1How do we talk when we are not sure what we are talking about? How do we construct meaning in the midst of a semantic crisis? These questions focus on the role of speech in understanding an extra-ordinary situation. The case of a major ongoing event is one of these situations and provides the relevant corpus: the first four hours of CNN’s live broadcast on 9/11 were studied to address the problem of meaning construction mechanisms when meaning making processes are challenged. The corpus is constituted of unscripted, spontaneous, naturally occurring speech typical of the breaking news format: over sixty speakers - journalists, eyewitnesses, experts, officials or reporters - talk with one another, together, one after the other as to understand what is going on. If discourse analysis sheds light on the particular meaning making processes in such situations, the analysis of prosody, and more broadly of speech semantics, can account for fine-grained phenomena. Of particular interest in this paper is the sense of urgency that unmistakably emerges from the interaction.
2Our approach of the data can be described as sociocognitive semantics. Meaning is viewed as both the product of an interactional and collaborative construction (in line with Sacks et al. 1974, Du Bois 2014) and as a set of cognitive processes (in line with cognitive linguistics [Langacker 1987, Talmy 2000] or psycholinguistics approaches [Pickering et Garrod 2013]) involving representational mechanisms in relation with such faculties as perception and memory. Our research question incites us to consider linguistic productions as the results of conceptualizations and underlying cognitive processes aiming at making sense of the information retrieved from the environment. But, because the data comes from media speech, it implies a discursive construct (such as television codes) and is interactional in nature. That is why a sociocognitive approach seems best fit to tackle the data (Danino 2014 for a detailed account). Both aspects of the corpus make use of prosody as a semantic component. We must however adopt a view of prosody that allows for an account both in interactional and in semantic terms, in line with a three-fold modeling of the corpus. This analytical model aims at distinguishing the constraints that weigh on utterance productions in order to account for their form and meaning.
3We isolated three types of constraints, each corresponding to an analytical level: situation, interaction and proposition. The situation refers to the environment, as it is perceived in the case of major ongoing event. The situation then provides one constant and shared attentional focus, creates shock and metal charge. It constrains in turn the interaction: the attentional focus "becomes" a constant and shared topic for the ongoing interaction whose goal is to describe what is going on and understand what it means. Because speakers have socially marked identities - as we will develop further - their responsibilities and roles within the interaction vary. This constrains what we called proposition: the actual utterances. One example is question design: journalists and reporters do not phrase their questions similarly, and the addressee (witness or official spokesperson) also impacts the question forms (Danino 2014 for a detailed account).
4Prosody is relevant at each of the three levels of analysis (situation, interaction, proposition). A parallel can be drawn with scales of linguisticity, after Di Cristo (2013) and Crystal (1976). Indeed, in our corpus, differences must be made between prosodic phenomena that can be attributed to technical conditions, to the organization of turn-at-talk (Sacks et al. 1974) or to a process of information structuring.
5We argue that elements in each level contribute to building a sense of emergency in the discourse. The breaking news format, which we will characterize in a first part, is immediately recognizable due to a series of audiovisual codes in which prosody plays a role. In a second part, we will try to determine the interaction of prosody with phenomena of saliency, both perceptual and conceptual. In a third part, we will examine the role of prosody in interaction management and speaker selection. A fourth and final part will focus on dysfluencies, namely silent pauses. In other words, we will successively confront prosody to multimodality, pragmatics, semantics and information structuring.
6The breaking news format itself creates a sense of urgency. The interruption of regular programs, the obvious lack of script on the part of the journalists and the different visual elements on the screen (starting with the mention “live”) all contribute to a multimodal construction (Steen and Turner 2013) that signals the specific format. The discourse has an open-ended structure (after Montgomery 2007), which means that it has no composition or planned structure according to which it is developed. The breaking news format is completely mapped onto the events to which it is consubstantial. The format itself implies the urgency and emergency of a crisis situation from which we have no distance since it is (or is presented as) ongoing. The construction of the split screen (featuring two places in two frames on the one screen) is typical: the broadcast tries to be everywhere the events are, all the time. Another indication that the discourse is open-ended and unscripted is the lack of convergence between the discourse and the images: speakers are usually not talking about what is shown. This disjunction is a major deviation from the broadcast news format (Montgomery 2007 for a detailed characterization). In breaking news broadcast, very few images are in fact available.
