CHAPTER 25
FOCUS
A news article in The Guardian from July 2016 attributes the statement in (1) about new Prime Minister Theresa May to a ‘liberal Tory’ (Behr, 2016).
(1) ‘Theresa will at least stab you in the front.’
The article goes on to clarify that the statement was made as a compliment. Indeed, any speaker of English could only interpret (1) by taking its focus structure into account: the phrase in the front is contrastively focused with an implicit alternative, in the back, yielding an overall interpretation that could be paraphrased as ‘Given circumstances where stabbing is inevitable, Theresa will take the alternative of stabbing you in the front, rather than stabbing you in the back.’ Interpreting the statement as a compliment therefore crucially relies on a comprehender’s ability to recognize the focus structure of the sentence, and infer the (in this case very small) contrast set with respect to which the focused element must be interpreted.
Contemporary research on focus within linguistic theory falls primarily into two strands: the study of the semantics of focus—its contribution to truth conditional and non-truth conditional meaning (Chafe, 1976: Rooth, 1985, 1992; Krifka, 1992; Schwarzschild, 1999), and the study of the intonational realization of focus in speech (Ladd, 1980; Gussenhoven, 1983; Selkirk, 1984; Pierrehumbert & Hirschberg, 1990; Selkirk, 1995). Experimental investigations into focus have been going on alongside theoretical work throughout. Along the way, they have occasionally become entangled in important debates within psycholinguistics, such as the debate over staged versus constraint-based parsing models (Ni et al., 1996: Sedivy, 2002; Paterson et al., 2007). Experimental questions have also often come from theoretical frameworks: for instance, experiments on focus projection and on the nature of focus alternatives were inspired by theoretical claims.
This chapter surveys the main lines of experimental research into focus. It is organized thematically rather than chronologically or by methodology, though sometimes certain questions attract certain experimental paradigms. It is therefore important to keep differences in methodology in mind, particularly when studies asking similar questions use different tasks, and report different outcomes. Also important are differences in how focus is ‘operationalized’ in experimental studies. While manipulating the preceding context, using focusing syntactic constructions such as clefts, using focus particles like only, and manipulating the type and placement of pitch accents are all meant to manipulate focus, these different ways of manipulating focus come with their own linguistic properties, and quite often do not yield identical results.
Experimental studies also differ importantly in how they characterize the abstract notion of focus that these manipulations are meant to probe. Some studies define focus in terms of discourse status (new versus given information, focus versus background). Others strictly tie focus to its surface realization, whether that is the presence of a pitch accent, the use of a focusing syntactic construction, or correspondence to explicit wh-elements in the discourse. Taking such an approach, a problem then becomes how to discover the focus structure of a sentence by way of overt cues to focus and principles about how focus can project from overtly-marked elements to unmarked ones. For many studies, ‘focus’ is used to mean contrastive focus—information that is contrastive with another (explicit or implicit) discourse element, as opposed to informational focus—focus used to mark new information. In such cases, contrast is often defined in terms of intended meaning, and inferring the intended contrast set is a matter affecting a sentence’s truth-conditions. As we will see, even how the notion ‘contrastive’ is implemented can vary from study to study.
Finally, ‘focus’ is sometimes used in experimental studies to mean ‘discourse focus’, which is related to attentional focus rather than the notion of focus defined in terms of linguistic representations. While we might want to set research into discourse focus aside because it is defined differently from linguistically-characterized focus, it is not always clear-cut which cases should be set aside: notions like discourse focus have been invoked in explanations of some of the same linguistic phenomena that linguistic focus is used to explain. For example, one influential theory of discourse structure (Grosz & Sidner, 1986) proposes that two different levels of focus structure are responsible for different interpretive phenomena: Centering Theory works at the sentential/inter-sentential level and is responsible for resolving anaphoric reference, while discourse-level intentional structure is responsible for the interpretation of definite descriptions. Such an explanation would seem to have much in common with e.g. the proposal in Cummins & Rohde (2015), which does appeal to contrastive focus in connection with pronoun resolution.
The organization of this chapter cross-cuts the earlier issues, and instead focus on lines of research that are held together by the questions they ask about memory representations, time-course, interpretation, and cues to focus structure. The final section concludes with discussion of some issues that are likely to be important in future experimental research on focus.
25.1 ATTENTION, MEMORY, AND DEPTH OF PROCESSING
Experimental studies going back to the 1970s have used recall paradigms to show that focused material is remembered better than unfocused material, whether focus is manipulated using pitch accent (Cutler, 1976; Cutler & Foss, 1977; Hornby, 1974), using the discourse context (Cutler & Fodor, 1979), or using syntactic focusing constructions such as clefts (Bredart & Modolo, 1988; Morris & Folk, 1998; Birch et al., 2000). Enhanced memory representations in such studies were construed as evidence that more attention was dedicated to processing focused material. Increased attention at the point of memory encoding was also consistent with focused material being easier to integrate into an existing discourse model (Morris & Folk, 1998), and the increased likelihood of referring to focused elements in story continuations (Birch et al., 2000) due to increased availability of these items in memory.
