Processing Focus in Turkish

Abstract

The immediately preverbal position has been argued to be the default focus position in Turkish. In absence of any overt focus markers, the constituent in this position is considered to carry sentential stress and neutral information for canonical word-order sentences and focus is projected to the whole sentence in the form of broad focus. In non-canonical word-order sentences, the immediately preverbal constituent is presumed to carry focal stress and the focused constituent would receive narrow focus. This paper tested this claim experimentally. The paper also investigated if there were any differences in the cognitive operations associated with processing and revising focus in canonical and non-canonical sentences. There were a sentence completion task and an eye-tracking experiment. The sentence completion data and the eye-tracking data supported the theoretical predictions: the immediately preverbal position was associated with default focus in Turkish when no pitch accentuation or other focus markers were available. The eye-tracking data further showed that changes to word-order were perceived as cues for broad versus narrow focus marking. The participants’ processing of and revision from narrow focus were costlier than processing broad focus and assigning narrow focus for the first time. We argue, in line with previous research, that this may be due to deeper encoding of focused information in memory or heavier memory load resulting from keeping a set of alternatives of the focused constituent when it has contrastive meaning.

1. Introduction

Focus plays a significant role in spoken or written language comprehension (Gussenhoven 2007) as it indicates new or emphasized information in a sentence (Chomsky 1971; Jackendoff 1972). There are different bifurcations of focus in the literature. It is described as broad versus narrow focus based on its scope (Gussenhoven 2007; Kahnemuyipour 2009) and informational versus contrastive focus (Göksel and Özsoy 2003; İşsever 2003; Reichle and Birdsong 2014) based on its meaning. In a sentence with broad focus, a constituent’s focus is neutral and its importance is equal to other constituents in the sentence. In a sentence with narrow focus, a particular constituent is focused and it conveys more prominent information than the other constituents in the sentence (Lowder and Gordon 2015). Narrow focus can be informational when new information is introduced, or contrastive when a constituent’s truth value is contrasted with the other elements in its semantic alternatives set (Rooth 1992). This study aims to manipulate focus through its scope (broad vs. narrow) and the target words are assumed to carry neutral information in sentences with broad focus and focused information in sentences with narrow focus.

In both broad focus and narrow focus, the focused constituent is phonologically marked through stress. Broad focus is generally associated with sentential stress which is assigned to the highest element in the stress domain in a sentence with canonical word-order (Kahnemuyipour 2009). When a constituent has sentential stress, focus is projected to the whole sentence. Then, the whole sentence, not the stressed constituent, conveys new information (Gussenhoven 2007). Narrow focus, on the other hand, is associated with focal stress which can be assigned to any constituent in a sentence (Kahnemuyipour 2009), and in that case, the stressed constituent alone conveys prominent information.

In Turkish, broad focus (presentational focus or p-focus) is considered to be marked in the immediately preverbal position when no other constituent bears prominent information and when the sentence has canonical word-order (Göksel and Özsoy 2003; İşsever 2006; Özsoy 2019). To mark narrow focus, different strategies, such as prosodic, morphological or syntactic, are utilized across languages (Büring 2010). Turkish employs both syntactic and prosodic means for narrow focus marking. Some argue that narrow focus can be marked through pitch accentuation in situ or through syntactic movement which makes the immediately preverbal position focused; but these two strategies cannot be used at the same time (Erguvanlı 1984; İşsever 2003; Kılıçaslan 1994). It has also been claimed that Turkish can mark focus in situ in any preverbal position regardless of scrambling (Göksel and Özsoy 2000; Özge 2003). It has been shown in other research (Stolterfoht et al. 2007) that readers (or speakers) create expectations for focus positions in the languages that they speak. Given the previous work (e.g., Erguvanlı 1984), it is possible that the immediately preverbal position could be the expected focus position in Turkish. This study investigates if that is so, in absence of other markers of focus such as pitch accentuation, background context or focus particles. More specifically, we examine whether or not the immediately preverbal position in Turkish is associated with broad focus (i.e., neutral information) in canonical word-order sentences (Göksel and Özsoy 2000, 2003; İşsever 2006) and with narrow focus (i.e., focused information) in non-canonical word-order sentences (Erguvanlı 1984; İşsever 2003; Kılıçaslan 1994). This was examined through a sentence-completion task and an eye-tracking experiment.

The present study also examines processing and revision of information which has varying degrees of focus (neutral vs. focused in the current study). The mental operations associated with different degrees of focus have been argued to be due to the depth of encoding focused information into memory and heavier memory load for higher degrees of focus (Benatar and Clifton 2014; Lowder and Gordon 2015; Reichle and Birdsong 2014). Lowder and Gordon (2015) for instance found that as the level of focus increased (focused > neutral > defocused), the readers’ fixation times also increased presumably due to the deeper encoding of the focused information. Similarly, Benatar and Clifton (2014) showed that recovery from increased focus degree (i.e., revision of contrastive focus compared to informational focus) caused longer reading times. The present study examines processing and revision of broad and narrow focus with these predictions in mind through what we consider as a firm manipulation of focus marking. Previous studies investigating focus processing manipulated focus marking through changes to background information (Benatar and Clifton 2014; Reichle and Birdsong 2014), use of it-cleft/pseudo-cleft sentences (Lowder and Gordon 2015) or focus particles (Stolterfoht et al. 2007). But these manipulations provided either differential context, added new words or changed the syntactic complexity of sentences which may have induced processing differences that are not all necessarily due to focus type. The present study utilizes word-order manipulations in Turkish to induce broad and narrow focus. This ensures that the background context, sentence complexity, words and the number of words across comparison conditions remain the same while only the focus structure is manipulated.

The study was conducted through reading, and as such we do not specifically examine assignment or interpretation of pitch accentuation, but we concur with Fodor (2002) in that “in silent reading, a default prosodic contour is projected onto the stimulus, and it may influence syntactic ambiguity resolution. Other things being equal, the parser favors the syntactic analysis associated with the most natural (default) prosodic contour for the construction.” (p. 113). Eye-tracking measures have been shown to be sensitive to projection of prosody in reading in the form of lexical stress (Ashby and Clifton 2005; Breen and Clifton 2011) and focus (Kentner and Vasishth 2016). We thus believe that the eye tracking data can also inform on how Turkish readers project focus in canonical and non-canonical sentences.

2. Focus Marking in Turkish

As mentioned above, Turkish marks focus through syntax via changes to the word order (Erguvanlı 1984; İşsever 2003; Kılıçaslan 1994), through prosody via pitch accentuation (Göksel and Özsoy 2000; Özge 2003), through focus particles such as bile ‘even’ (Göksel and Kerslake 2005) or de/da ‘also’ (Göksel and Özsoy 2003) and background context (Özsoy 2019). The present study investigated how robust the changes to word-order were for focus marking in Turkish; therefore its details will be highlighted below.

Kahnemuyipour (2009) argues that two rules, namely Sentential Stress Rule (SSR) and Focus Stress Rule, interact in marking focus. Sentential Stress Rule (SSR) applies for sentences with canonical word order in a neutral discourse. In such contexts, sentential stress is assigned to the highest element in the stress domain. A constituent with sentential stress has neutral focus; it does not have more emphasis or importance than the other constituents in the sentence; and its focus projects to the whole sentence and the sentence conveys new information and carries broad focus. Focus Stress Rule (FSR) applies to non-neutral contexts and non-canonical word-order sentences. In these contexts, focus is not projected to the whole sentence. The focused constituent receives focal stress and it carries the most prominent information in the sentence.

As mentioned above, in Turkish canonical word-order sentences sentential stress is marked on the immediately preverbal constituent (Göksel and Özsoy 2003; İşsever 2006). As such, the immediately preverbal constituent is assumed to carry broad focus where the whole sentence conveys new information if no other constituent is marked for focus through an overt focus marker (e.g., pitch accentuation) or if there is no background information (Özsoy 2019; see presentational focus in Göksel and Özsoy 2003).

In a non-canonical sentence in Turkish in which elements are moved out of the focus domain and thus are de-focused (İşsever 2003), the constituent in the immediately preverbal position cannot carry sentential stress as the context is not neutral. In such cases, if there is no other overt focus marker for another constituent, the immediately preverbal constituent is perceived to carry narrow focus and it cannot project its focus to the whole sentence (Erguvanlı 1984; İşsever 2003, 2006; Kılıçaslan 1994). When narrowly focused, the immediately preverbal constituent either conveys new information (Göksel and Özsoy 2003) or makes a contrast against the alternatives in its semantic set (İşsever 2003, 2006; Rooth 1992).