7From the discursive point of view, the breaking news format (BNF hereafter) has a peculiar "narrative" progression. Lacking an oriented structure (toward an end point), the discourse is based on a series of echoes. Because it is structured around the studio, the BNF is spatially organized: going out to where reporters and witnesses are to gather information before coming back to the studio where anchors (journalists) sum up the information. Temporally, it is stuck in an overbearing present tense. Just as the discourse structure is open-ended, the representation of temporality is that of an unbounded interval centered around now. Langacker’s (2011) notion of epistemic immediacy, built upon by De Wit and Brisard’s (2014) epistemic contingency explains the space-time construct in our corpus. Time of speech is constantly relevant and the events are considered (and linguistically represented) as coincident with the unfolding speech. The present tense has a "modal import" (Langacker op.cit.) because it necessarily refers to the act of speech and the speaker themselves.
- 1 All instances are from the corpus. Transcription is orthographic and includes the following symbols (...)
8In this context, the discourse can only be redundant: information comes from different sources. Information reliability is augmented if different sources converge. In the same line of ideas, since information is fragmentary, scarce and uncertain at best, it can only be built incrementally. Speakers in the corpus rightly use the metaphor of the different pieces of a puzzle when describing the state of information available at time of speech. In order for the puzzle to emerge as a clear image, journalists in the studio regularly offer recaps, usually followed by a description of the current situation viewed as a result of the summed up events, as in (1) and (3)1:
(1) M1 (...) we believe a commercial jet / has crashed into one of the towers of the world trade center / and you can see the smoke billowing out there there are flames billowing out there / anda a commercial jet . crashing / into one of the towers /
(2) W: (...) ok once again if you’re just warning us- -if you’re just joining us- the: the breaking story that we’re following out of New York City / within the span of 18 minutes two separate planes crashing into the world trade center // em the rescue operation-s underway- not clear- -we’re gonna show you this is the second plane after the first tower xx already on fire--this happened just minutes ago / let the pictures tell the story
(3) M: (...) for those of you just joining us let(’s) us briefly recap / what we know. About an hour ago about 8:45 eastern time / one plane crashed in-to / erm the tower the world trade center tower on the right-the first of those towers such you can see behind me erm and then about a half hour later erm a second plane crashed into the tower number two that’s the one to the left / where the darker smoke is billowing out right now / erm # /
9These recurring recaps are informational checkpoints organized in an echoic network of significant redundancy. They are significant despite being redundant because the repetitions are never strictly identical. Information becomes progressively more complete and more stable. (4) and (5) are the first two turns of the corpus: from the first to the second one, differences (and echoes) in the verbalizations start building the process of incremental information gathering:
(4) W: Yeah /// (1,36 s) This just in - you are looking at erm ov’ obviously a very disturbing live shot there - that is the world trade center and we have unconfirmed reports this morning / that a plane crashed into one of the towers / of the World Trade Center - CNN center right now is just beginning to work on this story / obviously calling our sources and trying to figure out exactly what happened but / clearly something / relatively devastating happening this morning there / on the south end of the island of Manhattan. / That is once again a picture / of one of the towers of the World Trade Center. /// (1,23 s)
(5) M1: We can see these pictures, it’s obviously em // something devastating is happening again unconfirmed report / that a plane has crashed in one of the towers there we are efforting / more information on this subject / as it becomes available to you /// (1,37 s)
10The quest for information starts in W’s turn and continues in M’s (efforting more information) so that the preterit on crash becomes a present perfect; the several unconfirmed reports become a singular converging piece of information in the second turn. Echoes, whether structural or lexical, strengthen the impact of theses differences by contrast. Information becomes more stable. The World Trade Center, fully referred to in the first turn is already considered given information in the second turn with the implicit turn of phrase one of the towers Ø. In just one turn, some common ground and shared knowledge has been built.
