Asaaaa

Journal of Pragmatics 35 (2003) 695–721 www. elsevier. com/locate/pragma On newspaper headlines as relevance optimizers Daniel Dor* Department of Communications, Tel Aviv University, Tel Aviv, Israel Abstract This paper suggests an explanatory functional characterization of newspaper headlines. Couched within Sperber and Wilson’s (1986) relevance theory, the paper makes the claim that headlines are designed to optimize the relevance of their stories for their readers: Headlines provide the readers with the optimal ratio between contextual e? ect and processing e? ort, and direct readers to construct the optimal context for interpretation.The paper presents the results of an empirical study conducted in the news-desk of one daily newspaper. It shows that the set of intuitive professional imperatives, shared by news-editors and copy-editors, which dictates the choice of headlines for speci? c stories, can naturally be reduced to the notion of relevance optimization. The a nalysis explains why the construction of a successful headline requires an understanding of the readers—their state-of-knowledge, their beliefs and expectations and their cognitive styles—no less than it requires an understanding of the story.It also explains the fact that skilled newspaper readers spend most of their reading time scanning the headlines—rather than reading the stories. # 2002 Elsevier Science B. V. All rights reserved. Keywords: Headlines; Relevance theory; Pragmatics; News value; News framing; Media, communication 1. Introduction This paper is an attempt to suggest an explicit and generalized answer to a very fundamental question in the study of the mass media, i. e. , the question of the communicative function of newspaper headlines.The importance of the role of headlines in the communicative act performed by newspapers can hardly be exaggerated, yet the nature of this role has virtually never been explicated in the literature. As we shall see below, the regular strategy adopted in the literature has been to make ? negrained descriptive distinctions between di? erent types of headlines—news headlines in ‘quality newspapers’; news headlines in ‘tabloid newspapers’; ‘summarizing * Tel. : +972-3–6406521; fax: +972-3-6406032. E-mail address: [email  protected] tau. ac. il (D. Dor). 0378-2166/02/$ – see front matter # 2002 Elsevier Science B.V. All rights reserved. PII: S0378-2166(02)00134-0 696 D. Dor / Journal of Pragmatics 35 (2003) 695–721 headlines’; ‘localizing headlines’, ‘quotation headlines’, etc. —and assign them different types of communicative functions. In this paper, I will suggest an explanatory functional de? nition of newspaper headlines which attempts to transcend the above distinctions in type and explain the very fact that newspapers—all types of newspapers—have headlines in them. The functio nal de? nition to be developed in this paper relies very heavily on Sperber and Wilson’s (1986) technical notion of relevance.Newspaper headlines will be functionally de? ned as relevance optimizers: Newspaper headlines are relevance optimizers: They are designed to optimize the relevance of their stories for their readers. This functional de? nition positions the headline in its appropriate role as a textual negotiator between the story and its readers. It explains why the construction of a successful headline requires an understanding of the readers—their state-of-knowledge, their beliefs and expectations and their cognitive styles—no less than it requires an understanding of the story. It reduces the di? rences between the di? erent subtypes of headlines mentioned above to a matter of tactical choice: As we shall see, all the di? erent subtypes target the same functional goal, that of relevance optimization, although they do it in di? erent ways. The literatu re on newspaper headlines covers a wide range of theoretical and empirical topics, all the way from the grammar of English headlines to the e? ects of headlines on news comprehension and recall. 1 Surprisingly, however, the literature dealing directly with the communicative function of headlines is rather sparse.I will review it in the next section. In Section 3, I will brie? y introduce Sperber and Wilson’s theory, and then develop the notion of relevance optimization. In Section 4, I will apply the notion of relevance optimization to newspaper headlines. In Section 5, I will present the results of an empirical study conducted in the news-desk of the Israeli national newspaper Ma’ariv, where I followed the process of headline production from close range. 2 I will show that the set of intuitive professional imperatives, shared by news-editors and copy-editors, which dictates the choice of headlines for speci? stories, can naturally be reduced to one meta-imperative: Ma ke the headline such that it renders the story optimally-relevant for the readers. In Section 6, I will apply the relevance-based conception to the analysis of tabloid headlines. In Section 7, I will deal with the role of the reader in this framework, and show that my relevance-based theory explains some of the more intriguing behavioral patterns manifested by newspaper readers—especially the fact that many skilled readers On headline reading, interpretation and recall, see Henley et al. 1995), Leon (1997), Lindemann (1989), Perfetti et al. (1987), Pfau (1995) and van Dijk (1988 and references therein); on headline production, see Bell (1984, 1991), Fasold (1987) and Chang et al. (1992); on the grammar of headlines, see Bell (1984), Jenkins (1990) and Mardh (1980); on metaphors in headlines, see de Knop (1985); on headlines from a cross-linguistic perspective, see Dierick (1987) and Sidiropoulou (1995). 2 Between 1996 and 1998, I worked as a senior news-editor and head of the news-desk in Ma’ariv.This was a period of very intensive participant observation: I was involved in the decision-making process concerning the formulation of thousands of headlines. The e-mail exchanges which were analyzed for this paper were randomly collected throughout this period—from other senior editors. 1 D. Dor / Journal of Pragmatics 35 (2003) 695–721 697 spend most of their reading time scanning the headlines rather than reading the stories. In the concluding section, I will sketch some of the larger-scale implications of my theory, and suggest some directions for further research. . Multiple types, multiple functions Traditionally, newspaper headlines have been functionally characterized as short, telegram-like summaries of their news items. This is especially true with respect to news headlines. Van Dijk (1988) couches this traditional insight within his discourseanalytic framework: ‘‘Each news item in the press has a Headline . . . and many have a Lead, whether marked o? by special printing type or not. We also have an elementary rule for them: Headline precedes Lead, and together they precede the rest of the news item.Their structural function is also clear: Together they express the major topics of the text. That is, they function as an initial summary. Hence, as in natural stories, we may also introduce the category Summary, dominating Headline and Lead. The semantic constraint is obvious: Headline+Lead summarize the news text and express the semantic macrostructure. ’’ Obviously, some newspaper headlines do provide what seems to be a summary (or abstract) of their stories, but the general theoretical conception which takes this to be the essential function of the headline seems to be too narrow, for at least three complementary reasons.First, even the most prototypical news headlines, those which appear in what is sometimes called ‘quality newspapers’, do not always summarize their s tories. Some headlines highlight a single detail extracted out of the story, and others contain a quotation which the editor decided should be promoted to the foreground. As we shall see below, some headlines even contain material which does not appear in the news item itself. The fact that headlines do not always summarize, but sometimes highlight or quote, has been noted by di? erent writers.Bell (1991), for example, makes a distinction between headlines which ‘‘abstract the main event of the story’’, and headlines which ‘‘focus on a secondary event or a detail’’ (p. 188–9). Nir (1993) distinguishes between headlines which function as ‘‘a summary of the story’’ and ‘‘headlines which, rather than summarize the story, promote one of the details of the story’’ (p. 25). 