The experimental sentences in the present study had ditransitive verbs and they had either canonical or non-canonical word orders. The theoretical literature does not seem to agree on a default, i.e., canonical, word order for ditransitive verbs Kornfilt (2003), for instance, considers it to be Direct Object (DO)-Indirect Object (IO) order, whereas Georgala (2011) argues that it is IO-DO. Göksel and Kerslake (2005) and Öztürk (2005) claim that both orders are acceptable and attribute preferences for a certain order to the information structure (Göksel and Kerslake 2005) or animacy of the indirect object (Öztürk 2005). We considered DO-IO order as the canonical order of accusative marked direct object and indirect object for ditransitive verbs given Kornfilt’s (2003) analyses (see p. 140) and Kurt and Bakay’s (2021) experimental evidence. Kurt and Bakay (2021) asked their participants to form sentences with a ditransitive verb and its three arguments where the animacy of the objects were controlled and no background context was available. Their results showed that 88% of the sentences were produced with a DO-IO order.

No background context or overt focus markers were used. Thus, the canonical word-order sentences were predicted to receive broad focus assignment; and the immediately preverbal constituent in these sentences would carry neutral information while the whole sentence would convey prominent information. Non-canonical word order sentences were predicted to yield narrow focus projection on the immediately preverbal constituent, not the whole sentence; and the immediately preverbal constituent would carry focused information.

3. Previous Work on Focus Processing

Previous work examined processing of different focus degrees from the perspectives of information encoding (Lowder and Gordon 2015), memory load (Reichle and Birdsong 2014), revision of focus structure (Stolterfoht et al. 2007) and revision of information type (Benatar and Clifton 2014). Their findings overall showed that both processing and revision of higher degrees of focus (focused vs. neutral or contrastive vs. informational) were costlier than those of lower degrees of focus.

In an eye-tracking experiment Lowder and Gordon (2015) reported that focused constituents in English were processed differently from neutral or defocused constituents. There were three conditions in their experiment, namely focused as in (1a), neutral as in (1b) and defocused as in (1c). In the focused conditions they used pseudo-cleft constructions to assign narrow focus to a target item, the official memo in (1). In the neutral conditions the sentences had a declarative form with broad focus, in which the target item, the official memo, received neutral focus and the whole sentence conveyed prominent information. In the defocused conditions, another constituent, the secretary in (1), was focused through it-clefting and the target item was defocused.

(1).	a.	Focused: What the secretary typed was the official memo about the new office policy.
	b.	Neutral: Yesterday the secretary typed the official memo about the new office policy.
	c.	Defocused: It was the secretary that typed the official memo about the new office policy.
		(Lowder and Gordon 2015, p. 1736)

The eye-tracking data showed that as the focus level of a constituent increased (focused > neutral > defocused), the participants spent more time to read it. Lowder and Gordon proposed that “the longer times on linguistically focused information reflect deeper encoding” (Lowder and Gordon 2015, p. 1738) of that information and deeper integration to its context.

Reichle and Birdsong (2014) investigated, in an ERP study, processing of two types of narrow focus (contrastive and informational) in French as the first language (L1) and the second language (L2). As was briefly mentioned above, contrastive focus is when narrow focus evokes other semantic alternatives in mind and emphasizes the truth value of the focused constituent among its alternatives. Informational focus, however, refers to the conditions where the focused constituent only carries new information. They manipulated experimental context through questions such as (2a) and (2b), so as to have the target sentence such as (2c) carry either contrastive focus or informational focus. When the target sentence in (2c) followed a question such as in (2a) it would carry contrastive focus. When it followed a question such as in (2b) it would carry informational focus. The focused constituent was presented in capital letters.

(2).	a.	Contrastive Focus:
		Est-ce que c’est un verrre ou ın marteau qu’ on voit sur la table?
		‘Is it a glass or a hammer that we see on the table?’
	b.	Informational Focus:
		C’est quoi qu’on voit sur la table?
		‘What do we see on the table?’
	c.	Target Sentence:
		C’est un marteau qu’on voit sur la table.
		‘It’s a HAMMER that we see on the table.’
		(Reichle and Birdsong 2014, p. 13)

The participants read question-response pairs such as (2a,c) or (2b,c) silently. The results for the L1 French group indicated that the contrastive focus conditions caused an increase in LAN (left anterior negativity) elicitation which is associated with an increase in working memory load (Cowles 2003). This was not the case for the informational focus conditions. Reichle and Birdsong (2014) attributed the higher cost of contrastive focus to its heavier burden on working memory because “contrastive focus requires a set of referents to remain active in working memory…” and it “…should therefore place a greater burden on working memory compared to informational focus” (p. 5).

Benatar and Clifton (2014) investigated if there is a difference between encoding new information for the first time and revision of the previously focused information. The experimental items in the study involved dialogues between two speakers as in (3a,b). The questions addressed by Speaker A would cause the target item, John, in Speaker B’s responses to carry either informational focus as in (3a) or contrastive focus as in (3b). In conditions with informational focus such as (3a) the readers would need to add new information (John) to the discourse for the first time, whereas in contrastive focus conditions such as (3b) they would need to contrast and replace old information (Mary) with new information (John).

(3).	a.	Speaker A: Did you tell someone to go home early?
		Speaker B: I told John, but I don’t know if it was a good idea.
	b.	Speaker A: Did you tell Mary to go home early?
		Speaker B: I told John, but I don’t know if it was a good idea.
		(Benatar and Clifton 2014, p. 8)

The eye-tracking data showed higher processing cost for contrastive focus conditions than informational focus conditions because in contrastive focus conditions content required to update the discourse representation. That is, the information from the previous discourse needed to be “amended or changed … in the case of corrective (i.e., contrastive) focus” (p. 10).

Stolterfoht et al. (2007) examined, in an ERP study, processing of and revision to focus structure (broad and narrow focus) in German. Their experimental sentences consisted of a main clause which had either wide focus (i.e., broad focus) or narrow focus and of a replacive phrase such as und nicht den Lehrer ‘not the teacher’ which contrasted with a constituent in the main clause and required, in some conditions, revision of the main clause focus structure. The sentences were manipulated for broad or narrow focus through the use of a focus particle (nur ‘only’) in the main clause as in (4a,b) and (4c,d), respectively. The sentences without a focus particle such as (4a,b) were broad focus conditions. The sentences with a focus particle such as (4c,d) were narrow focus conditions in which the focus particle would mark focus on the subject or the object. The replacive phrase contrasted either with the object as in (4a,c) or the subject as in (4b,d). In (4c,d) it was congruous with the constituents marked by the focus particle and did not require any revisions either to the focused constituent or to the focus structure. In (4a) the replacive phrase was congruous with the focused constituent, so no revision to the focused constituent was required; revision to the focus structure of the main clause (from broad to narrow) was necessary. In (4b) it was incongruous with the focused constituent and required revision to the focused constituent (from object to the subject) and revision to focus structure (from broad to narrow).

(4)	a.	No focus particle, Object-replacive: Am Dienstag hat der
		Direktor [den SCHÜler]F getadelt, und nicht [den LEHrer]F
		‘On Tuesday, the principal criticized the pupil, and the principal did
		not criticize the teacher.’
	b.	No focus particle, Subject-replacive: Am Dienstag hat [der
		DiREKtor]F den Schüler getadelt, und nicht [der LEHrer]F
		‘On Tuesday, the principal criticized the pupil, and the teacher did not
		criticize the pupil.’
	c.	Focus particle, Object-replacive: Am Dienstag hat der Direktor
		nur [den SCHÜler]F getadelt, und nicht [den LEHrer]F
		‘On Tuesday, the principal criticized only the pupil, and the principal
		did not criticize the teacher.’
	d.	Focus particle, Subject-replacive: Am Dienstag hat nur [der
		DiREKtor]F den Schüler getadelt, und nicht [der LEHrer]F
		‘On Tuesday, only the principal criticized the pupil, and the teacher
		did not criticize the pupil.’
		(Stolterfoht et al. 2007, p. 574 )

Revisions to the focus structure (from broad to narrow) were predicted to yield positive-going waveform as a correlate of focus structural revision it required; revisions to the focused constituent were predicted to elicit negative-going waveform as a correlate of prosodic changes. A bilateral positivity (350–1100 ms) was observed for conditions that required focus structure revisions from broad to narrow but not for conditions that already had narrow focus. A bilateral sustained positivity (350–1100 ms) and a widely distributed negativity (450–650 ms) were observed for conditions that required revisions to both focus structure (from broad to narrow) and the focused constituent (from object to subject). The authors attributed the positive-going waveforms observed for the conditions that required focus structure revisions to syntactic reanalysis. The negative-going waveforms observed for conditions that required revisions to the focused constituent were associated with implicit prosodic revision during silent reading.