11The frames of discourse (Fillmore 1977, 1985, Goffman 1974), be it the technical and semiotic dimension of the BNF or the echoic discourse based on significant repetitions, create a sense of emergency conditioning speech production. This creates a sense of hammering rhythm and temporal contraction around the time of speech. Time keeps repeating itself: the adverb again epitomizes this specific discourse progression. This is especially so in the case of 9/11 because each iteration of similar sub-events (the four crashes) modifies the interaction and its pace. The discourse appears to be constantly catching up with the events it deals with. Time of speech is mapped onto the events (and not the reversal). Legallois (2011: 7) sums up this idea when he distinguishes between the historical present, which presents the events as coincident with time of speech, and the reporting present, where time of speech is represented as coincident with the events. In this case, the events are the organizing factor.
12The discourse is indeed concerned entirely by the events. If it can be analyzed as an example of urgent speech (under temporal and informational pressure), it can also be considered a discourse about emergency as a principle of thematic coherence. From this point of view, does prosody, as marking the materiality of the spoken word, factor in? Or is emergency restricted to the lexical content?
13Prosody can refer to a number of phenomena. In this paper, the term is used along the lines of Di Cristo’s approach:
Les projets communicatifs sont conçus préalablement sur la base de critères socio-pragmatiques et rhétoriques, avant d’être configurés en ces structures conceptuelles linguistiquement encodables (...) [la prosodie] participe aussi à l’élaboration d’autres types d’informations qui seraient utiles à la production de la parole impromptue et qui se rapporteraient notamment à l’attitude dialogique du locuteur et à la mise en place de la structure informationnelle des messages (en particulier la distinction focus/fond) (Di Cristo 2013 : 58)
14The second part of the quote is of particular interest to us. If we agree with a model of linguistic production that involves prosody early on and grants this component decisive semantic contributions, the idea of its taking part in the dialogic attitude of the speaker and the construction of interactional meaning is particularly seductive. So is the idea that it takes part in the information structuring process especially through the partition between foreground and background. Prosody might just contribute to the levels of interaction and proposition in the three-level model we proposed for our corpus. Let’s start with words.
15Considering their key meanings both in semantic and discursive terms, temporal deixis adverbs such as again and (right) now have been analyzed. If the second features far more occurrences than the first (286 vs. 77), again still is crucial as it signals echoes and recaps in the discourse. Because it frames a particular sequence (recaps), we look at the prosodic patterns. The adverb is mostly used in sentence- and sequence-initial position. It does not always receive an emphatic stress but it is usually preceded by a pause, at least 0.2 second long. It is our contention that this pause reinforces its role as a discourse particle, opening a new interactional sequence. Thus, when it is used as a discourse particle (Aijmer 2002 for a definition), position and speech segmentation correlate a specific function of the word. Position and function are clearly different when again is an adverb expressing iteration of a process (it happened again).
16The case of (right) now is less clear-cut. Its position varies as we find instances in sentence initial, medial or final positions. If right now is rather adverbial, now can be a discourse particle as well where it mostly contributes to text segmentation and topic transition (Aijmer 2002 for a study on now). In our corpus, we found no significant correlation of these uses with prosodic patterns. There is no systematic or even recurrent pausing before or after the target word. In the same line of ideas, it can be detached from the rest of the sentence or turn unit (which in our case corresponds to an interpausal segment) but it can also be prosodically integrated. Even the topic shift use of now does not entail demarcation by pausing. On the contrary it is frequent that speech in the relevant segment is quite leveled (prosodic integration, no peaks in F0 or pitch). It can even co-occur with acceleration on the part of the speaker as if to make the transition smoother or quicker.