A line of research attempts to use effects of focusing on attention to explain enhanced sensitivity in change-detection tasks. Bredart & Modolo (1988) found that the ‘Moses illusion’ described by Erickson & Mattson (1981)—in which many participants provide an answer for the question ‘How many animals of each kind did Moses take on the Ark?’ and fail to notice that ‘Moses’ should have been ‘Noah’—was sensitive to clefting: prticipants in a sentence verification task were more susceptible to the illusion when material other than the problematic name was the focus of a cleft (e.g. ‘It was two animals of each kind that Moses took on the Ark.’).
Sturt et al. (2004) (also Sanford & Sturt, 2002) suggest such change-blindness phenomena have to do with the influence of focus on depth of processing: focused material is more likely to be processed fully, rather than being left underspecified. In a text-change detection task, participants had to detect changes (e.g. ‘cider’ being replaced by ‘beer’; changes underlined in (2)–(5)) between two repetitions of a sentence. Changes were more likely to be noticed when the changed word was focused by clefting or pseudoclefting (2) than when a different word was focused (3).
(2) What Jamie really liked was the CIDER/BEER.
(3) It was JAMIE who really liked the cider/beer.
Similar results obtained when focus was manipulated using the prior context instead of syntactically (Sturt et al., 2004) (4), and using both prosody and prior context (Sanford et al., 2006) (5).
(4)a. Everybody was wondering which man got into trouble./Everybody was wondering what was going on that night.
b. In fact, the man with the hat/cap was arrested.
(5)a. They wanted to know which money had been stolen. /They wanted to find out what had happened.
b. The money from the WALLET/PURSE had gone missing.
c. Thefts in the area were becoming all too common.
Eye-tracking during the same change-detection task showed more fixations and longer viewing times on a changed than on an unchanged word, only when the changed word was in focus (Ward & Sturt, 2007).
Explanations in terms of depth of processing are in principle compatible with either focus leading to more detailed processing of lexical material, or focused material requiring more processing effort. According to the first type of explanation, unfocused material has a less detailed (perhaps underspecified) memory representation to begin with, leading to differences in retrievability between focused and unfocused material. By contrast, according to an effort-based explanation, the same information may be encoded better for focused compared to unfocused material. In addition, according to Birch et al. (2000; see also Birch & Garnsey, 1995), the effects of focusing on memory only appear after a delay: in probe recognition experiments, when an element in the subject position of a sentence was probed immediately, syntactic clefting had no effect on recognition latency. However, when the probe was delayed, syntactically focused words were recognized faster than those that had not been focused. Similarly, Foraker & McElree (2007) argued on the basis of Speed-Accuracy Tradeoff experiments that clefting did not affect the speed of accessing an antecedent representation, but did increase the likelihood of retrieving the antecedent representation. These results were taken to suggest that focusing using clefts strengthens the focused material’s memory representation, but does not result in focused elements being actively maintained in focal attention (see Onea, Chapter 24 in this volume, for more on clefts).
Turning to effects on the processing of the focused element itself, there have been mixed findings regarding the impact of focus on reading time. Birch & Rayner (2010) used clefting to manipulate focus/prominence as in (6)–(7).
(6) The tenants at the complex were sick and tired of all the noise coming from #204.
(7)a. It was the landlady who confronted the woman who lived there.
b. The landlady confronted the woman who lived there.
They reported fewer fixations and shorter reading times for focused words, in both early measures of reading (first-fixation time, gaze duration) and late measures (total reading time, total number of fixations), suggesting that both encoding and integration processes were facilitated for syntactically focused words.
In another reading study using clefts and pseudoclefts to manipulate focus, however, Lowder & Gordon (2015) found increased reading times on a word (memo) when it was syntactically focused (8a) compared to conditions without clefting or pseudoclefting (8b) or conditions where a different word was focused (8c), as well as more time regressing to words immediately preceding the focused word.
(8)a. What the secretary typed was the official memo about the new office policy.
b. Yesterday the secretary typed the official memo about the new office policy.
c. It was the secretary that typed the official memo about the new office policy.
The authors suggested that these findings reflected deeper encoding of focused material, and deeper integration of focused material with its sentential context. The difference in findings between Birch & Rayner (2010) and Lowder & Gordon (2015) might be related to the difficulty of integrating new information into the discourse representation: while the focused element was discourse-new in both studies, only Birch & Rayner (2010) provided a preceding context sentence (6). Processing the focused element might have been facilitated in that study because the focus (the landlady) was implicitly associated with the individuals mentioned in the context sentence.
In fact, a study by Chen et al. (2012) independently manipulated newness and focus, and suggested that being new in the discourse makes information more difficult to integrate, while focused information is processed more quickly. In a reading study in Chinese, focus was marked by the focus-particle shi (10). Newness was manipulated independently using the discourse context (9).
(9)a. Heren was persuading his friends to go on an outing. (He) ignored that the weather forecast had predicted bad weather.
b. Heren was persuading Zhongying and others to go on an outing. (He) ignored that the weather forecast had predicted a bad weather.
(10)a. At that time shi Zhongying reasonably opposed him.
‘At that time it was Zhongying who opposed him reasonably.’
b. At that time Zhongying reasonably opposed him.
‘At that time Zhongying opposed him reasonably.’
In the critical region (Zhongying), focused information took less time to read than non- focused information, while new information took longer to read than given information. In the immediately post-critical region, fixation durations were longer for new information than given information, only in the focus conditions. Thus, it may be important to take into account the extent to which focused elements are implicitly or explicitly represented in the discourse context when asking about the effects of focusing on processing time or complexity.