3 Second, the traditional notion of headlines-as-summaries de? nitely does not capture the function of head lines in more popular newspapers, and especially in tabloids. This point has been made by di? rent writers, most notably by Lindemann (1990). As Lindemann shows, tabloid headlines rarely summarize their stories, are not always telegram-like, and in many cases are not even informative. Lindemann discusses the 3 Note that none of the above writers goes beyond the descriptive labeling of the di? erent types of headlines to suggest explicit theoretical de? nitions and explanations. This fact is most clearly demonstrated by Bernstein and Garst (1982), quoted in Lindemann (1990), who claim that ‘‘†¦ the headline contains the main highlight of the story. Since it is the most conspicuous part and the part that is read ? st, the copy editor must present the essence of the news before he goes further’’. In this short quotation, Bernstein and Garst seem to equate the essence of the story with its highlight, thus equating the function of summarizing with that of h ighlighting. 698 D. Dor / Journal of Pragmatics 35 (2003) 695–721 function of tabloid headlines in poetic terms: They present the reader with a ‘‘fairly complex riddle’’, which, ? rst, triggers frames and belief systems in the reader’s mind, and, then, gets resolved in the ensuing text. Thus, the following headline, (1) NO-LA-LA! The Frogs Get Bored with Bed raps the reader ‘‘in the treadmill of well-established cliches and prejudice’’, through the use of such expressions as frogs, no-la-la and bed, and is then informationally resolved in the intro: ‘‘The days of the great French lovers are over—froggies just don’t fancy it any more. A third of women and a quarter of men told a nationwide survey they found bedtime one big yawn’’. Implicit in Lindemann’s analysis is the assumption, that the function of tabloid headlines is so radically di? erent from their function in qual ity newspapers, that the two cannot be theoretically uni? ed.As I will show below, the relevance-based analysis will allow exactly for that—to my mind, a very welcome theoretical result. The third reason to reject the traditional conception is the simple fact that headlines seem to have an additional, pragmatic function, beyond the semanticallyoriented function which is supposed to be captured by the headline-as-summary analysis. Bell (1991) says that headlines are a ‘‘part of news rhetoric whose function is to attract the reader’’ (p. 189). Nir (1993) claims that the headline has ‘‘to attract the attention of the reader and provoke the reader to read the whole story’’.In a sophisticated analysis of the semiotics of headlines, Iarovici and Amel (1989) explicitly contend that the headline has a ‘‘double function’’: ‘‘The implicit convention between author and reader regarding the intentio n of correlating a text to another text as a headline, and regarding the formal marking of this quality by a privileged position, concerns the double function of the headline: a semantic function, regarding the referential text, and a pragmatic function, regarding the reader (the receiver) to whom the text is addressed.The two functions are simultaneous, the semantic function being included in and justi? ed by the pragmatic function. †¦ The main function of the headline †¦ is to alert the reader (receiver) to the nature or the content of the text. This is the pragmatic function of the headline, and it includes the semantic one. The headline enables the reader to grasp the meaning of the text. The headline functions as a plurality of speech acts (urging, warning, and informing)’’ (p. 441–443). The challenge posed by the above assertions is that of theoretical uni? cation.At least two questions are involved: First, can we functionally de? ne the headline in a way which would transcend the above distinctions between the di? erent semantically-oriented functions? In other words, is there a generalized function which summarizing headlines, localizing headlines and quotation headlines have in common? Second, can we de? ne the headline in a way which would transcend the distinction between the above semantic function and the parallel pragmatic function which D. Dor / Journal of Pragmatics 35 (2003) 695–721 699 headlines ful? ll?I would like to claim that this theoretical move becomes possible once we couch the functional analysis of headlines within the framework of Sperber and Wilson’s (1986) theory of relevance. 4 3. Relevance theory and relevance optimization Sperber and Wilson’s theory is an attempt to reduce a very complex set of phenomena having to do with communication and interpretation to a very constrained set of explanatory, cognitive notions. In its essence, the theory is one of cognitive cost-e? ectiven ess: It claims that human cognitive processes are geared to achieving the greatest possible cognitive e? ct for the smallest processing e? ort. This metaprinciple is incarnated in Sperber and Wilson’s technical notion of relevance. Let us take a look at the fundamental tenets of this framework: Our starting point is the individual mind: Every individual mentally represents in his or her mind a huge set of assumptions. Assumptions are propositional entitiesthey are the type of entities that can be believed to be true. Our assumptions may include, among other things, information on the immediate physical environment, expectations about the future, scienti? hypotheses, religious beliefs, anecdotal memories, general cultural assumptions, beliefs about the personal lives of our acquaintances, knowledge about politics and history, beliefs about our own emotions, fears and hopes, and so on. Each of the assumptions represented by the individual has a ‘‘strength’â⠂¬â„¢ for that individual. The strength of the assumption for the individual is the level of con? dence with which the individual holds to the belief that the assumption is true. The strength of the assumption is a function of its cognitive processing history.Thus, for example, ‘‘assumptions based on a clear perceptual experience tend to be very strong; assumptions based on the acceptance of somebody’s word have a strength commensurate with one’s con? dence in the speaker; the strength of assumptions arrived at by deduction depends on the strength of the premises from which they were derived’’ (p. 77) Note that the strength of an assumption for the individual has nothing to do with its objective validity—individuals may have a very strong belief in assumptions which are totally false, and vice versa.When an individual hears, or reads, a novel assumption, he or she always interprets it in a context. The notion of context is used here as a psychological construct: It is a subset of the assumptions which the hearer already represents in his or her long-term memory. Informally, what the mind of the individual does in the process of interpretation may be thought of as a comparison of the new assumption with the subset of assumptions represented in the individual’s memory. Sperber and Wilson name the cognitive apparatus responsible for this process of comparison- ‘‘the deductive device’’.The comparison of the novel assumption with the existing 4 The general notion of pragmatic relevance, which is not to be equated with Sperber and Wilson’s technical one, plays some role in van Dijk’s (1988) analysis of news selection. However, van Dijk does not make the connection between his notion of relevance and the function of headlines, which he takes to be summaries of their texts. 