposed to produce. To obtain a quantitative measure of correct identification, we used d-prime (d’), a sensitivity index used in Signal Detection Theory (Macmillan & Creelman 1991). This is given by the equation in (11):

H is the proportion of correct responses to one stimulus type (hit rate), F is the proportion of incorrect responses to the other stimulus type (false alarm rate), and z() gives the z-score of these variables. The measure d’ is suitable for our purposes, because it eliminates any biases in the response rates that may have arisen due to the decision rules a subject uses. For our identification data, we calculated hits and false alarms for individual minimal pairs in each experimental block. d’ values were obtained for each listener. In general, higher d’ scores indicate better performance. In our analysis, we took d’=1.35 to be the baseline performance. This d’ value corresponds with correct performance on 75% of the trials (Macmillan & Creelman 1991), which is often seen as the correct rate for a just-noticeable-difference (JND) (Durrant & Lovrinic 1977).

The maximum possible d’ score was 3.06 and the range of scores that the listeners achieved extended from –1.15 to 3.06. In line with common practice (Macmillan & Kaplan 1985), where d’ would otherwise be undefined (a hit or false alarm rate of zero or 1), rates of 0 were replaced with [0.5/n], and rates of 1 were replaced with [(n – 0.5)/n], where n is the maximum number of hits or false alarms. For our data, in which n=8, these values were 0.062 and 0.937, respectively.

We report averaged d’ values with corresponding standard errors pooled over 21 listeners in Figures 10 and 11 for contrastive focus and factive verb construction, respectively. There were no missing data in our analysis. Nor was it ever the case that an item was picked from the wrong minimal pair.

Figure 10

Mean d’ scores in an identification task for the minimal pair [tɒ́beʃ/tɒbéʃ] (panel (a)), and the minimal pair [dídan/didán] (panel (b)), as obtained in neutral, focal and post-focal conditions related to contrastive focus, pooled over 21 listeners.

Figure 11

Mean d’ scores in an identification task for the minimal pair [tɒ́beʃ/tɒbéʃ] (panel (a)), and the minimal pair [dídan/didán] (panel (b)), as obtained in non-presupposed and presupposed conditions related to factive verb construction, pooled over 21 listeners.

d’ values for each block were subjected to a separate repeated-measures ANOVA with minimal pair ([tɒ́beʃ/tɒbéʃ], [dídan/didán]) and information structure (neutral, focal, post-focal) as factors for contrastive focus, and minimal pair ([tɒ́beʃ/tɒbéʃ], [dídan/didán]) and information structure (non-presupposed, presupposed) as factors for factive verb construction. Huynh-Feldt corrected p-values are reported where appropriate.

For contrastive focus, the analysis revealed highly significant effects of minimal pair , and information structure , as well as a significant interaction effect between these factors . The Sidak post-hoc test showed that post-focal was significantly different from neutral (p < .001) and focal (p < .001). Taken as a whole, participants performed substantially worse in the post-focal condition than in the other two conditions, and the overall performance with the pair [tɒ́beʃ/tɒbéʃ] was significantly better than that with the pair [dídan/didán]. As indicated by the interaction between the two factors, the effect of focus was stronger on the identifications involving [tɒ́beʃ/tɒbéʃ].

As for factive verb construction, we again found significant effects of minimal pair , and information structure , as well as a significant interaction effect . The effects related to factive verb construction on the recognition are thus hardly distinguishable from the effects related to contrastive focus.

Finally, we used one-sample t-tests to compare different performance conditions against baseline (i.e. d’=1.35). According to the results, the identification of the pair [tɒ́beʃ/tɒbéʃ] was significantly better than the baseline in the neutral (t₂₀ = 5.41, p < .000) and the focal conditions (t₂₀ = 4.63, p < .000) related to contrastive focus, as well as in the non-presupposed condition related to factive verb construction (t₂₀ = 8.39, p < .000). None of the other conditions were significantly above the baseline (see Table 9).

Our finding that the identification scores in the post-focal condition are significantly lower than those in the neutral and the focal conditions is consistent with the perception data presented in Abolhasanizadeh et al. (2012). The post-focal result in that study was however above some chance level in a way that our result is not. They obtained a score of 73% for the post-focal items, which was interpreted as being considerably above the chance level of 50%. In our data the mean scores (percentages correct) for the post-focal and the presupposed items are 54% for post-focal [tɒ́beʃ/tɒbéʃ], 53% for post-focal [dídan/didán], 53% for presupposed [tɒ́beʃ/tɒbéʃ] and 50% for presupposed [dídan/didán].

The results of Experiment I and Experiment II thus converge to suggest that accentual contrasts are neutralized in post-focal regions and presupposed dependent clauses. Given that there were no filler items in any of the two tasks, it is very likely that participants were aware of the minimal pair contrasts. It is remarkable that, despite the obviousness of the contrasts, speakers and listeners failed to differentiate between initial and final accent conditions in the post-focal and presupposed conditions (the unsuccessful identification of [dídan/didán] in all experimental conditions will be discussed in Subsection 4.1).