Overall, the previous studies showed that as the degree of focus of a constituent increased, its processing and revision also became costlier. The present study aims to contribute to this literature with data from Turkish. To our knowledge, there has been no previous work testing how focus is processed in Turkish. In the present study, we first tested the theoretical claims on default focus position in Turkish through a sentence completion task. We followed up with an eye-tracking experiment which examined the processing and revision of focus in canonical and non-canonical word-order sentences in Turkish. The experimental sentences included ditransitive verbs which allowed for focus manipulation through word-order thanks to the alternation in object position. We believe that manipulation of focus structure through word order can control for any possible effects of contextual or structural differences across comparison conditions. The observed effects could then be attributed only to the focus structure manipulations.

4. The Present Study

The present study investigated focus assignment in Turkish through a sentence completion task and an eye-tracking experiment. The sentence completion task was included to test the default focus position in Turkish. The eye-tracking experiment investigated the real-time comprehension and revision processes of focus in canonical vs. non-canonical word order, which would be associated with broad and narrow focus (i.e., neutral and focused information), respectively. The eye-tracking methodology can inform on the cognitive processes that take place during early and later stages of processing (Pickering et al. 2004; Roberts and Siyanova-Chanturia 2013). It also provides a relatively more natural reading experience compared to other methods such as self-paced reading (Conklin and Pellicer-Sánchez 2016; Rayner and Sereno 1994). As mentioned above, the experimental sentences included ditransitive verbs, which would allow us to use word-order to manipulate focus structure (See Brown et al. 2012 for a similar strategy in English).

4.1. The Sentence Completion Task

The sentence completion task examined whether or not the immediately preverbal position would be perceived as the default focus position in Turkish. The test sentences employed ditransitive verbs which would allow for the alternation in object position. In canonical sentences, the indirect object was in the immediately preverbal position. The prediction was that it would receive sentential stress and its focus would be neutral as the focus would be projected to the whole sentence in the form of broad focus. In non-canonical sentences the direct object occurred in the immediately preverbal position. It was predicted to receive focal stress, would be narrowly focused and deliver prominent information. The experimental materials did not include any background context.

4.1.1. Participants

Thirty-eight native speakers of Turkish (24 females), with a mean age of 20.76 (SD = 2.19), participated in the sentence completion task. They were bachelor or master’s students at Boğaziçi University. The bachelor students received course credit for their participation; the master’s students did not receive any compensation. Informed consent was obtained from all the participants before data collection. The participants were naïve to the purpose of the study. They all had normal or corrected-to-normal vision.

4.1.2. Materials

The experimental sentences included a main clause with a ditransitive verb and a partially blank replacive phrase such as ….. değil (‘….. not’) which could be completed with a constituent that could contrast with the subject, the direct object or the indirect object in the main clause. The participants were presented with the experimental sentences as in (5a,b) and were asked to choose the best option to complete the sentence with one of the arguments of the verb as in (5c) below. Any of these arguments could potentially receive focus assignment. The experimental sentences had either canonical word order as in (5a) or non-canonical word order as in (5b).

(5)	a.	Canonical Word-Order
		Defne Hanım	çiçeğ-i	yönetmen-e	ver-di, _______	değil.
		Defne Miss-NOM	flower-ACC	director-DAT	give-PST.3SG	not
	b.	Non-Canonical Word-Order
		Defne Hanım	yönetmen-e	çiçeğ-i	ver-di,_______	değil.
		Defne Miss-NOM	director-DAT	flower-ACC give-PST.3SG		not
		‘Miss Defne gave [the flowers to the director]/[the director the flowers], not ______.’
	c.	Cümledeki boşluğu hangi kelime en iyi şekilde tamamlar? ‘Which word would complete the sentence the best?’
		○ aktör-e (actor- DAT) ‘to the actor’			(indirect object)
		○ ödül-ü (the award-ACC) ‘the award’			(direct object)
		○ Selin Hanım (Selin Miss- NOM) ‘Miss Selin’			(subject)

It was predicted that the participants would choose to complete the sentence with a constituent that would contrast with the immediately preverbal constituent, which would be the indirect object in canonical order sentences and the direct object in non-canonical sentences. If so, default focus in Turkish would be confirmed to be associated with the immediately preverbal position. The options also included the subject. It was predicted that assigning focus to the subject (i.e., the sentence-initial position) would be the least preferable as per the given-before-new principle (Halliday 1967; Clark and Haviland 1977) which maintains that focused information is generally expected in later positions in a sentence. Unless they are focused through syntactic movements in the focus domain, Turkish subjects occur in sentence initial positions. Thus, the sentence-initial subject position in Turkish was expected to carry given information (Erguvanlı 1984). The subject option was included to provide data for that.

There were 24 experimental sentence sets, each in two versions as illustrated in (5a,b). These were distributed across two lists controlling for word-order (canonical vs. non-canonical). The order of the direct object, the indirect object, and the subject were counterbalanced in the options in order to prevent any bias. Each list consisted of 24 experimental sentences and 36 fillers. 24 of the fillers aimed to disguise the experimental sentences. 16 of the disguise fillers included a ditransitive verb in the main clause, but their replacive phrase emphasized the truth value of the whole main clause, rather than a specific constituent. (e.g., Aylin arkadaşına barmeni gösterdi, ki bu ________ değil ‘Aylin showed the bartender to her friend, which is not _____’). These fillers were intended to prevent the participants from expecting a narrow focus reading in the main clause in the experimental sentences. The eight of the disguise fillers had a replacive phrase requiring a narrow focus reading in the main clause, but the main clause did not include a ditransitive verb. These fillers were included to prevent expectation of a ditransitive verb in the main clause (e.g., Mehmet çarşıda lise arkadaşına rastladı, ______ değil. ‘Mehmet came across his high school friend, not ____’). The remaining 12 fillers had varying syntactic structures.

4.1.3. Procedure

The sentence completion task was designed using the online platform IBEX Farm (https://spellout.net/ibexfarm/ accessed on 30 September 2021). The participants were sent an e-mail with a link for the task. In the task, the participants saw each experimental sentence with three options. They were asked to choose, with a mouse-click, the best option to complete the replacive phrase. After they responded, the next experimental sentence and three options to complete it appeared. There was no time limit. Note that the sentence completion task was administered two weeks after the eye-tracking experiment since the possibility of recall is higher in an untimed task. But we present the eye-tracking experiment later given the complexity of its test conditions.

4.1.4. Results

The percentage of replacive phrase completion preferences are presented in

Table 1. Mean Percentages of (In)Congruous Completions for the Replacive Phrase Region.

Condition	Replacive Phrase Completion	Percent Preference
Canonical Order	Congruous Incongruous Object Incongruous Subject	65.11 28.22 6.66
Non-Canonical Order	Congruous Incongruous Object Incongruous Subject	58.22 33.33 8.44

. The table shows if the choice was (in)congruous with the constituent in the immediately pre-verbal region.

As seen in Table 1, the participants mostly preferred to complete the replacive phrase with an argument contrasting with the constituent in the immediately preverbal region (M = 61.66%) both for the canonical sentence conditions (M = 65.11%) and for the non-canonical sentence conditions (M = 58.22%). The percentage of incongruous completions was smaller (M = 38.32%). Although the participants chose to complete some portion of the replacive phrases with an incongruous object (M = 28, 22% for canonical order sentences and M = 33.33 for non-canonical sentences), the percent of choices for the subject as the focus position was very low (M = 7.55 %).

A generalized linear mixed model (Marschner 2011) and the R statistical computing software (R Core Team 2020) were used to analyze if word-order predicted sentence completion preferences. Word-order (canonical vs. non-canonical) was entered as the fixed effect and subjects and items were random effects. The analysis showed no effect of word order in replacive phrase completion preferences, β = 0.21, SE = 0.15, t = 1.36, p = 0.17.

