2.1 The phrasal approach

A potential path for implementing the idea described above is to account for pitch accent placement through phrasal prominence. Motivated by observations in English and several other languages, the idea would be that words are those that are prominent in their respective phrases receive pitch accents. Depending on the specific implementation of the idea, “phrase” could be construed as syntactic phrase or (more commonly) Phonological Phrase, which is a phonological entity above the word level that only partially corresponds to the syntactic phrase and constitutes a level of the Prosodic Hierarchy (Selkirk 1980; 1984; Nespor & Vogel 1986). Under theories that adopt this approach, the position of nuclear stress may simply be determined as the last word of the sentence (or, more precisely, of the Intonational Phrase) that bears phrasal stress.

Let us now see how the position of phrasal stress inside each phrase is determined in phrasal approaches. In some theories, phrasal stress would be assigned to an edgemost word within a phrase (e.g. Chomsky & Halle 1968; Halle & Vergnaud 1987; Hayes & Lahiri 1991; Kahnemuyipour 2003). For instance, the word “rug” in the English phrase “the old red rug” would receive phrasal stress because English assigns phrasal stress on the right end of the phrase. A more language-neutral version of the phrasal approach (Truckenbrodt 1995; 2019) relies on the constraint “Stress-XP” and tends to assign stress to the elements that are most deeply embedded in nested phrase structures. Roughly stated, Stress-XP requires each phrase to contain one beat of stress. Moreover, it is assumed that phrasal stress “is assigned minimally but enough to satisfy Stress-XP for all XPs” (Büring & Truckenbrodt 2021). Since a stress on an inner phrase satisfies the constraint on outer phrases too, the constraint may be minimally satisfied by just a stress on the innermost element in a nested phrasal structure. For instance, in [X [Y [Z]]], a stress on Z is sufficient to satisfy the constraint for all three phrases. This successfully captures the cross-linguistic observation that in both [V DP] and [DP V] constructions, nuclear stress appears inside the object DP. In the simplified structure in (4), this mechanism correctly predicts a stress on the object.

In spite of its appeal, the phrasal approach does not seem to be suitable at all for handling stress above the word level in Persian. This is not immediately evident by looking at the way Persian nuclear stress behaves, but becomes clear in light of the fact that the concept of phrasal stress hardly has any manifestation in this language. Unlike many languages, Persian generally does not exhibit any kind of rhythmic alternation throughout the sentence. This is clearly shown in Sadat-Tehrani’s (2007) experiment-based study and has been highlighted in (Rahmani 2019: Ch. 3) and Gussenhoven (2022). Factoring out a few special cases of sentence-medial deaccenting,⁴ all lexical words preceding nuclear stress are equally accented. Outside of the verbal domain, in virtually all environments where one expects a sequence of words to form a single Phonological Phrase given common assumptions regarding the prosodic hierarchy, we observe that every lexical word receives its own pitch accent and no hierarchy of prominence is formed. A few examples are shown (5).

The items of interest in each sentence are the words before the nuclear stress. Example (5a) features a regular possessive construction. Possession is expressed in Persian with the possessor following the possessee with the morpheme e (called the ezafe morpheme) intervening between them. The possessor and the possessee both receive pitch accents (see Hosseini 2014 for examples and discussion). Example (5b) shows that a noun and its PP modifier both receive pitch accent. Finally, in examples (5c–e) we see that numerals, quantifiers, and demonstratives receive pitch accent and are therefore accentually as strong as the noun they modify. The same stress patterns are reported for similar constructions in the experimental study by Sadat-Tehrani (2007).⁵ In all of these examples, the existence of the accent on the modifying words can be confirmed by replacing them with unaccented prepositions and making a comparison. For instance, replacing ye zendâni ‘a prisoner’ with bâ zendâni ‘with the prisoner’ results in a visibly different pitch contour since the first pitch accent would be lost.

Obviously, the above facts do not make the prosodic hierarchy (or the concept of Phonological Phrases in particular) irrelevant in Persian phonology in general. What these facts suggest is that the question of whether a word is deaccented or receives a pitch accent in Persian cannot be answered through phrasal prominence. This makes phrasal prominence unusable for determining the position of nuclear stress.

In spite of what was said, Kahnemuyipour (2003) attempts to offer a phrasal account of the type described earlier for Persian. However, the success of this theory is limited to the verbal domain (Kahnemuyipour 2018; Rahmani 2019). The only type of nominal phrases Kahnemuyipour (2003) invokes for supporting his analysis are DPs containing demonstratives. For these, he reports accent patterns that do not match those of Sadat-Tehrani (2007: 123) and example (5f), but this turns out to be the result of an inaccuracy in reporting the data as explained in Rahmani (2019: Sec. 3.4.3).