700 D. Dor / Journal of Pragmatics 35 (2003) 695–721 assumptions may have di? erent types of out puts: It may turn out, for example, that the novel assumption already exists in the individual’s long term-memory, in which case it is not new for the individual.Or it may be new, in which case it may either be in line, or in contradiction with some of the already existing assumptions. If, for example, the novel assumption contradicts existing assumptions, and if it is strong enough, the process of comparison will end up with the weakening of the existing assumptions. In some cases, it may even end up with the erasure of those assumptions. If the new information is in line with some existing assumptions, it may serve to strengthen them some more. Moreover, the union of the new assumption with some existing assumptions may lead to the deduction of additional assumptions.Thus, for example, if the individual already represents the assumption that ‘‘whenever Peter goes to a party, it becomes a success’’, and he or she now learns that ‘‘Peter came to Bill’s party’’, then the deductive device deduces an additional assumption, namely that ‘‘Bill’s party was a success’’. To the extent that the comparison of the new assumption with the old ones results in a change to the individual’s set of prior assumptions (if it either adds new assumptions, or weakens or strengthens existing ones), we say that the new information has a contextual e? ect for the individual.Now, the following point is crucial: The deductive device does not compare every novel assumption to the entire set of assumptions represented in the individual’s longterm memory. Doing this would be cognitively impossible. This means that the comparison is done with some subset of existing assumptions. This, in turn, raises a very important question: How does the deductive device choose this subset? Traditionally, pragmaticists have assumed that the context for the interpretation of an utterance is simp ly given: It consists of the immediate environment and the information explicitly mentioned in the conversation prior to the utterance.Sperber and Wilson ? ip this assumption on its head and suggest a radical alternative: They show that the deductive device has to update the context for the interpretation for each new assumption, and that the speci? c subset of existing assumptions which is chosen for the context is determined, at least partially, by the content of the new assumption. In cognitive terms, this means that the order of events in comprehension is reversed: It is not that the deductive device ? rst sets the context, and then interprets the new assumption.On the contrary, the deductive device has to partially ? gure out the meaning of the new assumption, retrieve a speci? c subset of assumptions from long-term memory, store them in its own short-term memory, and then make the comparison. An example should make this radical conception rather intuitive. Take a look at the f ollowing exchanges: (2) A: How are you? B: Not so good, Mary has that ear-infection again, I’m worried. (3) A: How are you? B: Great, I just bought the tickets. We’re ? ying to Beijing in exactly four weeks.In order to interpret B’s answer in each of these exchanges, A has to compare them to a subset of existing assumptions. The proper context in (2) should include assumptions about the identity of Mary, her relation to B, her medical history, ear- D. Dor / Journal of Pragmatics 35 (2003) 695–721 701 infections, and so on and so forth. The proper context in (3) should include assumptions about B’s travel plans, the identity of her companion, or companions, whatever assumptions A has about Beijing, and so on and so forth. Obviously, these assumptions are not stored in A’s short-term memory on a permanent basis.A’s deductive device has to retrieve these assumptions from long-term memory, and only then make the comparison and deduce the contextual e? ects. We may now make two parallel cognitive assumptions regarding the process I have described. First, we may assume that in its appropriate context, a new piece of information has a certain number of contextual e? ects, which, at least theoretically, can be counted. Practically speaking, when we deal with interpretations of actual utterances by real people, we do not know exactly how to make the measurement, but the dea itself is intuitive enough for us to accept. We may be pretty certain that in di? erent contexts, the same piece of information may yield di? erent amounts of contextual e? ects, and that in the same context, some pieces of information would yield more contextual e? ects than others. Second, we may assume that the work of the deductive device involves some mental e? ort, which—theoretically speaking, again- may be measured. 5 Other things being equal, for example, the computation of a more complex piece of information will take more e? ort than the computation of a simpler one.Moreover, the construction of a new context for interpretation also involves some mental e? ort: To the extent that the interpretation of the novel piece of information necessitates the retrieval of a larger set of assumptions from long-term memory, the mental e? ort involved in the interpretation process would be greater. The measurements of contextual e? ect and mental e? ort constitute the basis of Sperber and Wilson’s notion of relevance: (4) Relevance for an individual (p. 145): a. An assumption is relevant to an individual to the extent that the contextual e? cts achieved when it is optimally processed are large. b. An assumption is relevant to an individual to the extent that the e? ort required to process it optimally is small. It is crucial to understand that this is not a de? nition of relevance in some objective sense, but a claim concerning the way our minds make relevance judgments about new assumptions: We consider new assumptio ns to be relevant if they carry a contextual e? ect at a reasonable cognitive price. We judge new assumptions to be irrelevant if they do not carry a contextual e? ect, or if the computation of the contextual e? ct entails too much of a mental e? ort. Note that this is a comparative, gradual conception of relevance, rather than a binary one: New assumptions are not either relevant or not; they are more or less relevant than others, in di? erent contexts, for di? erent people. Thus, for example, a regular newspaper reader will prob5 Sperber and Wilson (1986, p. 130) conceptualize about the measurement of contextual e? ects and cognitive e? ort in terms of physico-chemical changes which occur in the brain as a result of the processing of the contextual e? ects. 702 D. Dor / Journal of Pragmatics 35 (2003) 695–721 bly judge a piece of local news to be more relevant than a piece of foreign news because (i) the potential contextual e? ect derivable from the local news would probab ly be larger; and (ii) the e? ort needed to interpret the foreign news (especially in terms of the retrieval of the appropriate set of assumptions from long-term memory) would probably be larger. This judgment need not be made consciously; the reader may simply skip the foreign-news page, or note that ‘foreign news is boring’. Note, however, that the very same reader may take the trouble to read the foreign news to the extent that their contextual e? cts would