4.1.5. Discussion

The results of the sentence completion task confirmed the theoretical arguments in that the immediately preverbal position was associated with focus in Turkish for both canonical and non-canonical word-order sentences. The participants chose to place focus mostly on the immediately preverbal constituent (the indirect object in canonical sentences and the direct object in non-canonical sentences). This was followed by the other object and the subject. Also, as predicted, the subject position was considered as the least possible candidate for focus assignment (Erguvanlı 1984; Topaloğlu and Nakipoğlu 2017). As an anonymous reviewer notes, the higher rate of preferences for a constituent which is in the alternative set of the immediately preverbal object to complete the replacive phrase might be attributed to a recency effect. That is, as the immediately preverbal constituent is the closest to the replacive phrase, its case and alternative set could have biased the replacive phrase completions. Recency effects are usually consequences of memory constraints and are observed in on-line (i.e., initial) parsing of sentences (e.g., Fernández 2002). Considering the off-line nature of the task where the readers had enough time to go back and reach a final decision without any time pressure, we do not think that this would have affected their sentence completion preferences. Regardless, this concern was inspected in the eye-tracking data by comparing the reading times of the first object and the immediately preverbal object (see further details in the results and discussion sections of the eye-tracking experiment).

Overall, although informative on Turkish speakers’ expectations for a default focus position (i.e., immediately pre-verbal region), the results of the sentence completion task do not shed light on whether there were any differences in the types of focus assigned to this position in canonical and non-canonical sentences. That is, the results of the sentence completion task cannot inform on whether the readers projected broad vs. narrow (or neutral vs. focused) information based on the word order of the sentences. Recall that the theoretical literature predicts for canonical word-order sentences to carry broad focus and the immediately preverbal constituent would then have a neutral reading in the absence of any other focus markers. And the sentences with non-canonical word order are predicted to carry narrow focus and the direct object would carry focused information. To examine if this is the case, an eye-tracking experiment was conducted.

4.2. The Eye-Tracking Experiment

The eye-tracking experiment tested whether or not word order manipulations changed the focus structure of Turkish sentences. That is, the experiment tested whether or not the immediately preverbal constituent had broad focus in canonical word-order sentences and narrow focus in non-canonical sentences. This was done through examining the participants’ reading behavior for the immediately preverbal position. In canonical word-order sentence conditions, the immediately preverbal constituent was predicted to carry broad focus and thus neutral information in absence of any overt focus markers or background context. In non-canonical word-order sentence conditions, the immediately preverbal constituent was predicted to carry narrow focus and thus focused information in absence of any other overt focus marker or background context. Recall that the previous research on focus processing found slower reading as the degree of focus increased. The eye-tracking measures are sensitive to the time-course of processing behavior and later measures of eye-tracking such as rereading duration or total duration (see Section 4.2.4 for details) are associated with discourse integration (Roberts and Siyanova-Chanturia 2013). Thus, they can inform on whether or not Turkish readers assign focused or neutral reading to the immediately preverbal region depending on the changes to the word order of the objects.

The eye-tracking experiment also investigated whether or not the immediately preverbal position was perceived as the default focus position in Turkish. This was examined via the participants’ reading behavior at a replacive phrase region that followed each main clause and was either congruous or incongruous with the immediately preverbal constituent. If the participants’ reading times increased for replacive phrases that were incongruous with the immediately preverbal constituent, it would further confirm that this region is the expected focus region in Turkish. The reading behavior at the replacive phrase region would also inform on the mental operations associated with revision of main clause focus structure (broad-to-narrow vs. narrow-to-narrow). The constituent in the replacive phrase region contrasted with one of the arguments of the verbs, requiring a narrow focus reading for that constituent. Thus, for canonical order conditions where the immediately preverbal constituent is predicted to receive a neutral reading due to broad focus assignment to the sentence, a certain constituent would need to be assigned narrow focus for the first time and the focus structure of the main clause would be revised from broad to narrow. For non-canonical order conditions, while the main clause focus structure would still be kept narrow due to narrow focus assignment to the immediately preverbal constituent, either the information that the focused constituent induced would need to be amended or focus would need to be deleted from a previously focused constituent and assigned to a new constituent.

4.2.1. Participants

Forty-three native speakers of Turkish (24 females), with a mean age of 20.67 (SD = 2.07), participated in the eye-tracking experiment. 38 of the participants also took part in the sentence completion task. To prevent recall, there was a two-week interval between the eye tracking experiment and the sentence completion task. The additional five students also reported (corrected-to-) normal vision. They were bachelor students at Boğaziçi University. All the participants were naïve to the purpose of the study.

4.2.2. Materials

The experimental sentences were the same as those in the sentence completion task except that there was no constituent missing in the replacive phrase. The word order of the main clause (canonical vs. non-canonical), the congruity of the replacive phrase region with the immediately preverbal region (congruous vs. incongruous) and focus structure revision (broad-to-narrow vs. narrow-to-narrow) were manipulated. As in the sentence completion task, word order was manipulated by reordering the direct and indirect object to elicit broad or narrow focus projection. In canonical word order, the immediately preverbal constituent was the indirect object, and the readers were predicted to assign broad focus to the sentence since there was no other focus marker available; and neutral reading to the immediately preverbal constituent as in (6a–c). In non-canonical word order, the immediately preverbal constituent was the direct object and the readers were predicted to assign it narrow focus due to the syntactic movement in the focus domain (İşsever 2003) and given the absence of any other focus marker. This would give focused reading to the immediately preverbal constituent as in (6d–f). The constituent in the replacive phrase, ____ değil ‘not ____’, contrasted with one of the arguments of the ditransitive verb in the main clause: the subject, the direct object or the indirect object. When the constituent in the replacive phrase had a contrastive alternative which is in the same semantic alternative set (Rooth 1992) as the stressed constituent (which, in our items was assumed to be the immediately preverbal constituent) there was prosodic congruency as in (6a,d) because the stress would still be on the immediately preverbal constituent in the main clause. When the constituent in the replacive phrase contrasted with a constituent that was not immediately preverbal —the direct object (DO) as in (6b), the indirect object (IO) as in (6e) or the subject (SU) as in (6c,f)—there was prosodic incongruency because the readers would then need to assign stress to another constituent in the main clause. Lastly, in all the conditions, the replacive phrase ensured the truth value of the focused constituent in the main clause as it emphasized that the semantic alternative in the main clause was the true alternative, not the one in the replacive phrase. In canonical word-order conditions as in (6a–c), the readers would need to assign focus to a certain constituent for the first time because at the replacive phrase region the readers would understand that not the whole sentence, but a specific constituent should have been focused as it was contrasted with an alternative in its set at the replacive phrase. In non-canonical word-order conditions as in (6d–f), the focus structure was assumed to be kept narrow, but there would be a revision for the information encoded at the immediately preverbal constituent. In (6d), the readers would receive information that the semantic alternative in the replacive phrase is in the same set as the focused constituent, but the semantic alternative they encoded before (e.g., çiçeği ‘the flower’) would need to be contrasted against the one in the replacive phrase (e.g., ödülü ‘the award’); thereby encoding a new alternative in memory. In (6e,f), the readers would need to delete focus from the immediately preverbal constituent and assign it to a different constituent with a different semantic alternative set, thereby both deleting the focused constituent from memory and encoding a new constituent and its alternatives.

(6)	a.	Canonical/Congruous/Broad-to-Narrow:
		Defne Hanım	çiçeğ-i	yönetmen-e	ver-di,	aktör-e	değil.
		Defne Miss-NOM	flower-ACC director-DAT		give-PST.3SG	actor-DAT not
	b.	Canonical/Incongruous (Direct Object)/Broad-to-Narrow:
		Defne Hanım	çiçeğ-i	yönetmen-e	ver-di,	ödül-ü	değil.
		Defne Miss-NOM flower-ACC		director-DAT	give-PST.3SG award-ACC		not
	c.	Canonical/Incongruous (Subject)/Broad-to-Narrow:
		Defne Hanım	çiçeğ-i	yönetmen-e	ver-di,	sunucu	değil.
		Defne Miss-NOM director-DAT flower-ACC		give-PST.3SG		award-ACC	not
	d.	Non-canonical/Congruous/Narrow-to-Narrow:
		Defne Hanım	yönetmen-e	çiçeğ-i	ver-di,	ödül-ü	değil.
		Defne Miss-NOM director-DAT flower-ACC			give-PST.3SG	award-ACC	not
	e.	Non-canonical/Incongruous (Indirect Object)/Narrow-to-Narrow:
		Defne Hanım	yönetmen-e	çiçeğ-i	ver-di,	aktör-e	değil.
		Defne Miss-NOM director-DAT flowers-ACC give-PAST.3SG actor-DAT					not
	f.	Non-canonical/Incongruous Subject/Narrow-to-Narrow:
		Defne Hanım	yönetmen-e	çiçeğ-i	ver-di,	sunucu	değil.
		Defne Miss-NOM director-DAT flower-ACC			give-PAST.3SG	presenter-NOM	not
		‘Miss Defne gave the flowers to the director/the director the flowers, not to the actor/ the trophy/ the presenter.’