Another related theory of nuclear stress that needs to be mentioned is that of Cinque (1993), which ultimately relies on phrasal prominence as characterized in Halle & Vergnaud (1987). Roughly put, the main intuition behind Cinque’s (1993) theory is that nuclear stress is assigned to the most deeply embedded element in the sentence. With a logic similar to what we saw in the case of one of its successor theories, i.e. Stress-XP, this predicts nuclear stress to go on the object in both [V DP] and [DP V] constructions. However, unlike Truckenbrodt’s (1995; 2019) Stress-XP theory, Cinque (1993) does not identify the position of nuclear stress based on it being the last pitch accent in the sentence and does not require all phrases to contain “beats of stress”. For instance, in a hypothetical structure such as (6), Cinque’s (1993) theory predicts C to receive nuclear stress in the most general case, since it is the most deeply embedded element. However, according to Truckenbrodt’s (1995; 2019) Stress-XP theory, to guarantee that every phrase receives a beat of stress, each of C and D must bear stress, and nuclear stress is eventually assigned to D as the rightmost element among the two.

It is not possible to present a detailed comparison of these theories and their predictions here. We may mention in passing that a few problems with Cinque’s proposal have been discussed in the literature (e.g. Kahnemuyipour 2009: 36–42), mostly related to the difficulties that arise in situations where a non-final constituent (e.g. the subject) has a multi-layered internal structure that is deeper than the rest of the sentence (for a discussion of how Cinque addresses these cases and the potential issues, see Zubizarreta 1998: 86–88). Moreover, as Kahnemuyipour (2009) has noted, this theory cannot account for the position of nuclear stress in Persian sentences involving manner adverbs (see Section 3.4). In fact, two other major theories that inherit important elements of Cinque’s (1993) work (i.e. Zubizarreta 1998; Legate 2003) suffer from the same problem in dealing with Persian manner adverbs (see Kahnemuyipour 2009 for a detailed discussion of the topic).

2.2 The phase-based approach

As far as the Persian data are concerned, the most successful theory of nuclear stress is that of Kahnemuyipour (2004; 2009). In this theory, Kahnemuyipour departs from the phrasal approach of Kahnemuyipour (2003) and argues for a purely syntactic theory of nuclear stress relying on phase theory. Adopting the common assumption that only CPs and (transitive or unergative) vPs constitute syntactic phases, his argument is that sentential stress is assigned to the highest element in the complement of a phase head. In this theory, the stress assigned at the vP phase is stronger than the one assigned at the CP phase.⁶ In practice, this puts nuclear stress on the first element of the verbal domain in most Persian sentences. Crucially, this analysis is sensitive only to the surface syntactic form. Adopting Kayne’s (1994) antisymmetry, he assumes that objects merge to the right of verbs in Persian but move higher to the specifier position of AspP. The Persian verb, however, does not move to the v head in his theory. This is shown in (7).

Let us see how this theory works in action. In the sentences in (8), the complement of the v head according to Kahnemuyipour (2009) is enclosed in brackets. Only nuclear stress is shown.

In (8a), the highest element in the complement of the v head is the verb itself. In (8b), it is the object. Thus, the appearance of nuclear stress on the verb in (8a) and the object in (8b) is correctly predicted. In (8c), relying on a vast literature on acc-marked objects in Persian and elsewhere, Kahnemuyipour assumes that the acc-marked object (marked by the morpheme ro) has moved out of the verbal domain, leaving the verb as the highest element in the complement of the v head and thereby allowing it to receive nuclear stress. The contrast between (8d) and (8e) follows smoothly from the assumption that manner adverbs are inside the verbal domain⁷ while temporal adverbs are not. Again, in both cases, nuclear stress falls on the first element inside the verbal domain. In (8f), the PP tu tale “in trap” receives nuclear stress because it is an argument of the verb and therefore inside the verbal domain. Conversely, the PP in (8g) does not receive nuclear stress because it is not an argument and falls outside of the verbal domain. Finally, in the unaccusative construction in (8h), there is no vP phase under the CP, making the subject the highest element in the complement of the first phase head (i.e. the C head).

Kahnemuyipour’s (2009) theory is strictly surface-sensitive. This allows him to account for the apparent difference (but see Section 4.2) between the behaviors of non-acc-marked and acc-marked objects in (8b) and (8c) relying on their different surface positions and regardless of their identical merge positions. Similarly, this allows Kahnemuyipour (2009) to adopt Kayne’s (1994) theory of antisymmetry and assume that Persian’s surface SOV structure starts off as SVO. Since it is only the surface structure that matters, the non-acc-marked object may receive nuclear stress as the highest element in the complement of the v head even though it is quite low at merge under antisymmetric assumptions.

It is now time to examine potential drawbacks of Kahnemuyipour’s (2009) theory and justify the need for an alternative approach. Conceptually, it seems problematic that in Kahnemuyipour (2009) and its expansion in Kratzer & Selkirk (2007), there is no systematic explanation for why the very elements that steal nuclear stress from the verb in Persian (objects, PP arguments, and manner adverbs) tend to do so in English too. These elements have different surface positions in the two languages and therefore the phase-based approach assigns them nuclear stress for different reasons in Persian and English. For instance, in (8b), the object “pizza” receives nuclear stress in the equivalent English sentence too even though it follows the verb. Kahnemuyipour’s (2009) explanation for the English pattern is that the verb is outside of the verbal domain in English. Thus, the similarity between the two languages is accidental. A similar situation holds in examples (8d) and (8f) with manner adverbs and PP arguments respectively. I believe that these correspondences are not accidental and need to find a systematic explanation in a successful theory of nuclear stress.