When integrating new material requires revising the discourse model, not simply adding to it, focused material is processed more slowly than unfocused material. In Benatar & Clifton (2014), a context question (11b) was followed by an answer in which a critical word was either new or contrastively (correctively) focused (11c).
(11)a. John and Mary are working today.
b. Speaker A: Did you tell someone to go home early?/Did you tell Mary to go home early?
c. Speaker B: I told JOHN, but I don’t know if it was a good idea.
Target words were read more slowly when the critical word corrected prior information than when it provided new information. By contrast, target words introducing new information were read more slowly than those providing given information, in sentence pairs like (12).
(12)a. Speaker A: I’m confused, does Kyle care about Natalie?/Natalie is confused, does Kyle care about someone?
b. Speaker B: Kyle cares about Natalie but he doesn’t show it.
The presence of explicit contrast sets in the discourse seems to facilitate integration of focused elements. Braun & Tagliapietra (2010) and Fraundorf et al. (2010) showed that focus alternatives explicitly introduced in the prior discourse are better remembered when the focused element receives a contrastive (L+H*) pitch accent. In a direct comparison of explicitly introduced and inferred contrast, Chen & Yang (2015) had participants read short discourses that introduced one or two characters, with the character being emphasized or non-emphasized in subsequent texts. Eye movements showed that early processing of the emphasized character was facilitated (reflected in first fixation duration and gaze duration), which may have been due to increased attention allocation, whereas late integration of the emphasized character was inhibited when the discourse did not explicitly introduce a contrastive character (reflected in total reading times, second pass reading times, total number of fixations, and regressions). The integration of emphasized names was therefore facilitated only when the context provided a contrastive character.
In addition to strengthening memory representations for focused elements themselves, focus has been shown to enhance memory for alternatives to the focused element—members of its contrast set. Fraundorf et al. (2010) showed that, in a sentence verification task following discourses which introduced an explicit contrast set followed by a focused element (13), a contrastive L+H* accent selectively facilitated correct rejections of explicitly-mentioned contrast set members (14b), and did not facilitate correct rejections of same-category associates that were not mentioned as part of the contrast set (14c).
(13)a. Both the British and the French biologists had been searching Malaysia and Indonesia for the endangered monkeys.
b. Finally, the BRITISH spotted one of the monkeys in MALAYSIA and planted a radio tag on it.
(14)a. The British scientists spotted the endangered monkey and tagged it.
b. The French scientists spotted the endangered monkey and tagged it.
c. The Portuguese scientists spotted the endangered monkey and tagged it.
Similar findings were reported by Spalek et al. (2014), who manipulated focus using the German exclusive particle nur (‘only’) and the inclusive particle sogar (‘even’). Participants heard discourses that explicitly introduced a contrast set, then continued by mentioning an element from that set with nur, sogar, or no particle. When recalling elements from the initial context sentence after a delay, participants recalled alternatives to the focused element better in both focus particle conditions, relative to the no particle condition. This was despite lexical differences between the two particles: while inclusive particles mark the alternatives to be part of the predication, exclusives express that the alternatives to the focused constituent do not hold. These findings suggest that focus and contrast are processed differently from unfocused material, but in a way that may not be sensitive to lexical differences.
25.2 FOCUS AND SYNTACTIC AMBIGUITY RESOLUTION
Focus, and particularly the presence of explicit contrast sets in the discourse, played an important role in the debates over structural versus constraint-based models of sentence processing (Frazier, 1987; MacDonald et al., 1994; Trueswell & Tanenhaus, 1994). Much of this literature centred around temporary syntactic ambiguities like (15), where loaned is temporarily compatible with being the main verb of the sentence (‘The businessmen loaned people money at low interest’) or starting a reduced relative clause (‘The businessmen who were loaned money at low interest…’).
(15) The businessmen loaned money at low interest were told to record their expenses.
(16) Only businessmen loaned money at low interest were told to record their expenses.
According to structural theories, any sentence should receive an initial parse based solely on structurally-defined principles. However, Ni et al. (1996) showed that replacing the definite determiner the with the focus particle only (as in (16)) reduced the garden path effect in the disambiguating region. What only does here is introduce an implicit contrast set into the discourse: the businessmen mentioned in the sentence are contrasted with some unmentioned set of individuals. A continuation where loaned introduces a restrictive relative clause would supply exactly the kind of contrast set needed (businessmen who were loaned money at low interest, businessmen who were not loaned money at low interest). By creating an expectation for contrastive information and thereby making the relative clause parse more likely, only makes it easier to revise an incorrect main verb parse to the correct relative clause parse later in the sentence.1 The increased reading times normally associated with making this revision is therefore reduced in sentences with only.
Sedivy (2002) replicated Ni et al.’s only effect, and built on this by manipulating whether an explicit contrast set was introduced in the prior discourse. The inclusion of an explicit contrast in the discourse context (17) eliminated the facilitatory effect of focus on reading of reduced relatives.
(17) All of the secretaries and accountants were made to take a tough computing course.
(18) Only the secretaries prepared for the exam …
a. … and earned significant pay raises. (main verb continuation)
b. … passed and earned pay raises. (relative clause continuation)
Paterson et al. (1999; see also Filik et al., 2005) argue that the reduced garden path observed for reduced relative clause sentences with only does not hold for all reduced relative clauses. In particular, they argue that relative clause sentences starting with a noun phrase, verb, noun phrase sequence (19) were resistant to only-induced contrast effects because of the overwhelming preference for a Subject-Verb-Object parse (cf. the SVO Heuristic from Bever, 1970).