The objects in the experimental sentences were controlled for their animacy (i.e., animate DO-animate IO, animate DO-inanimate IO, inanimate DO-animate IO and inanimate DO-inanimate IO). This was done to balance semantic plausibility for objects as inanimate NPs are more frequent themes than animate NPs (Dahl and Fraurud 1996), and the animacy has been considered and found to be influential in the ordering of objects in Turkish (Kahraman et al. 2010; Kahraman 2013; Öztürk 2005). Each experimental sentence was preceded with a lead-in phrase such as Film festivalinde (At the movie festival,) for a neutral context. Each sentence was also followed up with a three-word, content-neutral sentence such as Organizasyon öyle yapılmış (It was organized that way) to center the replacive phrase region on the screen as readers do not tend to fixate on line-final words (Rayner 1998). A content-neutral follow-up sentence would also provide a spillover region for the replacive phrase region which was a critical region.

24 experimental items, each with six conditions as in (6a–f), were created. They were distributed across six lists counterbalancing for main clause word order (canonical vs. non-canonical) and prosodic congruency of the replacive phrase with the immediately pre-verbal constituent (congruous, incongruous object and incongruous subject). Each list also included 72 fillers, 24 of them were the same as in the sentence completion task except that their replacive phrases were not blank. The remaining 48 fillers had varying syntactic structures. There were also five practice sentences and three warm-up sentences at the beginning of each list. Each sentence was followed with a yes/no comprehension question (e.g., X kelimesini gördünüz mü? ‘Did you see the word X?’) to ensure attention.

4.2.3. Procedure

The data were collected using EyeLink 1000 Plus eye-tracker (SR Research, Mississauga, Ontario, Canada) in a quiet laboratory. The participants were seated in front of a computer that would present the sentences, with their heads stabilized, via a chin rest, 55 cm away from the screen. The experiment began after successful calibration of eye-movements. During the experiment a drift check at the beginning of each sentence further ensured successful calibration. The data from the right eye were recorded.

The sentences were left-aligned on the screen. The participants’ task was to read each sentence at their own pace and answer the comprehension question that followed on a separate screen. The participants moved from the sentence to the question and answered the comprehension question through a mouse-click. The next sentence appeared once they answered the comprehension question. There was no feedback on accuracy. A break was offered half-way through the experiment but no participant took it. The procedure, including calibration and validation, lasted approximately 30 min.

4.2.4. Data Analysis

All the participants were >70% accurate in their responses. The data were first cleaned from fixations longer than 800 ms and shorter than 80 ms (Morrison 1984; Rayner 1978; Rayner and Pollatsek 1989) as it is argued to not take longer than 800 ms to process information (Rayner and Pollatsek 1989) and less than 80 ms does not provide sufficient time for word recognition (Rayner and Pollatsek 1989). R statistical computing software (R Core Team 2020) was used for all the analyses.

The six-standard eye-tracking measures, first fixation duration (FFD), gaze duration (GD), regression path duration (RPD), re-reading duration (RRD), total reading duration (TD) and the probability of regression out (PRO) were analyzed (Clifton et al. 2007; Liversedge et al. 1998; Rayner and Pollatsek 1989; Roberts and Siyanova-Chanturia 2013). FFD is defined as “the duration of the first fixation on a word, provided that the word was not skipped” (Clifton et al. 2007, p. 5). GD is the total amount of fixations on a region of interest before the eyes move from the region (Rayner and Duffy 1986). RPD refers to “the sum of all the fixations from the first fixation in a region up to but excluding the first fixation to the right of this region” (Liversedge et al. 1998, p. 63). RRD is “the regression path reading time for a region less the [gaze duration] for a region” (Liversedge et al. 1998, p. 63). TD refers to “all fixations made within a region of text, including those fixations made when rereading the region” (Liversedge et al. 1998, p. 58). PRO reflects the probability of a regressive saccade of a target region with the aim of visiting a region in the text which was previously read. While FFD and GD are considered to be measures related to the early processes of comprehension such as retrieval of a word (Liversedge et al. 1998; Rayner and Pollatsek 1989), RPD, RRD, TD and PRO are considered to be measures informing on later processes of comprehension such as word integration to the preceding context in the sentence or recovery from a processing difficulty (Liversedge et al. 1998).

Two critical regions and their spill-over regions were analyzed for each eye-tracking measure through mixed effects linear or logistic regression models (Bates et al. 2015) using the lme4 package (Bates et al. 2015). The fourth word, the constituent in the immediately preverbal region was examined as the first critical region; the fifth word, corresponding to the verb, was analyzed as its spill-over region. The second critical region was the replacive phrase region with the sixth and seventh words, DO/IO/SU değil (‘not DO/IO/SU’); its spill-over region was the eighth word, the first word of the context-neutral follow-up sentence.

In the analyses on the first critical region (the immediately preverbal constituent) and its spillover region, word order (canonical vs. non-canonical) was the predictor. In the analyses on the second critical region (the replacive phase region) and its spillover region congruency with the immediately preverbal region (congruous vs. incongruous) and focus revision (broad-to-narrow vs. narrow-to-narrow) were fixed effects; subjects and items were random effects. In addition to those, at the replacive phrase, another analysis was conducted with three levels of congruency (congruous, incongruous object and incongruous subject) to examine if there is a difference between processing the two incongruous conditions; incongruous object and incongruous subject. Following Baayen (2008), the longitudinal effects of familiarization or fatigue were also examined. As the experiment progressed, the probability of regressing out of the spillover region in the immediately preverbal region became smaller (PRO: β = −0.01, SE = 0.00, z = −3.11, p < 0.05) and the reading times became shorter in the spillover region of the replacive phrase for RRD (β = −0.64, SE = 0.29, z = −2.21, p < 0.05). Thus, when building models for the given regions and measures, Trial was included as a predictor variable after it was centered to prevent collinearity (Baayen 2008). The analyses started with the simpler models with single predictors and then continued with complex models in which either the predictor variables were fit together (e.g., word-order + congruency) or they interacted (e.g., word-order x congruency). The simple and complex models were compared through a log-likelihood ratio test and the best fitting models are reported. During model criticism, data points with standardized residuals above or below 2.5 standard errors (Baayen and Milin 2010) as well as overly influential subjects, items and individual data points (influence.ME; Nieuwenhuis et al. 2012) were removed. The analyses for each critical region are reported separately below.

4.2.5. Results

The Immediately Preverbal Region

Table 2. Mean Values for the Six Conditions (in milliseconds (msec.) for durations), with Standard Errors in Parentheses, for the Six Eye-tracking Measures for the Immediately Preverbal Region. Here, and in all the following tables: CO = Canonical Word-Order, NCO = Non-canonical Word-Order, F = Focus, C = Congruous, InC = Incongruous, IO = Indirect Object, DO = Direct Object, and S = Subject.

	FFD	GD	RPD	RRD	TD	PRO
CO-C. IO	255 (7.09)	336 (14.5)	518 (38.7)	379 (38.5)	667 (38.0)	0.23(0.03)
CO-InC. DO	241 (6.69)	297 (13.4)	564 (42.8)	463 (46.1)	736 (47.6)	0.34 (0.03)
CO-InC. S	264 (9.7)	352 (15.1)	591 (53.2)	420 (43.3)	713 (44.4)	0.25 (0.03)
NCO-C. DO	268 (7.67)	268 (13.2)	571 (41.4)	457 (36.4)	740 (37.4)	0.27 (0.03)
NCO-InC. IO	262 (7.82)	327 (12.4)	609 (45.9)	475 (41.1)	756 (41.7)	0.33 (0.03)
NCO-InC. S	263 (9.19)	361 (16.5)	597 (39.4)	517 (47.6)	814 (47.7)	0.28 (0.03)

presents the mean values and the standard errors (in parentheses) for each eye tracking measure for the immediately preverbal region for each condition.