A related issue concerns the relationship between finality and nuclear stress. In Kahnemuyipour’s (2004; 2009) theory and its revised version in Kratzer & Selkirk (2007), no such systematic relationship is acknowledged. Thus, it seems to be almost an accident that in a vast range of English sentences the last element receives nuclear stress (as is the case in all of the examples in 8). The importance of finality in nuclear stress assignment is discussed in more detail in the Section 4.

Putting aside these theoretical considerations, Kahnemuyipour’s (2009) theory suffers from at least three empirical issues in relation to Persian. The first issue is that contrary to what it predicts, the same word receives nuclear stress in scrambled and non-scrambled sentences in Persian, suggesting that the surface position of elements is not all that matters in nuclear stress assignment. This is discussed in Section 3. The second issue is that once we eliminate confounding information structure effects, we observe that contrary to what Kahnemuyipour (2009) and a few other scholars have assumed, specific and non-specific objects do not have different stress patterns. This is argued for extensively in Section 4.2. Finally, Kahnemuyipour (2009) does not discuss a large class of PP arguments that follow the verb. In Sections 4.4, I show that the stress behavior of these PPs in Persian does not match Kahnemuyipour’s (2009) predictions as they receive nuclear stress even though there is compelling evidence showing that the highest element in the verbal domain is the verb itself.

3.1 Nuclear stress in scrambled sentences

We may now begin looking at the interaction between scrambling and nuclear stress in Persian. Persian allows a wide range of scrambling options. This relatively free word order offers new insights into the behavior of stress above the word level in the language. Let us begin by looking at a few different ways the elements in a sentence can be scrambled, as shown in (10).

The first sentence shows the unmarked order. The other sentences are scrambled and vary in their degrees of markedness. However, all of them are certainly passable (unlike permutations such as *maryam hamishe mikhore pitzâ ârum), and the stress patterns given in (10) are indeed the most neutral stress patterns given these word orders. In other words, for none of these word orders is there another stress pattern that sounds more or equally neutral. The examples in (10) do not cover all acceptable permutations, but they must be sufficient for giving a sense of the general situation, which can be described roughly in the following two generalizations:

1. Across different permutations, nuclear stress is always on the same element.

2. Elements following the sentence’s nuclear stress are always deaccented.

To show that this pattern is not due to some form of focus effect specific to manner adverbs, another set of examples is given in (11) and (12). In both cases, the (a) sentence shows the unmarked order. In these sentences, nuclear stress is always on the direct object. To make the evaluation of the situation according to the phase-based theory easy, the complement of the v head is shown with brackets in the (a) examples.

The adverb diruz ‘yesterday’ in the first example is clearly not a verbal argument and falls outside of the complement of the v head. In fact, suggesting that it is inside the vP causes problems for the phase-based theory because in that case it is expected to receive nuclear stress in (11b). Progressive auxiliaries like dâram in (12) are also known to be outside the vP, as they normally appear before non-argument PPs (Ghomeshi 2001; Kahnemuyipour 2003). Thus, we observe that in both scrambled sentences (the (b) examples), the non-acc-marked object receives nuclear stress even though it has moved out of the vP.

The fact that the same element receives nuclear stress regardless of the ordering in these examples poses a challenge to surface-based theories. To save these accounts in the face of the present data, one might be tempted to suggest that the (b) sentences in (11) and (12) have narrow focus on the elements bearing nuclear stress, as Karimi (2005: 131–133) suggests for similar cases in Persian and Sato (2012) does with respect to similar data in Japanese. However, at least in Persian, this does not seem to be the case. It is true that the fronting of the object serves a (relatively elusive) pragmatic role in the discourse, but it does not involve focus on the object. All of these sentences can be uttered in contexts where focus on the entire event is expected. For instance, regarding (12b), consider the imaginary dialogue in (12b).

It must be noted that in the dialog in (13), as in most other contexts, the unscrambled order (as in 12b) is the unmarked form. But what matters to us here is the fact that the response in (13) is felicitous too. Given that this word order with a nuclear accent on the object is acceptable even when the context forces a wide focus on the entire event, it is clear that this sentence’s use is not limited to cases where there is narrow focus on the object. Putting the nuclear stress on any of the the two other words in (13) is completely unacceptable.

A theory of nuclear stress that relies on stress reconstruction can easily account for the data at hand. If an element that receives nuclear stress does so by virtue of its base position, it is not surprising for it to keep its nuclear stress wherever it moves because its unpronounced copy is still at the stress-attracting position. This is shown schematically in (14) for a sentence of the form in (11) with a temporal adverb and a non-acc-marked object.

To account for the same facts under a surface-based theory, one must probably argue that the movements under discussion constitute PF movement. However, note that everything following the nuclear accent is always deaccented. Thus, the choice of what to deaccent depends on the order created by these movements (e.g. in (11)–(12) some words receive pitch accent only when there is no scrambling). Therefore, under a PF movement scenario, it must be stipulated that nuclear stress assignment occurs before PF movement but the choice of what words receive regular pitch accent and what words are accentless occurs after PF movement. This is shown in (15) for the same sentence type as (14).