be worth his or her processing e? ort. This may be the case, for example, if the story is about a country which the reader intends to visit; if some people which the reader knows are there; if there is a local angle to the foreign story; if the foreign story has a global consequence which is felt locally, and so on. Finally, note that this technical notion of relevance should not be equated with relevance in the ordinary sense of the word. Relevance in this ordinary sense may be thought of as the measurement of th e association, or congruence, between some content and its context of interpretation.Thus, a news story will be relevant in this sense to the extent that it is about those issues which are directly related to the readers’ lives and interests. Indeed, relevance in this sense may play a role in news value judgments. Note, however, that a story may be relevant in this ordinary sense but very low on relevance in the technical sense (if it is long and complicated to read, for example, or if it does not carry a lot of new information); and it may be irrelevant in the ordinary sense, but high on relevance in the technical sense- if its potential contextual e? cts justify the construction of a new context for interpretation. Now, our technical de? nition of relevance is addressee-oriented, but it may actually tell us something of importance about the role of speakers in communicative contexts. Think about a speaker, Ann, who is trying to tell her addressee, John, a story. Being a coo perative communicator, Ann would like to make the story as relevant for John as possible. How should she go about achieving this goal? According to relevance theory, she has three principled strategies which she can try to dopt: (i) First, Ann can try to compress the largest possible number of new assumptions (those which are new for John) into her story: Other things being equal, the more new assumptions the story contains, the more contextual e? ects it may have for John. In the worst-case scenario, the story will not contain any assumptions which are new for John, in which case he is going to ? nd it totally irrelevant. Ann de? nitely needs to ? nd a way to do better than that. In the best-case scenario, on the other hand, the story will contain a very large number of new assumptions.As we shall see below, this is not always going to be possible. (ii) Second, Ann can try to minimize John’s processing e? ort: Other things being equal, the smaller the e? ort he has to put in , the greater the relevance of the story is going to be for him. In the worst-case scenario, the story is going to be too long and complicated, and John is going to lose interest. Ann de? nitely has to avoid that. In the best-case scenario, on the other hand, the D. Dor / Journal of Pragmatics 35 (2003) 695–721 703 story will take a very minimal e? ort to process. Again, this is not always going to be possible. iii) Third, Ann can try to manipulate the context in which John is going to interpret her story. Other things being equal, the closer the context of interpretation is to the optimal one, the more contextual e? ects the story is going to carry for John. (Note that in regular conversation we regularly take the trouble to manipulate our addressee’s context of interpretation, especially when we wish to ‘‘change the topic of conversation’’: We say things like ‘Oh, I wanted to tell you something about Bill’, or ‘talking abo ut school, did you hear about Bill’s exam? ’).In the worst-case scenario, John is going to try to interpret Ann’s story in the wrong context, and the interpretation is going to yield no contextual e? ects. In the best-case scenario, the story is going to be interpreted in the most appropriate context, yielding the maximal amount of contextual e? ects. This, again, is not always going to be possible. Now, it is very important to realize that the three strategies mentioned above are not only completely intertwined, but are also in direct competition with each other. This is why achieving the maximal results associated with each of the strategies is not always possible.This is so for the following reasons: (i) First, every new assumption which the speaker adds to the story does not only contribute to the overall number of contextual e? ects- it also adds to the overall processing e? ort. Thus, the new assumption adds to the overall relevance of the story only to th e extent that it clearly adds more contextual e? ect than processing e? ort. To the extent that the new assumption adds more to the processing e? ort than to the contextual e? ect, it actually reduces the overall relevance of the story. In this case, more information results in less relevance.So, the attempt to maximize relevance simply by maximizing the amount of new information is bound to end up in failure. The speaker has to ? gure out the optimal amount of information which would not result in relevance reduction due to processing e? ort. (ii) Second, Ann may de? nitely try to maximally reduce John’s processing e? ort by making her story short, simple and clear, but this reduction will not necessarily result in maximal relevance: This is so, because the reduction in the story’s complexity characteristically reduces the number of its potential contextual e? ects. The reduction of processing e? rt will enhance the relevance of Ann’s story only to the extent t hat the amount of e? ort saved is larger than the amount of contextual e? ects lost. So, again, Ann cannot simply reduce John’s processing e? ort to the minimum. She has to ? gure out the optimal amount of e? ort which would not result in relevance reduction due to loss of contextual e? ects. (iii) Third, the number of contextual e? ects which John may deduce from Ann’s story is not just a function of the sheer number of new assumptions in the story, but a function of the interaction between these new assumptions and the context of interpretation.This means that Ann should not just provide 704 D. Dor / Journal of Pragmatics 35 (2003) 695–721 John with the optimal number of new assumptions, but also take care to provide him with those speci? c assumptions which would yield the maximal amount of contextual e? ects in the appropriate context, and at the very same time direct John to construct that speci? c context. This complicates our relevance considerations to a considerable extent, because the construction of the appropriate context entails a signi? cant amount of processing e? ort.Consequently, in principle, the construction of the appropriate context may eventually result in relevance reduction due to the increase in processing e? ort. Thus, the construction of a partial context for interpretation may sometimes be the optimal strategy. As we have seen, Ann’s role as the story-teller is going to be that of relevance optimization: She will need to provide John with the optimal ratio of contextual e? ect and processing e? ort. This, I would like to claim, is exactly the generalized communicative function which newspaper headlines are supposed to ful? l: They are designed to optimize the relevance of their stories for their readers. 