17It seems that strong and robust prosodic patterns in our data have more to do with interactional and discursive function rather than with specific lexical tokens. Given our conception of meaning (encompassing semantic, syntactic and pragmatic phenomena, in line with foundations of cognitive linguistics [Langacker 1987, Talmy 2000, Fauconnier and Turner 2002, Col et al. 2012]), prosody is indeed a semantic component but seems to primarily concern the intersubjective dimension of language. That is to say that prosody impacts meaning construction because it impacts information structuring as a consequence of alignment and dialogic coordination. It has been noted that prosody is one element of linguistic representation to which participants align (Pickering and Garrod 2013). The notion of alignment, along with the discrepancy observed in the patterning between right now and again, points to the idea of routines and possibly local routinization. Pickering and Garrod (2004: 13) give the following definition:
A routine is an expression that is “fixed” to a relatively great extent. First, the expression occurs at a much higher frequency than the frequency of its component words would lead us to expect (e.g., Aijmer 1996). (In computational linguistics this corresponds to having what is called a high “mutual information” content; Charniak 1993.) Second, it has a particular analysis at each level of linguistic representation. Thus, it has a particular meaning, a particular syntactic analysis, a particular pragmatic use, and often particular phonological characteristics (e.g., a fixed intonation).
18In line with these elements, we suggest that again in media speech might just be one of these routines. Routines, they further remark, can be understood in terms of the highly repetitive quality of dialogue - interaction, in more linguistic terms. As outlined in the first section of this paper, echoes and repetitions are in our corpus both an automatic default configuration of the interaction, and a conscious cognitive construct. Because repetition imposes a rhythm onto the interaction, creating a sense of urgency, it is striking that some repetitive elements should relate to stable processes, reducing novelty and uncertainty in a corpus confronted with a major semantic crisis. This apparent contradiction tends to provide further argument to the necessity of considering prosody at different scales. At word – or phrase – level, the situation does not seem to be straightforward. From the point of view of interaction though, and building on the psycholinguistic notion of alignment, the spoken dimension of our data may yield a more contrasted picture.
19Prosody can be tackled at utterance level. Building on what has been said on right now, we can first observe that the information the adverb carries (strict coincidence with time of speech, minimally) is usually given elsewhere in the utterance if only in the verb tense marking. In the case of the BNF, the entire "speech arrangement", for lack of a better word, is built on this idea. The information that right now brings to the utterance is thus usually quite redundant. Here are a few examples to illustrate this view (6 to 9). They share the use of perception verbs in the present tense, sometimes even in the present progressive, which, by definition, implies coincidence of the act of perception with the perceived event. Other temporal deixis adverbs (still) and mentions of live pictures reinforce the temporal anchoring in time of speech with a default “as we speak” grounding. Elements pertaining to this grounding, or "viewing arrangement" in Langacker’s terms (2011) appear in bold face.
(6) W: iti- it is an- a remarkable scene as we are seeing right now fla:mes still coming out of the windows / black smoke billowing from what appears to be all si:des / em . obviously windows shattered and steel jetting out from the structure [right now]
(7) M4: (...) // we’re looking at live pictures right now /
(8) G29: (...) you can see the people jumping out the window they’re jumping out the window right now // oh my god //
20 Or again, in this passage where right now is repeatedly used:
(9) W: Right now we’ve got Sean Murtagh he is a CNN producer on the telephone right now - Sean what can you tell us about what you know?
G1: This is Sean Murtagh - I was em just standing on the em // em Vice president of finance /// (1,12 s)
M1: Sea[n-
G1: [-Vice president of finance for CNN
M1: Sean-Sean we’re on the air right now . - what can you tell us about the situation? /// (2,6 s)
G1: Hello?
M1: Yeah Sean you’re on the [air right now
- 2 Analor is a software created by the LaTTiCe lab (UMR 8094) providing automatic speech segmentation (...)