The item of interest here is the temporal adverb. If we simply assign regular pitch accent to all words preceding nuclear stress as soon as nuclear stress is assigned (i.e. the second line in 15), the temporal adverb must receive a pitch accent too, contrary to fact. Thus, it seems inevitable to separate the assignment of nuclear stress and the assignment of other pitch accents under this approach, as done in (15). Given these complications, it seems reasonable to argue that in the absence of other motivations, the reconstruction account is preferable.

4.1 Objects without case marking

Non-acc-marked and acc-marked objects show different syntactic behaviors in Persian. This section looks at non-acc-marked objects (acc-marked objects are discussed in Section 4.2). In general, the non-acc-marked object precedes the verb and attracts nuclear stress. In the absence of an object, however, the verb receives nuclear stress. We explain the nuclear stress on the object via movement. Assuming Kayne’s (1994) antisymmetry, we expect the object (being a complement) to merge to the right of the verb.

In (16a), nuclear accent is predicted to be sentence-final since there are no order-changing movements (note that a string-vacuous series of movements does not lead to changes in the stress pattern). In (16b), the lower copy of the object is the lowest element in the tree. The pronounced copy of the object receives stress by virtue of the position of its lower copy. This account unifies the stress facts of Persian and SVO languages like English by attributing the presence of nuclear stress on the object to its sentence-final base position.

It is worth noting that there are certain verbs (or certain uses of them) for which a sentence-final object is considered unmarked. In these sentences, nothing is deaccented and the object receives nuclear stress, as shown in (17).

This matches our predictions. In a sentence where the surface order matches the base order, a sentence-final stress is predicted. Kahnemuyipour (2009) does not discusses such cases in Persian. However, it seems that for his account to work, it must be assumed that the verb has moved out of the verbal domain. It must be mentioned that these data are easier to account for under Kratzer and Selkirk’s (2007) version of the phase-based theory. They argue that nuclear stress goes on the highest phrase in the complement of the v head. Since the verb is not a phrase, they correctly predict the object rather than the verb to receive stress here (without having to assume that the verb has moved out of the verbal domain).

4.2 Case-marked objects

Persian features differential object marking. According to a common view (Browning & Karimi 1994; Karimi 1999; Kahnemuyipour 2009), the presence of the accusative marker denotes specificity in objects. However, other factors such as animacy have been argued to play a role in Persian object marking too (see Samvelian 2018 for a review). The accusative marker is the enclitic ro (râ in Formal Persian), which is usually pronounced as o after consonants. acc-marked objects tend to precede manner adverbs and adjunct PPs. Thus, most generative syntacticians maintain that acc-marked objects are positioned outside of the verbal domain (Browning and E. Karimi 1994; S. Karimi 1999; Kahnemuyipour 2003; S. Karimi 2005).

It is often reported that in sentences with acc-marked objects, nuclear stress goes on the verb rather than the object (Vahidian Kamyar 2001; Kahnemuyipour 2009; Rahmani 2019). An example provided by Kahnemuyipour (2009) (with minor modifications in formatting and glossing) is given (18).

This is in contrast with the behavior of non-acc-marked objects, which generally receive nuclear stress and leave the verb unaccented, as shown in (19).

At first sight, these data pose a challenge to our theory. Our theory predicts the object to receive nuclear stress in information-structurally neutral sentences regardless of how high in the tree it has moved. It is beyond dispute that given (18) in written form and asked to read it aloud, Persian speakers normally read it with nuclear stress on the verb khord ‘ate’. Nevertheless, I argue in this section that this is the result of the fact that speakers normally attribute a non-neutral information structure to sentences of this form. I demonstrate that the stress on such sentences in focus-neutral contexts is in fact on the acc-marked object, just like non-acc-marked objects. In other words, this is a case where what speakers may identify as the “normal” or “default” accent pattern of a sentence does not correspond to truly wide focus.

The information-structural factor that I argue to be relevant here is topichood. In particular, acc-marked objects are normally treated as topics and therefore reject nuclear stress. However, once we enforce a reading where such topicalization does not occur, the focus becomes truly wide and acc-marked objects no longer reject nuclear stress. To see that this is the case, we must first review certain key assumptions regarding focus.

In the following discussion, I make a few simplifying assumptions that do not interfere with the issue at hand. First, the focus domain is assumed to be the entire sentence. Second, the focus is assumed to be a single constituent. Note that only one word within the focus receives nuclear stress. Finally, I concentrate only on cases where no independent givenness effects interfere with the prosodic structure.

As famously proposed by Jackendoff (1972), we may treat focus as a feature present in the syntactic representation. A single sentence with a fixed stress pattern can be ambiguous in terms of what its focus is. In Chomsky’s (1971) famous example in (20), for instance, even though nuclear stress is fixed on “shirt”, the sentence can be interpreted with the focus being any of the bracketed constituents.

Depending on which of the constituents a through d we take to be the focus, one of the four responses in (21) becomes felicitous.