As noted above the immediately preverbal region was analyzed only for word order as a predictor. The effects of prosodic congruency and focus revision are expected to be observed at or after the replacive phrase region. The analyses indicated, for GD, RRD and TD, that reading the immediately preverbal region took longer when the sentence had non-canonical word order than when it had canonical word order GD: β = 27.65, SE = 10.74, t = 2.57, p < 0.05; RRD: β = 63.34, SE = 27.20, t = 2.32, p < 0.05, and TD: β = 73.87, SE =23.38, t = 3.16, p < 0.05; all other measures: t’s ≤ 1.76, p’s ≥ 0.07 and z = 0.64, p = 0.51.

Table 3. Mean Values for the Six Conditions (in msec. for durations), with Standard Errors in Parentheses, for the Six Eye-tracking Measures for spillover-1.

	FFD	GD	RPD	RRD	TD	PRO
CO-C. IO	250 (9.69)	292 (15.3)	446 (34.8)	192 (25.8)	376 (25.9)	0.24 (0.04)
CO-InC. DO	242 (9.25)	281 (10.9)	449 (52.1)	201 (25.3)	381 (24.7)	0.22 (0.03)
CO-InC. S	250 (9.36)	280 (12.5)	499 (11.8)	218 (25.1)	395 (25.0)	0.18 (0.03)
NCO-C. DO	228 (6.91)	249 (8.91)	454 (34.2)	188 (21.4)	374 (21.1)	0.28 (0.03)
NCO-InC. IO	239 (8.45)	268 (10.9)	487 (49.4)	229 (29.7)	388 (26.7)	0.26 (0.04)
NCO-InC. S	235 (7.65)	271 (12.7)	478 (47.9)	283(37.4)	443 (33.4)	0.26 (0.04)

presents the mean values and the standard errors (in parentheses) for each eye tracking measure for the spill-over region of the immediately preverbal region (henceforth spillover-1).

As in the immediately preverbal region, word order was included in the model as the only predictor variable. The analyses showed, for the RPD, that it took longer to read the spillover-1 region in non-canonical word-order conditions than that in canonical word-order conditions, β = 66.29, SE = 31.27, t = 2.12, p < 0.05. (For all other fixation durations: t’s ≤ 1.35, p’s ≥ 0.17 and for PRO: z = 1.76, p = 0.07.)

The results of both the immediately preverbal region and its spillover region indicated that reading the region took longer when the sentence had non-canonical word order than when it had canonical word order. These results would indicate that the readers projected narrow focus to the immediately preverbal region in non-canonical word-order sentences and a neutral reading (i.e., broad focus) in canonical word-order sentences. To ensure that the effect is merely a reflection of focus projection but not a result of unexpected (i.e., non-canonical) word order, the reading times at the region preceding the immediately preverbal region (where the readers would have had information on the structure of the sentence) were also compared across the two conditions. The results did not show any reliable difference between the canonical and non-canonical order sentences in the preceding region: t’s ≤ 0.85, p’s ≥ 0.39 and for the PRO: z = 0.30, p = 0.75). This excludes an interpretation that attributes the reading time differences to a factor other than projected focus, such as non-default word order.

An additional analysis was conducted to rule out the possibility of a recency effect for the observations at the replacive phrase region. That is, the proximity between the immediately preverbal constituent and the replacive phrase could potentially explain the congruity effects observed at the replacive phrase region. To rule out the possibility that the readers may have assigned focus to another preverbal constituent but not to the immediately preverbal constituent, the reading times for the immediately preverbal object and the preceding object were compared. This comparison would reveal which argument was assigned focus before encountering the replacive phrase. The results showed that the immediately preverbal region took significantly longer to read than the preceding object; FFD: β = 11.76, SE = 4.26, t = 2.75, p < 0.05; GD: β = 20.75, SE = 7.25, t = 2.86, p < 0.05; RPD: β = 122.67, SE = 14.10, t = 8.7, p < 0.05; and PRO: β = 1.55, SE = 1.79, z = 8.66, p < 0.05 (For all other fixation durations: t’s ≤ −1.74, p’s ≥ 0.08). These results rule out the possibility that any preverbal constituent other than the immediately preverbal object could have received focus assignment. (Note that the objects across these comparison conditions crisscrossed in canonical and non-canonical order sentences. Thus, we do not think that these observations would be due to word-length or differential case marking.)

The Replacive Phrase Region

Table 4. Mean Values for the Six Conditions (in msec. for durations), with Standard Errors in Parentheses, for the Six Eye-tracking Measures at the Replacive Phrase Region.

	FFD	GD	RPD	RRD	TD	PRO
CO-C. IO	259 (7.55)	429 (15.1)	588 (33.4)	283 (30.5)	674 (34.2)	0.18 (0.03)
CO-InC. DO	242 (7.23)	483 (20.8)	733 (47.2)	348 (36.2)	773 (41.2)	0.23 (0.03)
CO-InC. S	251 (7.17)	420 (13.8)	749 (60.7)	352 (32.5)	757 (36.7)	0.22 (0.03)
NCO-C. DO	262 (8.24)	452 (16.1)	744 (56.6)	284 (26.5)	697 (30.2)	0.24 (0.03)
NCO-InC. IO	262 (9.54)	479 (17.5)	812 (63.2)	315 (30.4)	750 (34.7)	0.22 (0.03)
NCO-InC. S	249 (7.72)	454 (17.2)	857 (66.2)	425 (36.1)	810 (41.5)	0.31 (0.03)

presents the mean values and the standard errors (in parentheses) for the replacive phrase region.

Two sets of analyses were conducted for the replacive phrase region. The first set of analyses examined the effects of (in)congruency and focus revision. Only two levels of congruency (congruous vs. incongruous) were included in these analyses. The second set of analyses examined whether or not the replacive phrase’s incongruency with the subject or the (in)direct object would differ and hence included three levels for congruency (congruous object, incongruous object and incongruous subject).

The first set of analysis included congruency and focus revision as predictors. The results showed an effect of both predictors. Reading the replacive phrase region in incongruous conditions such as (6b,c,e,f) took longer (for GD, RPD, RRD and TD) than that in incongruous conditions such as (6a,d); GD: β = 30.53, SE = 12,85, t = 2.37, p < 0.05, RPD: β = 147.23, SE = 38.2, t = 3.84, p < 0.05, RRD: β = 67.90, SE = 22.48, t = 3.02, p < 0.05 and TD: β = 60.13, SE = 22.84, t = 2.63, p < 0.05. There were no significant differences between congruous and incongruous conditions in FFD (β = −11.83, SE = 6.50, t = −1.81, p = 0.06) and in PRO (β =0.2174, SE = 0.1783, z = 1.219, p = 0.22).

The participants also fixated longer (GD, RPD) or were more likely to regress out of the replacive phrase region when revising focus in non-canonical sentences such as (6d,e,f) than in canonical sentences such as (6a,b,c); GD: β = 26.54, SE = 11.77, t = 2.25, p < 0.05 and RPD: β = 96.10, SE = 36.18, t = 2.65, p < 0.05, PRO: β = 0.37, SE = 0.1708, z = 2.176, p < 0.05. Focus revision did not show a reliable effect in the remaining measures: t’s ≤ 1.48, p’s ≥ 0.13. Given the analyses in the immediately preverbal region, it would be safe to assume that the participants had assigned narrow focus in non-canonical order sentences and broad focus in canonical order sentences. Thus, the longer times to revise focus in non-canonical sentences would indicate that revising focus from narrow to narrow is costlier than revising focus from broad to narrow.

The complex models with two predictor variables (congruency and focus revision) explained the data better than the simple models with focus revision as a single predictor (RPD: χ² (1) = 8.74, p < 0.05, RRD: χ² (1) = 8.56, p < 0.05, TD: χ² (1) = 10.12, p < 0.05) or congruency as a single predictor (RPD χ² (1) = 6.25, p < 0.05). The data were split by congruency for these eye-tracking measures to better understand the nature of the interactions. For RPD, the analysis of the reading times for the replacive phrase region for congruous conditions indicated that revising focus from narrow to narrow took longer than revising focus from broad to narrow, β = 93.45, SE = 48.45, t = 1.92, p = 0.054). No reliable effect of focus revision was found in RRD and TD (t’s ≤ 0.65, p’s ≥ 0.51). A similar effect was observed in incongruous conditions. For RPD, the analysis of the reading times for the replacive phrase region for incongruous conditions indicated that revising focus from narrow to narrow took longer than revising it from broad to narrow, β = 93.89, SE = 48.25, t = 1.94, p = 0.052. No reliable difference between narrow-to-narrow and broad-to-narrow focus revision was found in incongruous conditions for RRD and TD (t’s ≤ 1.53, p’s ≥ 0.12). As above, this would suggest that revising focus from narrow to narrow is costlier compared to revision of focus from broad to narrow in both congruous and incongruous conditions.