4. Newspaper headlines as relevance optimizers Consider the following story, from the Israeli national newspaper Ma’ariv: (5) The bodies of John Kennedy Jr. , his wife Caroline and his sister-in-la w Lorraine were discovered yesterday in the ocean, at a depth of 30 meters, 10 kilometers away from Martha’s Vineyard Island, where they were headed on Saturday. Senator Edward Kennedy, John’s uncle, arrived at the site where the bodies were found, in order to identify them. Kennedy Jr. ill be buried in NY in the coming days. This news item requires a certain amount of mental e? ort to interpret. To begin with, the paragraph requires some e? ort to read: It consists of about 70 words, and is grammatically fairly complex. Moreover, the news item requires the construction of a context for interpretation- one which includes whatever the reader knows about John Kennedy Jr. , his family, their disappearance two days before, the relevant geography, and probably at least something about the Kennedys’ history. As we have said before, the construction of this context takes an additional e? rt. Let us assume, for the sake of simplicity, that the interpretation of the enti re story will require the ordinary reader to invest a certain amount of e? ort, let us dub it E(story). Now, to the extent that the reader manages to construct the appropriate context and read the passage, the story carries a certain amount of contextual e? ects: It changes a lot of factual assumptions the reader represented in his or her long-term memory (e. g. , the assumption ‘John Kennedy Jr. is alive and well’ is replaced by ‘John Kennedy Jr. died in an irplane accident’), and it changes, weakens or strengthens a great many related assumptions having to do with, for example, the inescapable tragedies of the Kennedy family, the life-styles of the rich and D. Dor / Journal of Pragmatics 35 (2003) 695–721 705 famous, the blindness of fate, the risks involved in ? ying your own plane, and so on and so forth. Obviously, di? erent readers will probably deduce di? erent sets of contextual e? ects from the story, but for the sake of simplicity, let us a ssume that the ordinary reader will deduce a certain amount of contextual e? cts, let us dub it C(story). The relevance of the story for the ordinary reader will thus be: R(story)=C(story)/E(story). Now, let us take a look at the headline the newspaper gave to the story: (6) John Kennedy Jr. ’s body found How much e? ort does the reader have to invest in interpreting the headline? Obviously, much less than E(story): The headline is a single, short and simple sentence, comprising ? ve words, and the e? ort needed to read it is insubstantial. The e? ort needed to construct the context for the interpretation of the headline is also signi? antly smaller—the reader does not need to retrieve the sets of assumptions having to do with the geography of the story, with Senator Ed Kennedy, and so on. For the sake of simplicity, let us make the arbitrary assumption that E(headline) equals 10% of E(story). Now, how many contextual e? ects can the reader deduce from the headline? Su rprisingly, when the headline is interpreted in its reduced context, a signi? cantly large subset of the contextual e? ects of the entire story survive. Obviously, some things are missing—for example, the fact that Kennedy’s wife and his sister-in-law were found too—but Kennedy’s death, its signi? ance within the tragic history of the Kennedy family, and the more general implications of the story are clear contextual e? ects of the headline. Let us adopt a conservative estimate: For the ordinary reader, C(headline) equals 50% of C(story). As a simple calculation clearly shows, our estimates entail that the headline multiplies the relevance of the story by ? ve (! ). It saves much more on the processing e? ort than it loses on the contextual e? ects. This is exactly what a headline should do. A short and simple text, it optimizes the relevance of the story by minimizing processing e? ort while making sure that a suf? ient amount of contextual e? ects are d educible within the most appropriate context possible. Just like Ann, our story-teller, the headline does not adopt an all-ornone strategy of either reducing processing e? ort to zero, or maximizing new information, or constructing the most appropriate context for interpretation. Rather, it attempts to optimize the ratio between processing e? ort and contextual e? ects- and thus optimally negotiate between the story and the ordinary reader. Note that for the optimization of relevance to be successful, the right material should be chosen for the headline.Consider, for example, the following three clauses as alternative headlines for the Kennedy story: (7) a. Caroline Kennedy’s body found b. Sen. Edward Kennedy arrived at the crash site. 706 D. Dor / Journal of Pragmatics 35 (2003) 695–721 c. The bodies of John Kennedy Jr. and his wife Caroline were discovered yesterday in the ocean, near Martha’s Vineyard Island. The ? rst two alternative headlines (7a and 7b) ar e probably as easy to read as the actual one, and we may assume that they require the construction of a very similar context for interpretation. However, they do not carry the same amount of contextual e? cts as the original. The third alternative (7c) carries a slightly larger number of contextual e? ects than the original, but it very obviously requires much more processing e? ort. Thus, all three alternative fall short of providing optimal relevance. Is the original headline in (6) a summarizing or a highlighting headline? It is hard to tell. The important point, however, is that from our theoretical point-of-view the summarizing-highlighting distinction is simply not that crucial: Summarizing the story is just one tactical approach to relevance optimization.Highlighting the most intriguing aspect of the story, or reproducing the most interesting statement quoted in the story, may have the very same result. It may turn out, for example, that the quotation or the highlighted aspec t carry more contextual e? ects than the summary of the whole narrative. In this case, the rational thing to do would be to promote them to the headline- and thus optimize the relevance of the story for the readers. The choice between these di? erent tactical approaches is in part a matter of the editorial style of the newspaper, and to a very large extent a matter of the experience and creativity of its editors.For every given story, some headline options are going to suggest themselves. The editor may opt for a summarizing headline, a highlighting headline or a quotation headline- depending on which type of headline will provide optimal relevance. 6 Moreover, the editor may manipulate the length and complexity of the headline, and its speci? c contents. And again, these manipulations, to a very large extent, are going to be relevance-oriented. 7 In the following section, I will present the results of an empirical study conducted in the years 1996–1998 in the news-desk of th e Israeli national newspaper Ma’ariv. In the study, I followed the decision-making process leading to the choice of headline for a large number of news items. As the results of the study clearly indicate, the set of professional intuitions shared by the editors, concerning the properties of the ‘right’ or ‘appropriate’ headline, are theoretically reduced to our notion of releI will discuss the choice of tabloid-type headlines later on. An anonymous referee notes that some text manipulations may not be relevance-oriented. Thus, for example, some manipulations may have to do with spacing on the page, and others with political considerations. I agree with the ? st point. In Dor (2001), however, I show that relevance-oriented manipulations play an extremely important role in processes of political framing. 8 There are currently three national newspapers in Israel: Yediot Ahronot and Ma’ariv are considered to be the popular newspapers, whereas Haâ €™aretz is considered to be the quality, high-brow one. Yediot Ahronot and Ma’ariv, however, are not tabloids in the regular, American-European sense. They contain a variety of ‘‘serious’’ news items which is not that di? erent from that of Ha’aretz, and are distinct from it especially in writing style and graphic design.In all three newspapers, headline formulation is considered part of the editorial process, and reporters do not formulate headlines for their stories. 