There is consensus on the simple generalization that at least under the simplifying assumptions we have adopted, the nuclear stress of the focus domain must be inside the focus. As for the more interesting question of where in the focus the nuclear stress appears, things get more complicated in fringe cases, but under our assumptions, we can confidently rely on an important generalization by Jackendoff (1972), which was later adopted by many others (e.g. Cinque 1993; Zubizarreta 1998; Büring 2016: and implicitly by Kahnemuyipour 2009). According to this generalization, there is no mechanism beside the language’s default stress assignment mechanisms for determining which word in the focus receives nuclear stress.

Looking at this from the opposite direction, we may conclude that the words that create focus ambiguity inside a constituent when they receive nuclear stress are those that are given prominence inside that constituent by the language’s default stress assignment rules. For instance, the ambiguity in (20) shows that the default stress assignment mechanisms of English give the word “shirt” prominence in any of the constituents a through d. As we shall see below, this generalization gives us a powerful tool to determine the default stress mechanisms of a language when interference from information structure effects is too subtle to easily eliminate.

With this introduction, we may return to the question of Persian acc-marked sentences. In (22), the word sequence used in the potential responses corresponds to Kahnemuyipour’s (2009) example in (18). Here, the question forces a focus on the entire activity that serves as an alternative to playing. Of the two potential responses, only the one with nuclear stress on the object is felicitous. The author’s judgment is confirmed by a survey of five Persian speakers,¹⁰ in which four respondents preferred the (a) answer and one gave both (a) and (b) the perfect score.

Relying on Jackendoff’s generalization, we may argue that it is the “default” rules that are responsible for assigning nuclear stress to the acc-marked object rather than the verb. Crucially, a nuclear accent on the verb is allowed only when there is narrow focus on the verb. The asymmetry between the verb and its object clearly shows the special status of the object as the rightful bearer of “default” stress.

Before looking at more examples, let us see how the stress pattern Kahnemuyipour (2009) observes for (18) and its contrast with the data in (22) must be explained. It seems undeniable that examples such as (22) reflect truly wide focus, meaning that cases such as (18) are the ones with non-neutral information structure in spite of sounding more “normal” in most situations. Identifying the exact information-structural mechanism that leads to (18) behaving in the way it does is not part of the immediate goal of this paper and I do not discuss it in detail, but it seems to suggest that the object rejects nuclear stress in (18) because it is topicalized. Many researchers have argued for a close connection between acc-marked Persian objects and topicality (e.g. Peterson 1974; Ghomeshi 1997; Dalrymple & Nikolaeva 2011; Samvelian 2018).¹¹

A somewhat similar situation holds in Turkish. Turkish acc-marked objects are similar to their Persian counterparts and have similarly been argued to mark specificity (Enç 1991) and move out of the vP (Diesing 1992; Zidani-Eroğlu 1997; Kelepir 2001). Interestingly, earlier work on Turkish reported that acc-marked objects do not receive nuclear stress (Orgun & Inkelas 2005), but it was shown later by authors such as Üntak Tarhan (2006) and Nakipoğlu (2009; 2019) that such cases all involve information-structural interference.

Given the similarities between Turkish and Persian in differential object marking and intonational patterns, the above point makes the claim being made in this paper regarding acc-marked objects receiving nuclear stress in the default case more plausible. Nevertheless, the situation is not necessarily identical in the two languages. I rely on topicalization in my account and assume that elements can be topics without being given in the discourse (e.g. see Jäger 2001; Kratzer & Selkirk 2007 for analyses with the same assumption). Thus, the stress-rejecting acc-marked objects in Persian are not necessarily discourse-given. This can explain why even isolated sentences presented to speakers in written form are typically read without nuclear stress on the object, even though no relevant discourse exists in such situations. This seems to be different from the case in Turkish, where presupposition has been argued to be what interferes with object nuclear stress assignment. As Dolatian (2022) shows in a comparative study, Persian speakers may place nuclear stress on the verb even in situations where Turkish and Western Armenian speakers place nuclear stress on the acc-marked objects.

It is now time to present more examples demonstrating the stress behavior of acc-marked objects in wide-focus contexts. The interaction between focus and acc-marked objects can be seen more clearly when a focus operator such as faqat ‘only’ is used. For (23), all of the survey’s respondents preferred (a) to (b) and gave it a perfect score.

In analyzing these sentences, we assume that the part of the sentence that is the scope of the operator ‘only’ receives focus (e.g. see Jackendoff 1972; Krifka 1992). The acceptability of (23a) shows that a nuclear stress on the acc-marked object (leaving the verb deaccented) allows a reading in which the focus is on the entire sequence qazâ ro khord ‘ate the food’. On the other hand, in (23b), the utterance is unnatural presumably because a nuclear stress on the verb signals that the verb has narrow focus. As another set of examples, two dialogues are presented in (24). We are primarily interested in comparing the acceptability of the third response in (a) with the third response in (b), given the different questions. All five respondents to the survey prefer the (b) case and give it the perfect score (i.e. 3). The average score for the (a) case is 0.8.