(19) The/Only teenagers (who were) allowed a party invited a juggler straightaway.
In an eye-tracking reading study, participants read reduced and unreduced sentences with and without the focus operator only. There were longer first-pass reading times in the critical region of reduced sentences than in the same region of unreduced sentences, regardless of the inclusion of only. The authors conclude from this result that only is sometimes unable to modulate a garden-path effect when one parse is very strongly preferred. However, it should be noted that Ni et al. (1996) and Sedivy (2002) did not claim that only should affect parsing uniformly; rather, they were arguing against processing models that excluded the possibility of discourse-level contrast influencing the initial parse of a sentence.
Prosodically-marked focus has also been shown to influence syntactic attachment (Speer et al., 1993; Schafer, 1996; Schafer et al., 2000; see also Tonhauser, Chapter 29 in this volume, for in-depth discussion of prosodically realized focus). Schafer (1996) showed that prosodically-signalled focus can influence syntactic attachment preferences. A H* accent on a noun increased the likelihood that a subsequent relative clause was interpreted as modifying that noun; when a sentence contained two nouns that could serve as the head of a relative clause, an accent on one of the nouns disambiguated the sentence in favour of the accented noun. Thus, the relative clause was more frequently interpreted as modifying the noun plane when it was accented (20) than when the noun propeller was accented (21), and vice versa.
(20) The sun sparkled on the propeller near the PLANE that the mechanic was so carefully examining.
(21) The sun sparkled on the PROPELLER near the plane that the mechanic was so carefully examining.
A similar attachment effect is observed with embedded wh-words that are ambiguous between an interrogative wh-element introducing an embedded question and a relativizer (Schafer et al., 2000): a H* pitch accent on the wh-word (22) led to more embedded question interpretations than when a different constituent was accented (23).
(22) I asked the pretty little girl WHO is cold.
(23) I asked the pretty little girl who is COLD.
Because interrogative wh-elements must be focused, accent on an embedded wh-word supports an analysis where the wh-word is functioning as an interrogative constituent and introducing an embedded question. On the other hand, wh-relativizers are not normally focused, therefore accenting would be incompatible with an analysis where the wh-word introduces a relative clause.
25.3 FOCUS AND INTERPRETATION
25.3.1 Focus as cue to discourse structure
One way in which focusing has been shown to influence interpretation is by serving as a cue to some aspect of the discourse structure. In an environment where an elided verb phrase had two potential antecedents, a contrastive pitch accent (L+H*) on a subject preceding one of the potential antecedents (‘said Maria went to the rally’ in (24a), ‘went to the rally’ in (24b)) made that verb phrase more likely to be interpreted as the antecedent of the elided verb phrase (25a)–(25b) (Frazier et al., 2007).
(24)a. JULIE said Maria went to the rally…
b. Julie said MARIA went to the rally…
(25)… and GREG did too.
a. (Greg said Maria went to the rally)
b. (Greg went to the rally)
As Kehler (2000) has argued (see also Kehler, 2002; Kertz, 2010), antecedent and ellipsis clause pairs like those in (24)–(25) exemplify the Resemblance coherence relation, supported by the use of the connective and, and the interpretation of the elided VP as syntactically parallel to the antecedent in terms of argument realization. What the contrastive pitch accents in (24) might be doing, then, is serving as a cue to the underlying discourse structure—whether that structure is ultimately construed in terms of coherence relations (Kehler, 2000) or information structure (Kertz, 2010).
The Frazier et al. (2007) findings might also be explained by appealing to Question Under Discussion (QUD; Roberts, 1996): the difference in interpretation between (25a) and (25b) could be due to a difference between the QUD ‘What did each person say?’ and ‘What did Julie say?’. Clifton & Frazier (2012) directly tested the effects of focus-evoked QUD on discourse interpretation. In an eye-tracking during reading study, reading was disrupted when an answer to an explicit question did not answer the question in a way that conformed to the subcategorization biases of the verb used in the question. Further, when a previously introduced question ended up not being addressed by the end of a discourse, reading was disrupted, suggesting that readers maintained an expectation for QUD resolution while reading a discourse. Relatedly, Cummins & Rohde (2015) demonstrated that focus placement could indirectly influence interpretation in a number of pragmatic phenomena by influencing the QUD inferred by listeners. They used a listening task where contrastive pitch accent was placed on a target element: a scalar adjective (26), an element embedded under a negated presupposition trigger (27), or a potential antecedent of a pronoun (28).
(26) The view from the hotel window is pretty/PRETTY.
(27) Bill doesn’t regret arguing with his boss/BOSS.
(28) Charles/CHARLES congratulated Simon. He had criticized Stephanie.