21The framed occurrence shows an echo within the utterance. In a few words, there are two prosodic prominences in the segment, one on air and the second on now. Volume peaks on both words to reach 78 db. Similarly, a tentative analysis with Analor2 (see reference) would tend to confirm these two prominences. Caution is all the more necessary that we could not normalize the Praat input due to constant ambient noise, to different recording technologies used throughout our corpus and to the diversity of speakers. However indicative the results are, the intonational contour remains quite clear: fall-rise and vowel lengthening on now. This suggests that the meaning of right now is to be redundant, that is, emphasizing already given but still crucial information: strict coincidence with time of speech. The same logic holds true for (7) where the prosodic echoes link live with the second half of right now. These elements are only preliminary and have no other ambition than suggesting that prosody might have to do with information processing, rather than information structuring. Since it signals what information bears repeating, prosody contributes to building patterns, favoring efficient processing. In a pressured context, redundancy and strong patterning could help speakers since it maintains crucial information in a state of activation, thus avoiding the costly cognitive task of activating information. It could also play a central role by creating saliency: by contrast, new information is immediately heard and can be allotted more processing resources. In order to test this hypothesis, within the limits of this paper, we would like to present a case of misunderstanding.
22If the analyses we have presented so far tend to show the convergence of semantic and prosodic markers, where both linguistic subsystems cooperate to mutually support each other, it is not always the case. The BNF is not a type of spoken media where speakers can rely on scripted contributions, pre-planned interventions. Everything is constructed on the go, from the wording of speakers’ contributions to the determination of who intervenes when in the broadcast. Planning efforts are made of course but fail explicitly in several instances of our corpus:
(10) M: (...) this is one of those situations that is extraordinarily chaotic / erm even . even in the best of planning I think it’s fair to say / that it is chaotic / (32 : 255-256)
(11) M: (...) erm we’ll try now to connect with KDKA TV / in Pittsburgh /// (1,28 s) well just / so much for planning we try we will try again / (...) (45 : 810-811)
23Not much is rehearsed or even prepared so that technical problems are part of the interaction and the constraints that shape it. But some problems are directly interactional. Let us consider the following example. The sequence presented at length in (12) features M closing a two-way interview with “Jamie” and opening for Allan (G23) to utter his report, using a single term of address, namely the intended cospeaker’s first name. G23 proceeds to deliver the information he has gathered so far: he first recounts what is in fact the second tower collapsing, then tries to describe the consequences of that event by describing the current situation in the street. Around him, since he is “on-site”, he can see the rescue workers coming and going, the firefighters trying to control a “roaring” fire while the street is “filled with smoke”, everyone covering their mouths. The interactional misunderstanding happens in the framed passage, at the end of G23’s report.
(12) M: Jamie thank you erm so we have planes hitting the pentagon or a plane / hitting the / pentagon two planes hitting / the world trade centers / em towers // i-in New York--Allan Dodds-Frank joins us on the phone / in lower Manhattan Allan?