In each dialogue, the three responses correspond to the following three different focuses: the verb (‘washing’), the object (‘the car’), and the combination of the object and the verb (‘washing the car’). In (24a), only the second response is available, showing that when nuclear stress is on the verb, the only available reading is a focus on the verb. In (24b) where the nuclear stress in the question is on the object, the sentence can be interpreted to have focus on the object (the first response) or on the sequence containing both the object and the verb (the third response). This means that the question sentence in (24b) is the one with wide focus, confirming that in the absence of interfering information-structural factors, the acc-marked object receives nuclear stress and leaves the material after it deaccented.

Even with a verb such as azyat kardan “to tease”, which is not expected in the context of the question, the same effect is observed, as shown in (25). For this example, all five respondents prefer the first response and give it a perfect score. The average score for the other response is 1.

Given existing assumptions regarding the position of the Persian acc-marked object, if we accept that acc-marked objects receive nuclear stress just like non-acc-marked objects in information-structurally neutral contexts, the phase-based theory faces a problem. In the phase-based theory, acc-marked objects must not receive nuclear stress given that they are outside of the verbal domain in the surface structure. To save the theory in the face of this issue, one may argue that acc-marked objects in these examples are low enough to fall inside the complement of the v head. This is what Üntak Tarhan (2006) suggests for Turkish acc-marked objects in order to reconcile the Turkish data with this theory (see Nakipoğlu 2019 for an objection). While not impossible, this proposal goes against Kahnemuyipour’s assumption that manner adverbs mark the left edge of the verbal domain, since acc-marked objects generally precede manner adverbs in both Turkish (Üntak Tarhan 2006) and Persian. A Persian example is given in (26), in which the object precedes the manner adverb bad (=“bad”) and still receives nuclear stress.

As before, it must be noted that nuclear stress normally goes on the manner adverb in the presence of an acc-marked object, because the normal reading in such sentence does not have wide focus. What triggers the nuclear stress going on the object in this particular example is that the event is contrasted with other events, thereby enforcing wide focus. Four of the survey’s respondents gave this sentence the perfect score, i.e. 3, and one gave it 2 in our four-value 0–3 scale.

4.3 Manner adverbs

Let us now see how manner adverbs are accounted for in the present theory. Manner adverbs are known to behave differently from higher adverbs. In English, they generally appear after the verb whereas other adverbs precede it, as shown in (27).

The appearance of manner adverbs after the verb is not a peculiarity of English. In fact, manner adverbs generally follow the verb in languages with VO surface order (Dryer 1992). In line with the framework of antisymmetry and the assumption that VO is the base order, I assume that the post-verbal position reflects the base position of manner adverbs, even in languages where it has a different surface position. It has long been argued (Larson 1988 building on the work of McConnell-Ginet 1982, also see Kayne 1994: Ch. 7; Alexiadou 1997) that manner adverbs are base-generated deep inside the verb phrase as a complement. In his work on phrase stress, Cinque (1993) uses this analysis to account for the stress facts of English and Italian. The adverb in (27a), on the other hand, is an adjunct (or the specifier of a functional head) above the v head, thus it is expected to be generated on the left of v under the assumptions of antisymmetry.

In Persian, unlike English, all adverbs precede the verb. However, if manner adverbs are base-generated lower than the verb as discussed above, our theory would predict it to receive nuclear stress. This is indeed the case, as shown in (28).

In (28a), the manner adverb receives nuclear stress because its lower copy has the lowest position in the tree. In (28b), the adverb is not a manner adverb. We may assume that it is adjoined to the left of the verb phrase, and therefore its base position matches its surface pre-verbal position. (28c) features both a manner adverb and an object preceding the verb. Here too, we assume that the English (and cross-linguistically common) surface order reflects the base order in Persian, meaning that the manner adverb is base-generated low and thereby attracts nuclear stress.

As discussed in Section 2.2, the phase-based theory of Kahnemuyipour (2009) predicts the presence of stress on Persian manner adverbs with no problem. This theory is sensitive only to the surface position of constituents. Even though Kahnemuyipour (2009) do not consider the manner adverb to have been generated lower than the verb, he does assume that the manner adverb (unlike other adverbs) is inside the verbal domain. Thus, the manner adverb receives stress as the highest element in the stress-receiving domain (i.e. the complement of v). However, unlike Persian, the stress on manner adverbs in English, shown in (29), poses a potential challenge for the phase-based theory.

At least at first glance, it is not clear why the manner adverb receives nuclear stress in (29) under the phase-based theory. After all, in these sentences, the highest element in the complement of the v head seems to be the verb itself. Kahnemuyipour (2009) argues (following Cinque 2004) that in English sentences with sentence-final manner adverbs, the verb and its adverb have both moved out of the complement of v. This means that his theory’s regular stress-assignment mechanism cannot not assign nuclear stress to any element in the sentence. The stress we see on the manner adverb, according to him, is the result of a backup mechanism that is activated when the complement of v is phonologically null, placing nuclear stress on the closest element to the complement of the v in such situations. Given the absence of independent motivation for this backup mechanism and the fact that the cases of Persian and English manner adverbs find unrelated accounts in his theory, his solution does not seem attractive.