In each case, focus highlighted the QUD, perhaps making listeners infer a contrastive question rather than a non-contrastive informational one. For scalar adjectives, this increased the availability of scalar alternatives, increasing the availability of scalar implicatures (‘The view from the hotel window is not gorgeous’). Similarly, contrastively focusing an element embedded under a presupposition trigger in (27) may have highlighted the QUD ‘Who is it that Bill doesn’t regret arguing with?’ (as opposed to e.g. ‘Does Bill regret arguing with his boss?’), making the presupposition that Bill argued with his boss less likely to project. Relative to a baseline preference for interpreting ‘Simon’ as the antecedent of ‘he’ in (28), contrastive focus on ‘Charles’ increased interpretations of ‘Charles’ as the antecedent. In all cases, focus placement could be argued to have influenced interpretation indirectly, by serving as a cue to QUD (see Degen & Tanenhaus, Chapter 3 in this volume, for discussion of the role of QUD in pragmatic processing, including the interpretation of scalar implicatures).
25.3.2 Anaphoric dependencies
Much of the discussion of focus in relation to pronoun resolution has centred on the time-course of implicit causality effects (see Rohde, Chapter 27 in this volume, for discussion of implicit causality effects on pronoun interpretation). The implicit causality of a verb refers to the bias to attribute causality to one of its arguments. When asked to provide a cause for a sentence like ‘Ann envied Bill’, for example, people tend to produce causes associated with Bill (e.g. ‘… because he had a beach house’). NP2-biasing verbs like envy contrast with NP1-biasing verbs like anger, which tend to elicit causes associated with the subject NP (‘Ann angered Bill because she knocked over his drink’). The influence of verb bias on pronoun resolution was demonstrated by studies in the 1970s (Garvey and Caramazza, 1974; Garvey et al., 1974; Brown and Fish, 1983; Au, 1986); for example, Caramazza et al. (1977) showed that participants’ interpretation of the pronoun in a sentence like (29) (from Stewart et al. (2000)) was consistent with the verb’s implicit causality, and that responses were faster when the subordinate clause was congruent with verb bias.
(29) Roy questioned Anthony because he wanted to learn the truth.
The debate over the mechanism underlying implicit causality effects has often boiled down to a difference in predictions about time-course. One prominent theory about the source of these effects appeals to the focusing properties of implicit causality: by focusing attention on the cause of an action, implicit causality facilitates comprehension of a subsequent pronoun referring to that individual. Because proponents of the Focusing account link implicit causality effects directly to verb bias, they predict that these effects should emerge in early stages of processing, very soon after the pronoun is encountered. By contrast, Integration accounts associate implicit causality effects with later-stage processing, when the two clauses are integrated based on their causal relationship; on such accounts, implicit causality effects on pronoun resolution are expected to emerge late.
McDonald & MacWhinney (1995) argued for the Focusing account using a cross-modal probe task involving auditory presentation of a sentence like (29) and visual presentation of a probe word. For NP2-biasing verbs, NP1 was processed faster than NP2 at points before the pronoun and 200 milliseconds (ms) after the pronoun (perhaps reflecting a first-mention advantage), but immediately after the pronoun, reaction times were similar for both NP1 and NP2. Greene & McKoon (1995) also reported a greater reduction in reaction times for NP2 probe words than for NP1 probe words, when probe words were presented after an NP2-biasing verb rather than before the verb. Because these effects occurred either immediately after the pronoun or before the pronoun but after the biasing verb, these results were used to argue for the Focusing account.
Garnham et al. (1996) employed an all-visual probe task, and tested before and after the pronoun and at the end of the sentence. Self-paced reading was used to assess effects of congruency—whether the implicit cause associated with the verb matched the explicit cause given in the sentence—and anaphor type (name or pronoun).
(30) Jean congratulated Rita (vigorously) because she/Rita had won the championship.
(congruent)
(31) Jean congratulated Rita (vigorously) because she/Jean was very impressed.
(incongruent)
The absence of a congruency-by-anaphor type interaction, with a congruency effect appearing for pronouns but not for names, was construed as evidence against the Focusing account, and for the Integration account.
Because the debate is about whether attentional focus plays a role in the mechanism underlying implicit causality effects, studies like these do not typically involve manipulating linguistic focus. One exception is Itzhak & Baum (2015): against the background of implicit causality effects, they asked how contrastive focus (conveyed by a contrastive pitch accent) interacted with verb bias to influence reference resolution in a visual world experiment (Tanenhaus et al., 1995). In critical conditions, implicit causality bias was pitted against contrastive accent: in (33), the verb envy biases the pronoun to be interpreted as referring to Bill, but the antecedent John is contrastively accented.
(32) John and Bill both care about money a lot.
(33) JOHN/John envied Bill when they were young because he came from a rich family.
While the bias-inconsistent referent (John) received fewer fixations overall, looks to the bias-inconsistent referent increased when the corresponding name was contrastively accented. This effect emerges at the pronoun, and prior to disambiguating information; contrastive focus therefore seems to be able to resolve the potentially misleading information from verb bias.
25.3.3 Focus and reference resolution in visual displays
The Itzhak & Baum (2015) study builds on a line of research investigating prosodically-marked contrast on reference resolution in visual displays. Dahan et al. (2002) found in a visual world task that participants rapidly interpret accented noun phrases as referring to discourse-new referents and deaccented noun phrases as referring to given referents. In a first instruction, participants were asked to move an object in a display (34); a second instruction used either an accented (L+H* or H*) or a deaccented noun to refer to the same object (candles), or to a discourse-new competitor (candy) (35a)–(35b).
(34) Put the candle above the triangle.
(35)a. Now put the CANDLE/CANDY above the square.
b. Now put the candle/candy ABOVE THE SQUARE.