7 6 D. Dor / Journal of Pragmatics 35 (2003) 695–721 707 vance: The most appropriate headline for a news item is the one which optimizes the relevance of the story for the readers of the newspaper. 5. The notion of relevance and the art of headline writing In general, news editors do not work with a very explicit de? nition of what headlines are, or of their communicative functions.What they do work with is a cluster of professional intuitions—gradually develop ed ‘in the ? eld’, and never seriously explicated—concerning the properties of what we might call, rather informally, the ‘right’, ‘appropriate’, or ‘good’ headline. When asked to provide an explicit de? nition of what a headline is, senior newspaper editors usually give an answer of the type: ‘I don’t know what headlines are, but I can tell a good one when I see it’. This answer is actually a pretty accurate rendition of a very fundamental sentiment: Professional knowledge is practical, not theoretical.However, when presented with a news-item, and asked to choose a headline out of a set of alternatives, experienced news editors do so with extreme ease and e? ciency. Moreover, senior editors in the same newspaper have a very high rate of agreement on the preferred headline. This means that experienced news editors know a great deal more about the functional properties of headlines than they ever explicate . In this sense, headline production is more similar to an artistic activity than, say, to the practice of an exact science. This a? ity with the arts is very clearly re? ected in the trial-and-error process which beginning copy-editors go through as part of their on-the-job training procedure. Rather than receive their professional education in the form of explicit lecturing, beginning copy-editors in Ma’ariv simply start out working: They are assigned a new-item, and are asked to rewrite it and suggest a headline for it. The result is then reviewed by the senior editor in charge, who, in most cases, rejects the suggested headline and writes a di? erent one, which eventually gets published.Sometimes, the copy-editor is asked to suggest the alternative headline, which is, again, reviewed by the editor in charge. Deadline pressure usually does not allow for long explanations: When the process is over, the copy-editor gets another story, suggests a headline, which usually gets rejected, and so on and so forth. This process goes on for years, and in a real sense never ends: In Ma’ariv, each and every suggested headline is sent to the senior editor in chief, in the form of an electronic message, to be approved or rejected, even if the copy-editors have years of experience behind them.Obviously, the rate of rejected headlines goes down with time, when the trained editor internalizes the set of implicit intuitions shared by the other, more experienced editors, but even very experienced editors get some of their headlines rejected some of the time: Sometimes, for example, the editor in chief knows something about the wider context of the story which the copy-editor was not aware of. The fact that these real-time negotiations about the headlines are done in writing, by e-mail messages, allowed me to follow the process of headline formulation from very close range.I collected 134 e-mail exchanges, concerning 134 news-items, and analyzed the semantic-pragm atic di? erences between the rejected and approved headlines. In some cases, I asked the editor in charge to reconstruct the reasons for 708 D. Dor / Journal of Pragmatics 35 (2003) 695–721 the rejection of the suggested headline. I then extracted a list of ten properties, which I shall call the properties of the appropriate headline. I submit that this list is an accurate rendition of the set of implicit intuitions shared by experienced news editors in Ma’ariv.In the following section, I will present the ten properties, each with its representative example, and show that the list is actually reducible to one professional meta-imperative: Make the headline such that it renders the story optimally-relevant. Three notes should be made at this point: First, the following discussion should not be thought of as an attempt to construct a theoretical framework, but as a description of a set of professional intuitions, shared by news editors, concerning the properties of the à ¢â‚¬ËœÃ¢â‚¬Ëœappropriate headline’’. In other words, I do not intend to make any signi? ant claim concerning the theoretical status of the ten properties to be discussed below. Quite obviously, some of the properties seem to bear close resemblance to some principles discussed in the literature under the rubric of news value (e. g. , in Galtung and Ruge, 1965; Bell, 1991); other properties may remind the reader of Grice’s conversational maxims. I will leave the elaboration of these resemblances for further research, and concentrate on the possibility of reducing the entire set of properties to the relevance-based meta-intuition mentioned above. Second, the properties are to be thought of as default conditions, rather than obligatory ones. It is not the case that every headline should have all 10 properties. It is the case that a headline which meets any of these conditions is better than a headline which does not, and a headline which meets a larger number of the conditions is better than a headline which meets a smaller number of them. Thus, for example, the ? rst property—‘headlines should be as short as possible’—should be read as saying: ‘other things being equal, a shorter headline is better than a longer one’.Finally, The headlines presented in the next section are translated from the Hebrew original. I chose to keep the translation as literally accurate as possible, and avoided translating the headlines into ‘‘headlinese’’, because Hebrew headlines do not usually have the telegraphic syntax characteristic of English headlines. 5. 1. The properties of the ‘‘appropriate headline’’ [1] ‘‘Headlines should be as short as possible’’. Newspaper headlines are, quite obviously, very short clauses.The actual length of each particular headline, however, is a matter of considerable debate and negotiation between senior editors an d copy editors: Copy-editors, especially the beginners, suggest longer headlines, attempting to ‘capture’ as much of the story as possible. The senior editors shorten the headlines to a considerable extent- leaving out whole chunks of information. One of the expertises mastered by experienced editors is the ability to decide which parts of the story should be left out of the headline. The following exchange is a very typical example. The I thank an anonymous referee for his/her discussion of this point.The referee also wondered whether any of the ten properties may be reducible to another. Thus, for example, the referee felt that properties [6] and [7] are mirror-images of each other, and should thus be put together. I assume that this can indeed be done. For me, however, the more important point was that the editors I talked to felt these were two separate, although obviously related, principles. As I am interested here in the description of intuitions, rather than in the construction of a theoretical framework, I will discuss the two principles separately. 