As mentioned earlier, in the modified version of the phase-based theory presented by Kratzer & Selkirk (2007), the stress in the vP phase goes on the highest phrase (rather than the highest constituent) in the complement of v. This solves the issue in (29a), but not (29b). In (29a), since the verb is not a phrase, the highest phrase in the verbal domain is the manner adverb. In (29b), however, the highest phrase in the verbal domain is presumably the object, but nuclear stress goes on the manner adverb. A similar problem facing this theory concerning post-verbal PPs is already discussed by Shaw 2009 and Rochemont 2013.

4.4 Argument PPs

As Kahnemuyipour (2009) has rightly pointed out, there is a contrast in stress behavior between argument and adjunct PPs in Persian. While argument PPs receive nuclear stress and leave the verb unaccented, adjunct PPs reject nuclear stress. This is demonstrated in the examples in (30) (repeated from (8).

Similar to the case of adverbs, to explain the above facts, we need to assume that the two types of PPs have different base positions. In (30a), the PP tu arusi ‘in the wedding’ is an adjunct and therefore base-generated to the left of the verb. Therefore, it does not have a lower copy after the verb and does not receive nuclear stress. In (30b), the PP tu tale ‘in trap’ is an argument of the verb. Therefore, it is generated to the right of the verb, which means that its lower copy earns it the nuclear stress.

For our account to be convincing, it is helpful to have further motivation for the proposal that the PPs that receive nuclear stress are indeed generated to the right of the verb. Fortunately, there is independent evidence for this in Persian, coming from directional and goal PP arguments. These arguments usually appear after the verb, i.e. in the position which we take to be their base position. The stress pattern of sentences of this type has been described by Sadat-Tehrani (2007), but has not been discussed by Kahnemuyipour (2003; 2009). Even in the case of (30b), it would be at least as acceptable to place the PP after the verb. Further examples are given below.

In Formal Persian, unlike Tehrani Persian, goal PPs usually precede the verb (and usually require prepositions, unlike 31a and 31d). In such sentences, the adverbial receives nuclear stress and the verb is left unaccented. In our theory, the fact that goal PPs have the same stress status regardless of their position is expected. The formal equivalents of the first two sentences in (31) are given in (32).

The problems faced by Kahnemuyipour (2009) in explaining the stress facts regarding sentence-final PPs in colloquial Persian (and English) are similar in nature to the ones mentioned in the previous subsection for sentence-final manner adverbs in English and have been discussed by Shaw (2009: 30) and Rochemont (2013). Kahnemuyipour (2009) does not mention these Persian examples, but he does mention sentence-final PPs in English. His solution for those is the same as the one he offers for sentence-final manner adverbs. Thus, his theory can account for these sentences if we assume that the verb and its argument are both outside of the verbal domain and that in the absence of a pronounced element inside the verbal domain, a backup mechanism assigns nuclear accent to the last element in the sentence. However, even this convoluted solution runs into problems when goal PPs and manner adverbs appear together, as in (33).

Recall that Kahnemuyipour’s (2009) account requires the verb and the PP to have moved outside of the vP in (31a). But in (33), we observe that the verb and the PP follow the manner adverb and are both deaccented, suggesting that they are inside the vP under his assumptions. Thus, unless through far-fetched movement scenarios, it seems impossible for his theory to account for (31a) and (33) simultaneously.¹² It is worth noting that the version of the phase-based theory proposed by Kratzer & Selkirk (2007), which avoids assigning nuclear stress to verbs because they are not phrases, does not suffer from a problem in these cases.

4.5 Complex predicates

In Persian, the vast majority of verbal constructions come in the form of complex predicates, constructions composed of a non-verbal element (NVE) which bears the main semantic content of the verb and a conjugated light verb (see Karimi 1997; Folli et al. 2005; Megerdoomian 2012, inter alia for thorough discussions of the syntax of complex predicates in Persian).

It is not always easy to tell apart an NVE from a regular non-acc-marked object. Nevertheless, in spite of the apparent difficulty in categorizing many particular instances, the conceptual distinction seems to be solid (Megerdoomian 2012). In any case, the analysis of the sentences in (34) seems straightforward. Like objects, an NVE is the complement of the light verb and is therefore merged to its right. This explains its stress behavior. A more interesting case occurs when NVEs and non-acc-marked objects co-occur in a sentence, as in the examples below.

In analyzing these sentences, we follow the proposal offered by Folli et al. (2005). Building on the work of Hale & Keyser (2002) on argument structure and relying on how the interaction between the identity of the NVE and the verb determines the argument and event structure of complex predicates in Persian, they argue that in sentences of this type the object and the NVE merge together to form an AP with the NVE adjective serving as the head. In the account presented by Folli et al. (2005), antisymmetry is not assumed and the merge positions match the surface positions. However, we may modify it to comply with Kayne’s (1994) ordering. For this purpose, the light verb, being a head, must precede the AP. Similarly, inside the AP, the adjective NVE must precede the object DP. Thus, for (35c), we get the pre-movement arrangement shown in (36).

This means that the order of the three elements at merge is the reverse of their surface order. Under our assumptions, this explains the assignment of nuclear stress to the object. The situation after movement is shown below. The object receives nuclear stress because its lower copy is the last element in the sentence. I do not discuss the mechanics of the movements and the exact landing sites.