Participants looked more often to the competitor (candy) when the noun in the second instruction was accented and less often when it was deaccented, indicating that listeners made immediate use of pitch accent information and its relationship to discourse status in order to resolve reference.
In another visual world eye-tracking study, Weber et al. (2006) found that German listeners looked earlier at the picture of a referent belonging to a contrast pair (red scissors, as opposed to red vase) when instructions to click on it carried a contrastive accent on the colour adjective (37a) than when the adjective was not accented (37b).
(36) Click on the purple scissors.
(37)a. Now click on the RED scissors/vase.
b. Now click on the red SCISSORS/VASE.
In addition, there was a general preference to interpret adjectives contrastively, whether the adjective was accented or not; this was reflected in a general preference for the contrast set member (red scissors) relative to its competitor (red vase). In a second experiment, however, introducing the first member of the contrast pair (purple scissors) with a contrastive accent resulted in the subsequent adjective being interpreted contrastively only when contrastively accented. This suggests that contrastive focus is sensitive to the prior prosodic context, in addition to the discourse status of potential referents.
Kurumada et al. (2014) provided further evidence for the incremental computation of contrastive inferences on the basis of visual context and contrastive prosody. Participants in a visual world experiment heard ‘It looks like an X’ pronounced with either a noncontrastive accent on the final noun (38) or a contrastive accent on the verb (39).
When the visual display contained a single contrast set (e.g. a zebra and a zebra-like animal), listeners made anticipatory eye movements to the contrast set upon hearing contrastively-accented LOOKS. Presumably, this is because the contrastive focus on looks made the contrast between ‘It LOOKS like X’ and ‘It IS X’ highly salient. As the effects of contrastive focus emerged before auditory input about the target noun (zebra) was available, this result shows that contrastive prosody, together with information from the visual context, gives rise to immediate inferences about reference.
Prosodic information is also used to signal which element under the scope of a focus particle should be focused. In a visual world study, Gennari et al. (2004) manipulated whether the element that associated with only was unstressed or stressed (40).
(40) The mother only brought some milk/SOME MILK to the boy.
Participants looked more often to the referent corresponding to the focused element and a contrast set member when the focused element was stressed, suggesting that prosodic information was integrated rapidly and used to identify focus under the scope of the focus particle. In a similar study in Dutch, Mulders & Szendrőoi (2016) used accent on one of two arguments of the verb to signal which argument should associate with the Dutch focus particle alleen (‘only’). In a picture matching task, participants judged whether sentences like (41)–(42) matched a picture.
(41) Ik heb alleen SELDERIJ aan de brandweerman gegeven.
‘I only gave CELERY to the fireman.’
(42) Ik heb alleen selderij aan de BRANDWEERMAN gegeven.
‘I only gave celery to the FIREMAN’.
25.3.4 Inferring implicit alternatives
As in the studies described earlier (section 25.1) on the impacts of focus on memory (e.g. Fraundorf et al., 2010), some recent research related to interpretive effects of focus has turned its attention to the status of the alternatives evoked by focus, rather than the focused element itself. In many prior studies, contrast sets associated with focused elements were explicitly introduced in the surrounding discourse: reading studies investigating effects of only on subsequent syntactic attachment introduced explicit contrast set members in a preceding sentence (e.g. Sedivy, 2002); other studies showing that focus-incongruent continuations disrupted reading did so by explicitly mentioning a contrast set member (e.g. Paterson et al., 2007; ‘Jane passed only the salt… and not the pepper’).
Kim et al. (2015) showed that, as in Sedivy (2002)’s self-paced reading study, compre-henders made rapid inferences about contrast based on the discourse context and the presence of only: listeners in a visual world experiment made anticipatory looks to a focused element in the scope of only, when explicit alternatives had been mentioned in the preceding discourse. In addition, listeners readily interpreted unmentioned associates based on either lexical (43) or situation-based (44) categories as focus alternatives.
(43)a. Neil has some pears and some oranges.
b. Alex only has some apples.
(44)a. Neil and Alex are at the baseball game.
b. Alex wants to buy some Coke and some nachos.
c. Neil only wants to buy some hot dogs.
Interestingly, the implicit associates showed general facilitation irrespective of the presence of only, unlike explicitly mentioned alternatives, which were selectively facilitated in sentences with only.
Using cross-modal priming, Husband & Ferreira (2016) investigated the time-course of activation of contrastive (46a) versus non-contrastive (46b) alternatives to a contrastively accented prime word (NURSE in (45)).
(45) The murderer killed the nurse/NURSE last Tuesday night.
(46)a. contrastive: doctor
b. non-contrastive: clinic
c. unrelated: plug
In early processing, contrastive focus led to activation of both contrastive and noncontrastive associates relative to unrelated items. By contrast, in later processing, contrastive associates remained activated while non-contrastive associates were deactivated.
Similarly, Götzner & Spalek (2016) used a probe recognition experiment to compare the activation of contrastive alternatives to a focused element with non-contrastive conceptual associates under the scope of a focus particle (German nur ‘only’ or auch ‘also’).
(47) Carsten wollte gern Obst essen und griff in einen Korb.
‘Carsten wanted to eat some fruit and reached into a basket’
(48) Er nahm sich {nur, auch,____} (Äpfel)F heraus.