The second set of analyses included three levels of congruency (congruous, incongruous object and incongruous subject). This would allow to examine if the grammatical category of the verb’s argument would have a differential effect on the (in)congruency of the replacive phrase region. The effects of the focus revision predictor were similar to those in the first set of analyses; hence, they will not be reported here for brevity.

The analyses showed that both incongruous object (6b,e) and incongruous subject conditions (6c,f) were read more slowly than the congruous condition (6a,d); incongruous object: GD: β = 36.27, SE = 14.15, t = 2.56, p < 0.05, RPD: β = 136.53, SE = 44.17, t = 3.09, p < 0.05 and RRD: β = 55.13, SE = 27.73, t = 1.98, p < 0.05; all other measures: t’s ≤ 1.92, p’s ≥ 0.055 and PRO: z = 0.44, p = 0.65; and incongruous subject RPD: β = 157.97, SE = 44.21, t = 3.57, p < 0.05, RRD: β = 91.92, SE = 39.72, t = 2.31, p < 0.05 and TD: β = 68.53, SE = 26.49 t = 2.58, p < 0.05; all other measures: t’s ≤ 1.61, p’s ≥ 0.10 and PRO: z = 1.7, p = 0.08.

The analyses comparing the incongruous object conditions to the incongruous subject condition showed for RRD that it took longer to read the replacive phrase region in incongruous subject conditions (6c,f) than that in incongruous object conditions (6b,e), β = 62.51, SE = 27.23, t = 2.29, p < 0.05. The opposite pattern was observed for GD. For GD, it took longer to read the replacive phrase region in incongruous object conditions (6b,e) than that in incongruous subject conditions (6c,f); GD: β = 36.21, SE = 14.56, t = −2.48 p < 0.05. There was no reliable difference in other measures: t’s ≤ 1.26, p’s ≥ 0.20, z = 1.1, p = 0.26. Given that a later measure such as RRD is more likely to be associated with recovery and integration processes during reading (Roberts and Siyanova-Chanturia 2013) re-reading data can be interpreted as more informative and suggest that recovery from mis-assignment of focus in incongruous subject conditions was more difficult than that in incongruous object conditions.

Similar procedures were followed in the analyses for the replacive phrase’s spill-over (spillover-2) region.

Table 5. Mean Values for the Six Conditions (in msec. for durations), with Standard Errors in Parentheses, for the Six Eye-tracking Measures for the spillover-2 Region.

	FFD	GD	RPD	RRD	TD	PRO
CO-C. IO	242 (8.68)	299 (12.0)	631 (143)	129 (19.9)	318 (20.3)	0.112 (0.03)
CO-InC. DO	226 (6.66)	274 (10.8)	593 (84.6)	160 (23.5)	307 (21.2)	0.142 (0.03)
CO-InC. S	237 (8.41)	290 (13.0)	732 (92.5)	116 (18.7)	326 (21.8)	0.154 (0.03)
NCO-InC. DO	244 (8.20)	281 (10.2)	516 (60.7)	107 (17.9)	301 (19.1)	0.134 (0.03)
NCO-InC. IO	235 (8.51)	279 (12.4)	521 (67.2)	148 (21.6)	327 (23.9)	0.155 (0.03)
NCO-InC. S	239 (8.18)	311 (15.8)	725 (114)	180 (24.4)	339 (24.0)	0.175 (0.03)

presents the mean values and the standard errors (in parentheses).

The results were similar to the replacive phrase region. The first set of analyses showed that reading spillover-2 region took longer (for RPD and RRD) or the readers were more likely to regress out of the region in incongruous conditions (6b,c,e,f) than that in congruous conditions (6a,d); RPD: β = 152.30, SE = 65.50, t = 2.32, p < 0.05 and RRD β = 34.20, SE = 16.67, t = 2.05, p < 0.05; PRO, β =0.7297, SE = 0.3093, z = 2.359, p < 0.05 (all other measures: t’s ≤ 0.75, p’s ≥ 0.44). Focus revision did not have a reliable effect in this region, t’s ≤ 0.88, p’s ≥ 0.37; z = 0.02, p = 0.98.

The second set of analyses showed a higher processing cost for incongruous subject conditions (TD, β = 38.47, SE = 15.53 t = 2.47, p < 0.05 and PRO, β = 0.8029, SE = 0.3414, z = 2.35, p < 0.05 (all other measures: t’s ≤ 1.36, p’s ≥ 0.17) but not for incongruous object conditions (t’s ≤ 1.85, p’s ≥ 0.06; z = 1.85, p = 0.06). As the influence of incongruous object faded away at this region, no further analyses were conducted to compare incongruous object and incongruous subject conditions.

4.2.6. Discussion

The results of the eye-tracking experiment showed that processing the immediately preverbal region took longer in sentences with non-canonical word order than those with canonical word order. This was observed in the eye-tracking measures of GD, RRD and TD for the immediately preverbal region and the RPD for its spill-over region. In absence of any other focus markers (e.g., pitch accentuation) or manipulation of background context, a canonical word-order sentence would receive broad focus and the immediately preverbal constituent is expected to have a neutral reading. In a non-canonical sentence, the immediately preverbal constituent was predicted to carry narrow focus and have a focused reading. The results of the present experiment further confirmed this to be so. That is, reading the immediately pre-verbal region took longer in non-canonical sentences than canonical order sentences. Given that there was no background context to induce a particular type of focus or any other focus marker in the text, we take these observations to indicate that the readers assigned narrow focus to the immediately pre-verbal constituent in non-canonical word-order sentences and broad focus in canonical word-order sentences. The results are in line with previous research which reported for English (Birch and Rayner 1997; Lowder and Gordon 2015) that processing narrow focus (i.e., focused information) takes longer than processing broad focus (i.e., neutral information). We conducted further analyses to ensure that these observations cannot be attributed to word order differences.

The slower processing of focused information could be due to several factors. Lowder and Gordon (2015) argue that sentence structure, similar to background context, “act as a powerful linguistic cue to the readers” as to which constituents are more important to focus than others and processing focused constituents requires their deeper integration to the preceding context compared to neutral constituents, presumably because there is more information, such as the constituent’s prominence in delivering information (Birch and Rayner 1997), the increase in its specificity (Sturt et al. 2004) or the alternatives in its semantic set (Sanford et al. 2006), that needs to be encoded for that constituent in memory. This would increase the memory load for the focused constituent compared to a neutral constituent especially when the former has contrastive meaning (Reichle and Birdsong 2014).

In line with the Implicit Prosody Hypothesis (Fodor 1998, 2002), we also predicted the readers to mentally project the default prosodic contours in their processing of written input. The differences between the reading times for broad focus and narrow focus can therefore be also due to their differential phonological encoding. In oral production, it has been shown in Turkish that narrow focus has longer duration compared to broad focus (İvoşeviç and Bekar 2015). It has also been shown that listeners are sensitive to the acoustic cues associated with narrow focus (as opposed to broad focus) (İpek 2011). Therefore, the differences observed for the immediately preverbal constituent in canonical and non-canonical order sentences could be due to differential phonological encoding of broad and narrow focus during silent reading (see Pauker et al. 2011 for evidence of mental projections of prosodic boundaries and Kentner and Vasishth 2016 for that of focus).

As explained above, the eye-tracking data indicated that the immediately preverbal constituent in non-canonical word-order sentences carried narrow focus. Although we did not manipulate background context, the narrowly focused constituent might have had contrastive meaning as well as informational meaning depending on the readers’ interpretation of the main clause. If they had assumed, for example, that the sentence Sunucu yönetmene çiçeği verdi ‘The presenter gave the director the flower’, was an answer to a question such as ‘What did the presenter give the director?’ which makes çiçeği ‘the flower’ new information in the context, then there would be informational focus. But if the readers had interpreted the sentence as an answer to a question such as ‘Did the presenter give the director the award (or any other semantic alternative)?’, then çiçeği ‘the flower’ would carry contrastive focus. In both informational and contrastive focus interpretations of narrow focus conditions, the focusing degree is higher than the neutral information. As in Lowder and Gordon’s (2015) study, this can explain the longer reading times for the sentences with focused information. But additionally, if our readers’ interpretation for narrow focus was due to an associated contrastive meaning, they might have had to keep other elements in the same alternative sets of çiçeği ‘flower’ in mind such as chocolate or award, and pay attention to the truth value of çiçeği ‘flower’ among the other candidates. This may have brought a heavier processing load for memory in sentences with focused information (Reichle and Birdsong 2014).