Let us now briefly discuss cases involving nuclear stress on the light verb. So far in this subsection, I argued that default nuclear stress goes on an element in the complement of the verb in the base structure. This implies that whenever the stress is on the verb itself, there is narrow focus on the verb. This even includes cases where the verb does not have an independent relevant meaning.

The stress pattern in (38a) represents the default case. In (38b), the stress is on the verbal element of the complex predicate, even though the complex predicate’s meaning is non-compositional, making focus on the verb unexpected at first sight. However, the English translation (which is normally read with narrow focus on the auxiliary “does”) helps in revealing the nature of the phenomenon. It seems clear that it is not the lexical meaning of the verb migarde, but its quality as a verb, presumably through having tense or aspect, that qualifies it for narrow focus in (38b) (see Sadat-Tehrani 2017 for a detailed demonstration of the facts and a different analysis). The prosodic role of the morpheme ke, which is brought up by an anonymous reviewer, can be understood in the same light.

A more rough translation of the first clause in (39) would be “As far as returning is concerned, she does return.” Thus, we may argue that when ke attaches to a constituent, it places it outside of the focus. The mechanics of this behavior are not simple, but given the English translation and what we saw in (38b), it should be clear that the morpheme’s pragmatic role affects the information structure of the sentence. The additive morpheme ham seems to have a similar role, but I refrain from discussing it here due to its more complicated nature.

4.6 Ditransitives

The stress pattern of sentences involving both PP arguments and objects is relatively complicated, and judgments are not clear in most cases. I start with the cases where the object is non-acc-marked. For these, giving examples like (40), Kahnemuyipour (2009) reports that nuclear stress goes on the direct object.

For Kahnemuyipour (2009), this stress pattern is predicted given that the object is the highest element in the complement of the v head according to standard assumptions. Our theory’s prediction depends on where we assume PP arguments to merge. Following the work of Barss & Lasnik (1986) and more importantly Larson (1988), it has usually been assumed that the PP argument merges lower than the direct object. For instance, in an English sentence such as “Mary gave the book to John.”, the low surface position of the PP “to John” is assumed to reflect its low merge position. If we adopt this assumption for Persian, which is also in accordance with Kayne’s proposal (1994: 70–71), we expect nuclear stress to go on the PP argument, contrary to what we see in (40). As we shall see, the facts are more complicated than what (40) suggests and if anything, support our assumptions.

First of all, it must be noted that if we replace “the table” with a phrase that is not easily presupposed in any discourse, the pattern in (40) becomes unacceptable.

It seems that (41) is felicitous only in situations where the PP is given. To express this proposition, two other sentence types, shown in (42), sound natural at first glance. All five survey respondents assign the score 3 (“perfect”) to both sentences in (42) but the average score for (41) is 1.6 (the respondents’ individual scores are are 1, 1, 1, 2, and 3).

These sentences place nuclear stress on different elements (one puts it on the PP argument and one on the object). It is not immediately clear which one must be viewed as the default case as far as stress and information structure is concerned. To find the answer, we may use contexts where focus on the entire event is enforced as we did in Section 4.2. The somewhat surprising result is that none of the stress patterns is felicitous for a situation where wide focus is needed. Our respondents’ average scores for (43) and (44) were as low as 0.2 and 0.6 respectively. Two other versions with the nuclear stress on the object but different word orders (Obj PP V, Obj V PP) were also given to the respondents and both received average scores of 0.4.

While both stress patterns are unacceptable in (43) and (44), a slight difference between the two types of stress pattern emerges in sentences with bare nouns.

The respondents’ average scores for (45) and (46) are 0.8 and 1.8 respectively, showing a preference for a nuclear stress on the PP argument, as our assumptions predict.¹³ Three respondents prefer (46), one prefers (45), and one respondent gives them both a score of 2 (in a 0–3 scale). The preference seems meaningful, but the result is not conclusive since the number of respondents is small and only one respondent finds the preferred pattern (i.e. 46) perfect. It may be worth adding, by the way, that the author’s own intuitions also strongly prefer (46) to (45).

To summarize, to the extent that wide-focus readings are available, they favor nuclear stress on the PP argument, as predicted by our assumptions. Let us now consider what happens when the object is acc-marked. In such cases, the stress is normally on the PP argument, as shown below. Like the other cases we saw in Section 4.4 where the PP argument follows the verb, Kahnemuyipour’s (2009) theory faces issues in this case. I do not repeat the arguments here.

As in the case of non-acc-marked objects, our theory easily predicts the nuclear stress on the PP in these examples. However, it is fair to ask what happens if we use a question-answer setting like we did above to enforce a wide-focus reading in these cases too.

The survey’s results are not completely conclusive, but encouraging. The average scores for the two responses are 1.6 and 2.4 (out of 3) respectively. One respondent finds both responses perfect, three strongly prefer the second response, and one strongly prefers the first response. What is perhaps more important is that four out of five respondents give the second response the perfect score. Thus, it seems that placing nuclear stress on the PP is indeed the preferred way for marking wide focus in these constructions, as our theory predicts.