‘He {only, also,____} took (apples)F out of it’
Contrastive probes (e.g. ‘berries’) took longer for participants to reject, while noncontrastive probes (nouns related to the focused element by situational knowledge, e.g. ‘maggots’) were unaffected. The authors concluded that the computation of focus alternatives takes into account whether an alternative can replace the focused element.
One question raised by these studies is why Braun & Tagliapietra (2010) and Husband & Ferreira (2016) observed (initial) facilitation for non-contrastive associates, while Götzner & Spalek (2016) observed inhibition. Götzner & Spalek (2016) suggest the difference is related to methodology: their study uses difficulty of rejecting probes as a measure of activation, whereas Husband & Ferreira (2016) use facilitation of a cross-modal prime. Differences in sensitivity to low-level activation between these paradigms could be responsible for the different findings for non-contrastive alternatives.
However, another possibly relevant factor is how the difference between contrastive and non-contrastive associates has been operationalized in different studies. Recall that, in Fraundorf et al. (2010)’s sentence verification experiment, a contrastive accent on a focused element increased correct rejections of explicitly mentioned contrast set members, but did not affect correct rejections of same-category unmentioned associates. In that experiment, non-contrastive alternatives were unmentioned associates that could replace the focused element (Portuguese vs. British). Kim et al. (2015) did not distinguish between contrastive and non-contrastive alternatives, but used conceptual associates (both lexical associates and associates evoked by situational information) which were able to replace the focused element. Husband & Ferreira (2016) used Latent Semantic Analysis to select contrastive associates that had higher similarity scores to prime words than noncontrastive associates did; in addition, contrastive associates were able to replace prime words and preserve both grammaticality and plausibility, whereas substituting non-contrastive associates for prime words would result in either ungrammaticality or implausibility. According to Götzner & Spalek (2016), their non-contrastive associates were related to the focused element by world knowledge, but the key criterion distinguishing contrastive from non-contrastive associates was replaceability—whether an associate could have replaced the focused element. Thus ‘contrastive’ in these studies is characterized by explicit mention, conceptual similarity, replaceability, or some combination of these properties. In evaluating future studies, it will be important to keep track of the way in which contrastive and non-contrastive alternatives are characterized.
25.4 CUES TO FOCUS STRUCTURE
25.4.1 Conflicting cues to focus
Most prior studies manipulated a single cue to focus—clefting, use of a focus particle, the prior discourse context—while holding other cues constant. When different cues signal different focus structures, how are these conflicts resolved?
Paterson et al. (2007) showed that reading was disrupted when an explicit contrast item introduced after the focus particle only was incompatible with the focus structure introduced by only. In dative or double object sentences where either the direct object or the indirect object was preceded by ‘only’, readers showed increased regression path and total reading times when reading a continuation mentioning an incongruent alternative to the focused element (e.g. (49a) continued with (50b)).
(49)a. At dinner last night, Jane passed only the salt to her mother…
b. At dinner last night, Jane passed the salt to only her mother …
(50)a. … but not the pepper as well because she couldn’t reach.
b. … but not her father as well because she couldn’t reach.
In another reading study using eye-tracking, Sauermann et al. (2013) similarly pitted contextual cues to focus against lexical cues. Embedded wh-questions were used to focus either the direct or indirect object in a double object sentence, which was followed by a continuation in which the placement of only signalled a focus structure that was either congruent or incongruent with the preceding context. When only preceded both objects as in (52), continuations with an explicitly excluded alternative (not the cherries/grownups) compatible with the focus structure associated with only (53a) resulted in slower reading times for both the excluded alternative and the following region, compared to continuations where the excluded alternative was compatible with the focus structure supported by the context (53b). This was taken to suggest that the conflict between lexical and contextual focus cues was resolved in favour of the contextually-cued focus structure.
(51) John wondered what Sally would pass the children.
(52) Sally passed only the children the apples …
(53)a. … but not the cherries, because they did not want them.
b. … but not the grownups, because they did not want them.
However, when only appeared after the indirect object and associated with the direct object as in (55), cue conflict disrupted processing earlier in the sentence, and reading times during and after the excluded alternative did not favour either lexically cued (56a) or contextually cued (56b) focus structures.
(54) John wondered who Sally would pass the apples.
(55) Sally passed the children only the apples …
(56)a. … but not the grownups, because they did not want them.
b. … but not the cherries, because they did not want them.
It remains unclear why the conflicting cues to focus structure seem to resolve in different ways depending on the placement of the focus particle. However, it may be relevant that, in sentences like (55), only can associate with the indirect object the children as well as with the direct object the apples, a difference that would have been signalled prosodically in auditory stimuli.
Event-related brain potential (ERP) studies have also investigated cases of conflicting cues to focus structure (Hruska et al., 2000; Stolterfoht et al., 2007). For example, Cowles et al. (2007) measured ERP responses to violations of focus structure, where focus was manipulated contextually, using wh-questions, and using syntactic clefts. In addition to finding a large positivity associated with clefting, in general consistent with sentence-final integration effects, they found that focusing a element incongruent with the focus structure determined by the wh-question elicited an N400-like effect.