9 D. Dor / Journal of Pragmatics 35 (2003) 695–721 709 tory is about a youth gang which was caught red-handed mutilating gravestones in a military cemetery in Haifa. The copy-editor suggested the following headline: (8) Haifa: A youth gang was caught mutilating gravestones in the city’s military cemetery The head of the news-desk ordered the copy-editor to shorten the headline, in the following way: (9) Haifa: A youth gang was caught mutilating gravestones Note that the decision to shorten the headline is not without its price: We have lost a piece of information, i. e. , that the gravestones were mutilated in a military cemetery, which means we have lost some contextual e? ects.We have, however, gained in reading e? ort. The shorter headline is simply easier to read. This is a very clear example of relevance optimization by e? ort reduction. The editor in charge decided that the loss in contextual e? ects is smaller than the gain in reading e? ort. [2] ‘‘Headlines should be clear, easy to understand, and unambiguous’’. Formulating a headline to a complex story is not an easy task. Copy-editors sometimes suggest headlines which come out unclear, di? cult to understand, or unintentionally ambiguous. Such headlines are rejected, and the copy-editor is asked to formulate a clearer, simpler, unambiguous headline.In the following example, the article tells the story of a police drama in the city of Ramat-Gan, where a single arsonist threatened the city for weeks, burning down vehicles every night. On that speci? c night, the police caught a suspect, but had to release him after the ‘real’ arsonist took out to the streets again, burning down more vehicles to prove that he was not caught. The copy-editor suggested the following headline: (10) The ‘real arsonist’ from Ramat-Gan proves: You Haven’t caught me The headline was rejected because it was considered unclear and unnecessarily ambiguous.It raises more questions and vaguenesses than it actually answers: Who is the ‘real arsonist’? Is there an ‘unreal arsonist’? How has the ‘real arsonist’ proven that he wasn’t caught? By whom? The copy-editor was ordered to formulate a clearer headline. This was his second attempt: (11) The arsonist ‘was caught’- and the vehicles in Ramat-Gan went on burning This version is much clearer: It makes clear that a claim was made that the arsonist was caught, which turned out to be false, and it makes clear that, on that day, 710 D. Dor / Journal of Pragmatics 35 (2003) 695–721 fter the capture incident ended, some vehicles in Ramat-Gan were still burning. There is, however, a potential ambiguity here: The headline can be read as making the claim that the arsonist set the vehicles on ? re before the police made the false claim, and that they went on burning after the incident. The headline was rejected again, and the copy-editor suggested the third version, in (12), which was ? nally accepted and published as it is. Note that the only di? erence between (11), the rejected headline, and (12), the approved one, is in the tense of the verb in the second clause. 12) The arsonist ‘was caught’- and the vehicles in Ramat-Gan go on burning This headline makes it clear that the arsonist is still on the loose, and is still in the habit of setting vehicles on ? re. Finally, the story is captured in a clear, simple and unambiguous fashion. This reduces processing e? ort to the necessary minimum— and optimizes the relevance of the story. [3] ‘‘Headlines should be interesting’’. This quality plays a central role in the negotiations between copy-editors and senior editors. Many suggested headlines are rejected on the grounds that they are ‘not interesting’.What is usuall y meant by this rather obscure phrase is that the editor imagines that the readers of the paper will not ? nd the headline interesting enough. In terms of our relevance-based theory, this means that the editor estimates that the amount of contextual e? ects carried by the headline will not justify the amount of reading e? ort. The copy-editor is then asked to read the article again, and look for a ‘more interesting’ piece of information to foreground to the headline. In the following example, the story includes an interview with Uri Lubrani, IDF’s Chief of Military Operations in Southern Lebanon.General Antoin Lahed, who is mentioned in the rejected headline, is the Commander in Chief of the South Lebanon Army (SLA), a Christian Militia which has traditionally been IDF’s ally in Lebanon. The context of the story is a wave of rumors, according to which the IDF plans to withdraw from Southern Lebanon, thus leaving General Lahed and his people on their own ag ainst their Islamic rivals: This is the headline which the copy-editor suggested: (13) Lubrani: ‘There was no secret meeting with General Lahed’ Whether or not there was a secret meeting between IDF o? ials and General Lahed on the previous day is hardly an interesting question. After all, IDF o? cials and General Lahed meet on a regular basis, and their meetings are usually kept secret. In our terms, the headline does not carry a substantial amount of contextual e? ects. The headline was rejected, and the copy-editor came up with the following alternative: (14) Lubrani: ‘There is no plan to evacuate SLA seniors to Europe’ Whether or not there is a secret plan to evacuate SLA seniors to Europe is very obviously much more interesting.If there was such a plan, this would be a pretty D. Dor / Journal of Pragmatics 35 (2003) 695–721 711 remarkable sign that the IDF is on its way out of Lebanon. Lubrani’s ? at denial can be interpreted in more than one way: We can take him for his word, or assume that he chose to deny the existence of the plan for tactical reasons- at any rate the denial has interesting implications. It is de? nitely more interesting than just another meeting- in our terms, it carries more contextual e? ects for the same amount of processing e? ort. Note that this is a very good example of the signi? ant role of headline writing in the workings of a newspaper. The two headlines, the rejected one and the suggested one, make it quite obvious that the Lubrani interview did not contain any remarkable scoops, and that the editor had to dig in to ? nd something which was worth promoting to the headline. As the senior editor’s decision makes clear, even negative statements, ? at denials of the type that Lubrani suggested as answers to the reporter’s questions, have di? erent amounts of relevance, and the one which was more relevant than the other was promoted. 4] ‘‘Headlines should contain new information’’. A major topic for negotiations between copy-editors and senior-editors has to do with the question of whether the readers already know what the copy-editor decided to promote to the headline. Obviously, editors do not really know what their readers know, but their estimates of their readers’ state of knowledge play a central role in the decision-making process. This makes perfect sense within our relevance-based framework: A headline which does not contain novel assumptions cannot bring about contextual e? ects, and is thus irrelevant.In Ma’ariv, as in any other daily newspaper, estimates of the readers’ state of knowledge are based primarily on what has already been communicated by the other mass-media, especially the evening news on TV. If the content of the proposed headline for the next morning has already appeared in the news the night before, most chances are it will be rejected. The following headline, for example, was rej ected on these grounds: (15) The Austrian Chancellor Arrived for a visit; will meet Netanyahu today The copy-editor had a hard time ? nding an alternative headline. This is what he came up with: (16) O? ials in Jerusalem hope for the Austrian Chancellor’s visit to run smoothly This headline was accepted, for two reasons: First, it carries the implication that o? cials in Jerusalem are worried that the visit might not run smoothly—an angle on the visit which was new. Second, it connects the story to prior events and expectations: The visit of the British foreign minister had just ended the day before, and that visit was full of political hurdles and diplomatic embarrassments. As we shall see below, connecting a story to its wider context is another important property of good headlines. 