4.1 The main result

The current study found that Persian word accent is deletable. Members of accentual minimal pairs become homophonous within post-focal regions and presupposed dependent clauses. This result is in accord with claims in the traditional Persian grammars that accent contrasts are neutralized in such contexts.

Our finding disconfirms the results of Abolhasanizadeh et al. (2012). Because their participants recognized the members of similar minimal pairs above chance, they tentatively concluded that constituents retain the accent after the focus. The explanation of the discrepancy most probably lies in the method of data collection. Instead of a contextualized elicitation task, they adopted a reading task in which the focused constituents were presented in bold print. The lack of an appropriate context may have led some speakers to pronounce the target items merely with more or less emphasis rather than with focus prosody. Note that focus, a grammatical concept, need not co-occur with (paralinguistic) emphasis. We return to this point below.

There is one finding in our data that was unexpected. While both minimal pairs ([tɒ́beʃ/tɒbéʃ] and [dídan/didán]) were found to be deaccented in the post-focal condition (related to contrastive focus) and in the presupposed condition (in dependent clauses of factive verbs), the overall recognition score was significantly lower for the pair [dídan/didán]. In fact, it did not even reach the baseline level. The question is why listeners performed poorly with the accented forms of this pair in both the neutral/focal conditions and the non-presupposed condition. Participants cannot have had any difficulty identifying accent locations, given that Persian speakers are widely exposed to the contrastive function of accent and will immediately spot incorrect accent placements. This position is supported by high recognition scores for the members of the minimal pair [tɒ́beʃ/tɒbéʃ]. We suggest that participants must have had some difficulties with the association of target items with the response buttons, despite the fact that they all indicated that they had mastered the association after the trial stimuli. Their confusion may have arisen because the colloquial minimal pair [dídan] ‘they saw’ vs. [didán] ‘they have seen’ involves a grammatical contrast between past and perfect forms of the same verbal stem, for which speakers may have a less clearly defined intuition than for lexical contrasts. Recall that these items appeared on the screen in their literary forms ([didand] vs. [dideand]). The members of the other minimal pair, by contrast, represented different lexemes with clearly different meanings. Moreover, there is no difference between the colloquial and literary realizations of these items, unlike the situation for [dídan/didán].²⁶

4.2 Focus accent vs. neutral accent

One observation in our production data (Experiment I) concerns the question of whether accent for focus is phonologically different from neutral accent in Persian. As noted in Subsection 1.3, previous experimental studies have shown that focused constituents are realized with a higher F0 (Sadat-Tehrani 2007; Taheri & Xu 2012; Hosseini 2013). It has been suggested that the extra high on (contrastively) focused elements might be phonological rather than phonetic (Scarborough 2007). To verify this idea, we show contrastive focus contours against neutral contours for each member of the two minimal pairs separately in Figure 12, which gives time-normalized mean curves for [tɒ́beʃ] (panel (a)) and [tɒbéʃ] (panel (b)), while Figure 13 does the same for [dídan] (panel (a)) and [didán] (panel (b)).

Figure 12

Mean F0 contours for [ún tɒ́beʃe] (panel (a)) and [ún tɒbéʃe] (panel (b)) on a normalized time scale for neutral and focal pronunciations separately, pooled over 8 speakers. The values are given in semitones (reference: 100 Hz).

Figure 13

Mean F0 contours for [unɒ́ dídaneʃ] (panel (a)) and [unɒ́ didáneʃ] (panel (b)) on a normalized time scale for neutral and focal pronunciations separately, pooled over 8 speakers. The values are given in semitones (reference: 100 Hz).

Our data do not confirm the claim that the accent for focus is consistently higher than the neutral accent, suggesting that F0-raising is not a consistent mechanism to mark focus in Persian. Perceptual data reported by Taheri et al. (2014) support this view of F0-raising for focus as a paralinguistic and optional speaker strategy, though these authors did not provide any discussion. Taheri et al. (2014) conducted a perception task in which Persian listeners were asked to identify the position of the focused item in a five-word sentence. While the mean and maximum F0 of all the focused constituents in their corpus were substantially higher than that of their neutral counterparts, the identification scores were lower for the sentence-final items (59%) compared to the non-final items (79%). This suggests that absence of PFC, which is not available in sentence-final position, inhibited the recognition of the focus for sentence-final target items. Since identification errors in final position were biased towards neutral focus (73%), this result supports the conclusion that there is no phonological difference between the accent for focus and neutral accent in Persian and that the two sentences have the same phonological structure.

To sum up, focus is marked by post-focal compression, such that elements with neutral focus and narrow focus are homophonous (see also the discussion in Subsection 4.3).