25.4.2 Cues to focus projection
Another important question has been how an accented element projects focus to a larger constituent (Gussenhoven, 1983; Selkirk, 1995; Gussenhoven, 1999; see also Schwarz, Chapter 6 in this volume). Birch & Clifton (1995) manipulated both accent placement and context in auditorily presented question-answer pairs; listeners made judgements about the appropriateness of the prosody or made judgements based on their comprehension of the exchange. In the appropriateness judgements, accenting a discourse-new noun phrase (58b) was judged less appropriate than accenting both the verb and noun phrase comprising the verb phrase (58a), suggesting that the accent on the noun phrase was not successful in projecting focus to the entire verb phrase.
(57) Isn’t Kerry pretty smart?
(58)a. Yes, she TEACHES MATH.
b. Yes, she teaches MATH.
By contrast, comprehension of sentences with VP focus was as good when only an argument NP was accented as when both the NP and V were accented. This disparity suggests a task-related difference: NP focus did project to the VP in terms of conveying the intended meaning to listeners, but listeners also showed a preference for both NP and V to be accented in conveying that meaning.
Using the same methodology, Birch & Clifton (2002) showed that accented adjuncts do not project to larger constituents they are part of. Linguistic analyses of intonational structure have indicated that some syntactic constituents, when accented, can project focus to an entire phrase, but other constituents, including adjuncts, cannot project focus. When both the verb and the adjunct were discourse-new, both appropriateness ratings and comprehensibility showed a preference for both verb and adjunct being accented (59a) compared to just the verb (59b).
(59)a. He DROVE SPEEDILY.
b. He DROVE speedily.
In addition to the presence of a pitch accent on the focused constituent, Welby (2003) shows that focus projection is sensitive to the deaccenting of following unfocused material.
25.5 FUTURE DIRECTIONS
25.5.1 Different focus operators
The focus particle only has been a prominent feature in reading time studies. However, the alternative-triggering properties of only are also meant to be shared by other focus-sensitive lexical items (e.g. also, even, too). More recently, studies have begun to investigate differences between focus particles in terms of the time-course of their influence of interpretation, and the nature of the relationship they specify between the focused element and its alternatives.
For example, Filik et al. (2009) compared sentences with either only or even with continuations that were either congruent or incongruent with expectations triggered by the focus particle, as in (60).
(60) Only/Even students taught by the best/worst teacher passed the examination in the summer.
While reading times in the only sentences reflected a difference between likely and unlikely continuations in the critical region (passed the examination), comparable differences only emerged in the post-critical region (in the summer) for sentences with even, suggesting that different focusing mechanisms—even different focus particles—can differ in the time-course and perhaps complexity of evoking contrast. In the case of even, the alternatives triggered may be more complex because they must be ordered on a scale of likelihood (or another contextually salient scale): ‘Even students taught by the worst teacher passed the examination’ gives rise to the inference that the students taught by the worst teacher were the least likely to pass the examination. The corresponding only sentences also encourage a scalar interpretation (the students taught by the best teacher were the most likely to pass the exam), but a non-scalar exclusive reading is also always possible. The complexity of the alternative set required to interpret a focus particle may in turn influence the richness of the context needed to support rapid online interpretation.
25.5.2 Contrastiveness and the sources of contrast sets
As discussed in section 25.3.4, recent studies have begun to investigate the mechanisms underlying the computation of focus alternatives. An important question in future research will be how contrastiveness is characterized—what makes contrastive alternatives different from non-contrastive ones? While alternatives explicitly introduced into the discourse are clearly facilitated in both processing and recall measures, the status of implicit alternatives that are either conceptual associates of the focused element or cannot substitute for the focus is less clear.
Part of understanding where contrast sets come from will involve understanding how non-linguistic context informs comprehenders’ inferences about linguistic alternatives. In reference resolution studies, the non-linguistic context is approximated by visual context. However, it will be important to understand how speakers and hearers estimate contextually salient alternatives based not only on visual information, but also prior knowledge, and knowledge about shared and unshared knowledge among interlocutors.
25.5.3 Cross-linguistic variation
Investigations into focus in English are necessarily constrained by properties of English, such as its rigid word order. Languages with different ordering, morphosyntactic, or tonal properties allow for more nuanced understanding of how the grammatical properties of a language influence how different cues to focus are used in a language. For example, Keller & Alexopoulou (2001) investigated the relative contributions of word order, accent placement, and clitic doubling to the realization of focus in Greek, a free word order language. Ouyang & Kaiser (2015) explore how information status and corrective focus are realized in the prosody of Mandarin, a tone language where acoustic cues are also used to distinguish lexical items.
Even across languages that share relative flexibility in terms of word order, focus phenomena may differ in terms of their sensitivity to discourse factors. In a comparative study, Skopeteas & Fanselow (2011) found that contrastive interpretation of object-fronting is context-dependent and sensitive to referent predictability in some languages (German, Spanish, Greek), but not others (Hungarian). Similarly, Turnbull et al. (2015) suggest that contextual predictability affects the prosodic realization of focus in a way that may vary across languages. More broadly, different cues to focus vary in terms of their inherent reliability—for example, prosodic cues are inherently more gradient than word order cues. Looking at typologically different languages, future research might also ask whether languages are efficient in how they combine the cues available to them for conveying focus.
1 Note that if (16) is initially construed as ‘Only businessmen and no other individuals’, a relative clause parse would be no more expected than in (15). The reduced garden path in (16) suggests that only creates an expectation for a more concrete set of implicit alternatives—in other words, {businessmen who were loaned money at low interest, businessmen who were not loaned money at low interest} is more expected as a contrast set than {businessmen, non-businessmen}.