The analyses of the measures for both the replacive phrase region (GD, RPD, RRD and TD) and its spill-over region (RPD, RRD and PRO) also indicated that the incongruency of the replacive phrase with the immediately preverbal constituent negatively affected the region’s processing. That is, when the constituent in the replacive phrase region was not congruous (i.e., did not contrast) with the immediately preverbal constituent, the participants slowed down. In addition, the measures of FFD, GD, RPD and PRO indicated that reading the immediately preverbal object caused longer reading times and higher probability of regressing out compared to the preceding object, which implies the availability of focus assignment to the former. These observations supported that the participants had a preference for the immediately preverbal position to be the default focus position in Turkish and are compatible with the sentence completion data as well as the theoretical literature (Erguvanlı 1984; İşsever 2003; Kılıçaslan 1994). The (in)congruency effects at the replacive phrase region could also indicate a prediction for the default focus position (Carlson et al. 2009; Clifton and Frazier 2016; Frazier and Clifton 1998) in Turkish. When the replacive phrase contrasted with a constituent other than the immediately preverbal constituent, it conflicted with the stress assignment the readers previously expected (i.e., stress assignment to the immediately preverbal constituent in both broad focus and narrow focus conditions due to the sentential stress assignment or scrambling as a focus marking strategy). This in turn resulted in longer fixation durations or increased the probability of regressing out of the region as it required revision to the stressed constituent in the main clause.

The grammatical category of the incongruous replacive phrase also affected the reading times. The participants read the replacive phrase region more slowly when it was incongruous with the subject than when it was incongruous with the (in)direct object (re-reading duration at the replacive phrase region; total duration and probability of regression out at its spill-over region). This concurs with earlier claims that the subject, i.e., the sentence initial, position in Turkish is typically reserved for background information and is not expected to carry new information (Erguvanlı 1984). That is probably why revising the focus of the sentence to a structure in which focus is assigned to the subject took longer. One might also argue that this result may be due to the subject’s distance to the default focus region, i.e., the immediately preverbal position. Since the subject is sentence-initial and more distant to the immediately preverbal position than the object, it could have been more difficult for the readers to retrieve cues associated with the subject from memory than retrieving cues associated with the object (Grodner and Gibson 2005; VanDyke and Lewis 2003). But given the results of the sentence completion task in which the subject was the least likely to receive focus, the distance effect may be an additional factor.

The analysis at the replacive phrase (GD, RPD and PRO) also indicated that revising focus structure of the main clause from narrow to narrow took longer than revising it from broad to narrow. This was the case in both congruous and incongruous conditions. We attribute this again to the focus structural differences between two different word-order sentences. In broad-to-narrow focus revision, all the constituents in the main clause were assumed to carry neutral information until the replacive phrase where the readers would realize that they should have assigned focus to a specific constituent. Thus, they had to revise the main clause focus structure from broad to narrow and assign narrow focus to a constituent in the main clause for the first time. In narrow-to-narrow focus revision, the readers would have already assigned focus to the immediately preverbal constituent in the main clause before the replacive phrase. There, the main clause focus structure needed to remain narrow, but the focused information that the immediately preverbal constituent induced needed to be amended. These effects could be due to different reasons in congruous and incongruous conditions.

In congruous conditions, the readers needed to compare the truth value of the semantic alternative in the replacive phrase to the one in the immediately preverbal constituent. In narrow-to-narrow focus revision, the narrowly focused constituent in the main clause would have been deeply integrated to its context as it was previously focused. Thus, its comparison to an alternative in its set would require encoding other information deeply and comparing the two deeply encoded constituents. In broad-to-narrow focus revision, the immediately preverbal constituent would have neutral information. Thus, comparing it to an alternative in its set would only require to compare it to an alternative in its set and deeply encode the constituent to memory for the first time. For the incongruous conditions, broad-to-narrow focus revision would require assigning narrow focus to the other object or the subject for the first time, hence deeply encoding that constituent in memory and comparing it against an alternative in its set at the replacive phrase region. Narrow-to-narrow focus revisions, on the other hand, would require the readers to delete the previously and deeply integrated information in the main clause and replace it with new information which also needed to be deeply integrated into the context alongside its new semantic alternative set which would also need to be activated and suppressed. (See Pauker et al. 2011 for a similar argument on prosodic boundary deletion vs. prosodic boundary assignment.)

These findings are similar to Stolterfoht et al. (2007) for incongruous conditions (i.e., a negativity in ERP measures when their readers needed to change the focus stress from the object to the subject) and to Benatar and Clifton (2014) (i.e., longer reading times when a change to discourse was required) for both congruous and incongruous conditions in the sense that we showed slow down for narrow-to-narrow focus revisions.

5. Summary and Conclusions

This study investigated whether or not the immediately preverbal position was associated with focus in Turkish and whether or not there were any processing or revision differences between canonical and non-canonical word-order sentences which were predicted to carry broad focus (or neutral information) and narrow focus (or focused information), respectively. These questions were addressed with a sentence completion task and an eye-tracking experiment.

The results of the sentence completion task showed, in line with the theoretical literature (Erguvanlı 1984; İşsever 2003; Kılıçaslan 1994), that both in canonical and non-canonical word order sentences the immediately preverbal position was preferred as the default focus position in Turkish. The results of the eye-tracking experiment indicated that reading the immediately preverbal region took longer when the sentence had non-canonical word order (hence narrow focus on the immediately preverbal region) than when it had canonical word order (hence broad focus on the whole sentence). This indicated that when the immediately preverbal constituent had focus, its processing took longer than when it was neutral in focus. This finding is in line with previous research (Lowder and Gordon 2015; Sturt et al. 2004; Sanford et al. 2006; and Reichle and Birdsong 2014) in that an increase in a constituent’s focus level (narrow > broad) also increased its processing time. This could be because the focused constituents require deeper encoding to the previous context (Lowder and Gordon 2015) due presumably to more semantic detail they introduce (Sturt et al. 2004), suppression of their semantic alternatives (Sanford et al. 2006) and heavier burden these additional operations create on memory (Reichle and Birdsong 2014). Additionally, the structural differences that occur during broad focus and narrow focus assignment (İşsever 2003), or differences in their phonological encoding during silent reading can also explain the longer reading times for narrow focus.

The eye-tracking data at the replacive phrase region provided further evidence (to the sentence completion task) that the immediately preverbal region is considered as the default focus position in Turkish for both canonical (Erguvanlı 1984; Göksel and Özsoy 2000, 2003; İşsever 2003, 2006) and non-canonical word-order sentences (Erguvanlı 1984; İşsever 2003; Kılıçaslan 1994) as the readers had greater difficulty when the region contrasted with a constituent that was not immediately preverbal than when it contrasted with a constituent that was immediately preverbal. In addition to providing empirical evidence for theoretical predictions for focus assignment in Turkish, this observation could also suggest that Turkish readers create expectations for the default focus position in Turkish (Carlson et al. 2009; Clifton and Frazier 2016; Frazier and Clifton 1998; Stolterfoht et al. 2007). When those expectations are not met (at the replacive phrase region), their processing slows down.

The results of the replacive phrase region also showed that the sentence-initial position was the least preferable position for focus assignment in Turkish because revising focus to sentence-initial position (the subject in the present materials) caused the longest reading times. This is compatible with the argument that the subject position is mostly reserved to present background information in Turkish (Erguvanlı 1984). The longer distance of the subject to the replacive phrase region than the objects could have also made it difficult for the readers to retrieve the cues associated with it due to activation-based decay (Grodner and Gibson 2005; VanDyke and Lewis 2003).

The readers’ behavior at the replacive phrase region also informed on revision of main clause focus structure. Revising focus from narrow to narrow was costlier compared to revising it from broad to narrow. This is similar to Benatar and Clifton’s (2014) findings and indicates that assigning focus for the first time is easier than deleting focused information and re-assigning it since deleting and re-assigning focus would require deletion of already encoded information and introduction of new information. And in some cases, (i.e., the incongruous conditions), revising focus from narrow to narrow would require encoding a new semantic alternative set to the context while deleting the previously encoded constituent and its semantic set.