A final observation in Figures 12 and 13 is that the pitch register of utterances with narrow focus was unexpectedly lower than in neutral utterances, as is visible in Figure 13 (panel (b)), where the focus contour appears to be shifted downward with respect to the neutral contour. This use of a lower register in focus utterances may be interpreted in terms of Ohala’s (1983) Frequency Code (FC), which is based on the fact that smaller larynxes produce higher fundamental frequency than larger ones, causing high pitch to sound uncertain or submissive and low pitch to sound authoritative and assertive. Conceivably, speakers used a lower register to sound more authoritative in contrastively focused utterances compared to neutral utterances. This observation is consistent with the findings of Kügler & Genzel (2012), who show that Akan’s H and L tones have lower pitch under corrective focus.²⁷

4.3 Focus vs. emphasis

The reason why the earlier literature has frequently reported extra high F0 for Persian focus may be due to experimental procedures that confound the effect of focus with that of (paralinguistic) emphasis. In such procedures, participants are instructed in a way that may give them the impression that the task is about emphasizing words, i.e. pronouncing words with a great energy, resulting in higher F0, higher intensity and greater duration. In many cases, experimenters present participants with bold or underlined letters in target items. Moreover, it is frequently the case that during the training sessions, the experimenters use the Arabic loanword [taʔkid] ‘emphasis’ to refer to focus, as for example in the case of Abolhasanizadeh et al. (2012) (Mahmood Bijankhan, p.c.). As we noted in Subsection 2.3, we did not highlight target items in our experiment, instead relying on context sentences to elicit the intended focus structure. In a series of pilot experiments, in which we did use bold print for target items, we found that some participants pronounced them at a higher-than-average pitch. To illustrate the distinction between the two procedures, we present four realizations of the target sentence in (12) in Figure 14.

Figure 14

F0 contours of four realizations of the sentence [nilio landan didim] ‘We saw Nili in London’ as variations of focus and/or emphasis, spoken by a male speaker. Panel (a) has focus on the whole sentence and no particular emphasis on any word. Panel (b) has focus on [landan] and no particular emphasis on any word. Panel (c) has focus and emphasis on [landan]. Panel (d) has focus on [landan] and emphasis on [nili].

The realization of (12) shown in panel (a) of Figure 14 was produced in response to What happened?, while that in panel (b) was spoken in response to Where did you see Nili? and has narrow focus on [landan] ‘London’. Neither was spoken with particular emphasis. To illustrate the prosodic effect of emphasis, panels (c) and (d) give realizations of (12) with focus for [landan], obtained using our contextual elicitation procedure. In panel (c), we observe F0-raising on the focused item compared with the pronunciation in panel (b). The contour in panel (d) shows F0-raising is not necessarily applied to the focused item, since here non-focused [nili] has a significantly raised F0 peak as compared to the other contours in Figure 14 and to the focused item in the same utterance. These data suggest that the realization of focus and the application of emphasis are independent. Focus is indicated by PFC, regardless of the presence of emphatic pronunciation before the focus.

4.4 Prosodic vs. morphosyntactic expression of focus

The prompts and stimuli in our experiments illustrate prosodic focus, meaning that there is no marking of focus other than through prosody. Karimi (2005) in particular has argued that there is a morphosyntactic strategy to express (contrastive) focus through the use of exclusive focus adverbs like [faɢat] ‘only’, which in the usual cases appears to the left of the focus constituent, as in (13), where the object [junɒn] ‘Greece’ represents the focus constituent.

Karimi claims that contrastive focus is realized either morphosyntactically or prosodically, which suggests that the focus for [junɒn] in (14) can either be expressed through deaccentuation of the verb [bord], as in (14a), or by attaching the focus adverb to the object, as in (14b).

However, the use of a focus adverb like [faɢat] ‘only’ in no way absolves the speaker from employing prosodic focus marking in the form of post-focus deaccentuation, as shown in (15) with its F0 contour depicted in Figure 15 (cf. panel (c) in Figure 1 which gives the F0 contour of (14a)). Prosodic focus marking is thus obligatory and independent of the morphosyntactic focus marker.

Figure 15

F0 contour of [ɒlmɒ́n faɢát junɒ́n=o bord] ‘It was only Greece that Germany beat’, spoken by a male speaker.

Moreover, prosodic focus may be the only way in which the scope of the adverb is signalled, as shown in (16). The focus adverb here takes scope over the preceding subject, since the elements after the focus adverb are deaccented.²⁸²⁹

These observations strongly suggest that adverbs like [faɢat] cannot express focus independently of the prosodic focus marking.³⁰ This finding, along with our observations in Subsection 1.3 regarding strict morphosyntactic constraints on the realization of prosodic focus, highlight the close link between prosodic focus and morphosyntactic structure in Persian.