712 D. Dor / Journal of Pragmatics 35 (2003) 695–721 5] ‘‘Headlines should not presuppose information unknown to the readers’’. This principle, in a sense , is the mirror-image of the previous one: The information in the headline should de? nitely be new—but it cannot be ‘overly new’. Headlines should only presuppose information which is already part of the mutual knowledge established between the newspaper and its readers. In terms of our relevance-based framework, every presupposition in the headline should already be available within the readers’ context of interpretation. Otherwise, the computation of the headline will result in zero contextual e? cts. Consider, then, the following headline: (17) Advanced negotiations on the establishment of the second Israeli-owned casino in Jericho This headline presupposes the existence, or at least the potential existence, of the ? rst Israeli-owned casino in Jericho. The ? rst news concerning the plans to build this casino, the ? rst one, were published only a few days before the above headline was suggested. According to the editor in chief, the readers had not yet registered the future existence of the ? rst casino in their long-term memory—it was premature to treat it as a presupposition.The copy-editor was asked to change the headline, and came up with the following alternative: (18) The ? rst casino in Jericho will be operational in February In this headline, the establishment of the ? rst casino in Jericho is not presupposed, but reported as part of the news. This is much better. But the editor in chief asked the copy-editor to rephrase the headline again, this time for a di? erent reason: The proposed headline forces the reader to calculate the amount of time it will take till the casino will be operational. This adds to the processing e? ort. The alternative, which was eventually published, reduces this e? rt, thus optimizing the relevance of the story: (19) The ? rst casino in Jericho will be operational in a year [6] ‘‘Headlines should include names and concepts with high ‘news value’ for the readersâ €™â€™. 10 Experienced editors develop a sense of the ‘news value’ of names and 10 This property, and the next one, reminded one anonymous referee of Ariel’s (1988, 1991) accessibility theory (see also Kronrod and Engel 2000). The resemblance, however, is rather super? cial. Ariel is not interested in the speci? c contents of the referring expressions, but in their general cognitive and structural properties.Thus, for example, referring expressions which function as high accessibility markers (e. g. , personal pronouns, ? rst names) are used by speakers when they assume that the referents are highly accessible for their addressees; lower accessibility markers (e. g. , long de? nite descriptions, full names etc. ) are used when the speaker assumes low mental accessibility for their addressees. The point I am making here, however, is not that newspaper editors prefer certain types of referring expression (e. g. , full names) to others (e. g. , last names), but that they prefer certain referents (e. g. , famous ? gures) to others. D.Dor / Journal of Pragmatics 35 (2003) 695–721 713 concepts: They very easily identify names and concepts which should appear in headlines- and those which do not. In terms of our relevance-based theory, experienced editors know, or at least believe they know, which names and concepts will carry a large number of contextual e? ects for their readers. The following example demonstrates this very clearly. Some background: Two days before the following story was to be published, a story in one of the national newspapers revealed that the popular musicians contracted to perform in Israel’s 50th Jubilee were paid high sums of money- at the ax-payers’ expense. In our story, some other popular musicians reacted to the revelation and angrily declared that they were willing to perform in the Jubilee for free. This was the headline suggested by the copy-editor: (20) A group of artists suggests an alternative for the Jubilee: ‘‘we are willing to perform for free’’ This headline, a classic summarizing headline, was rejected in favor of the following, which replaces the expression ‘a group of artists’ with the names of two celebrity musicians, Shimi Tavori and Margalit Tsan’ani: (21) Shimi Tavori and Margalit Tsan’ani: ‘‘we are willing to perform in the jubilee for free’’Note, ? rst, that the replacement headline in (21) actually loses some of the information we had in (20): The group of artists included many more musicians than just the two mentioned in (21). The point, however, is that the two are the most famous members of the group, and names of well-known popular musicians always carry a lot of contextual e? ects: This is so, because their names direct the readers to construct a much wider context for interpretation, which includes whatever we know about them, their personalities, their views, t heir social background, their wealth, their life styles, and so on. 20) is a headline suited for a regular news story about the 50th Jubilee; (21), on the other hand, is a headline for what is basically a gossip story about Shimi Tavori and Margalit Tsan’ani. [7] ‘‘Headlines should not contain names and concepts with low ‘news value’ for the readers’’. This is the mirror-image of the last property: Some names and concepts do not have ‘‘news value’’ for the readers, and experienced editors avoid promoting them to the headline. In our terms, these names and concepts do not help the reader construct the optimal context for the interpretation of the headline.In the following example, the copy-editor attempted to promote such a name to the headline, and was intercepted by the editor in charge. Some background: The story has to do with an accident in which two military helicopters collided in the air on their way to I srael’s northern border, an accident in which more than 70 soldiers were killed. New immigrants to Israel receive an immigration grant from the Jewish agency, and it turned out that the agency asked the parents of one of the soldiers killed in the accident, a newly-arrived immigrant, to pay back his grant.The copy editor suggested the following headline: 714 D. Dor / Journal of Pragmatics 35 (2003) 695–721 (22) The Michaelov family was ordered to return the immigration grant given to their son- who was killed in the Galilee The editor in charge rejected the headline, and ordered the copy-editor to produce another one, which would not contain the name. This was the alternative headline, which was ? nally published: (23) The Jewish agency refused to let a family, whose son was killed in the helicopter accident, keep his immigration grantNote that the move from the de? nite ‘the Michaelov family’ to the inde? nite ‘a family’ made it di? cult for the copy-editor to keep it in subject position, and dictated an overall grammatical change from passive to active voice. [8] ‘‘Headlines should ‘connect’ the story to previously known facts and events’’. Just like the last two principles, this one has to do with the construction of the appropriate context for interpretation. A story interpreted on its own, as an isolated event, will carry a certain amount of contextual e? ects.The same story can carry more contextual e? ects to the extent that the readers interprets it within a wider context, which includes previously known facts and events. Consider the following example, which is a report on a violent taxi robbery in the city of Haifa: (24) The driver was beaten and thrown out- and the stolen taxi was later found stuck in the mud The editor in charge, who rejected this headline, asked the copy-editor to connect the incident in Haifa to the rising trend of taxi robbery throughout Israel. The f ol