G23: Aaron em just erm / two or three minutes ago there was yet another erm collapse or explosion--I’m now out of sight a good Samaritan’s taken me in on Dwayne street // xxx at quarter to eleven there was // another collapse or explosion following the ten thirty collapse // of the second tower // and a firefighter rushed by us / estimated that fifty stories went down // erm / the street filled with smoke // it was like a fire erm forest fire roaring down a canyon // now is ??I think Patty Sagba told you // all of Manhattan is covered em downtown Manhattan is covered with // thick white ash ’n / building material // em the ambulances have been coming now from as far as Long Island // erm all the rescue workers have been equipped with gas ?or / face builder masks // and erm firefighters have been arriving even by
pick-up truck /// (1,2 s)
M: e[rm]
G23: [other]wise the streets are now deserted
M: em Allan thank you Allan Dodds-Frank / in Manhattan
24 G23 stops his report at “pick-up truck”. If the silence that follows the turn is indicative that G23 reached the end of his contribution, it is also indicative that M did not expect it. 1,2 second is a long time to say nothing, especially on air, even more so in an interaction between two media professionals. Something went wrong but prosody is not to blame: G23’s last words are pronounced with a very clear falling intonation (low fall even). This is all the more so remarkable that this pattern is quite rare in the data. It is only typically found at the end of reporters’ long one-turn contributions (reports) like this one. Here is the last segment visually represented with Praat:
25“Face builder masks” was clearly pronounced with a continuative intonation, visible in the plateau-shaped contour, thus excluding that idiosyncratic intonation could challenge interaction. G23 masters the codes associated with prosodic cues. So when he concludes with a fall on pick-up trucks (green arrow), it is clear he has reached the end of his turn. M, as the anchor, is automatically selected as next speaker. But M does not take the floor, leaving a 1.2 second silence before volunteering to take the floor again. It is remarkable that M and G23 resume speaking at the exact same time: their judgment of inappropriately long silence is strikingly synchronized (as evidenced by the overlapping talk). M might have been distracted by some producers speaking in his earpiece when G23 concluded his turn; it might not be that he did not pick up on G23’s cues; it might have been that he was not paying attention. Whatever the reason may be, he was distracted enough at this point to think he had a few seconds ahead of him before being selected as next speaker, and when G23 reacted to the misunderstanding and consequent silence, he volunteered a piece of information that was not randomly selected. The semantics of his next utterance sheds light on what the speakers thought went wrong.
26Indeed, he offers a characterization of the streets around him. G23’s second turn is not only building on his previous turn (adverb now, definite article before streets), he is offering a conclusion of sorts. His first turn stopped as he was describing activity in the streets, reinforced but the last verbal form, a present prefect progressive. In a sense, we can assume that M was expecting further description of what was going on around G23 in the current situation. This interpretation seems licensed by G23’s use of otherwise in his second turn: despite the activity previously described, there no longer is anything to comment upon since everyone fled. The end of his first turn presented a conflict between a conclusive prosody and semantic contents calling for further development. As it were, semantics won, and further characterization is offered in his second turn. It offers a conclusion of sorts (this is how things stand now) and an explanation of what went wrong at turn transition point.
27What this example points to is the fundamentally interactional quality of spoken media which constantly rely on an array of resources from different modalities. Because the visual modality is not controlled or prepared in the BNF, because images arrive with a delay to the studio, because the cameras were where they were (and not positioned with care and reflection), the broadcasters have to rely on what they have. They cannot rely on graphics or visual aids that weather presenters or newsreaders have at their disposal (Montgomery 2007, Steen and Turner 2013). If we go back to the hypothesis suggested above, the fact that semantics won over prosody tends to confirm that prosody helps processing meaning, rather than structuring. Indeed, in terms of structure, prosody clearly indicated the end of sequence. M however expected more information based on the contents of G23’s turn. The temporary misalignment can be due to a conflict between prosody and semantics, but it can also be considered as an indicator that prosody has to do with processing rather than structuring or segmenting. These elements invite us to consider the type of alignment that prosody can help create to ease the cost of information processing.
28Interaction is paramount in our corpus, and crucial to the BNF characterized by a stronger than ever reliance on the collaborative construction of collective speech. If prosody is not necessarily concerned by resonance, as we will describe it, spoken media, the BNF included, are. Based on the Question & Answer model, the BNF clearly exhibits dialogic passages and some others that are in fact monologic. Just like Smith (2003) establishes "modes of discourse" for written texts based on criteria such as temporal progression or primary referent, it is possible to correlate modes of discourse in our corpus with the sociodiscursive profiles of speakers. Table 1 presents the categories of speakers pertinent to our data. Interactional dyads in spoken media are determined by the identity of the two speakers and their degree of mastery of public speaking. Briefly put, the less professional a co-speaker is, the more scaffolding of the interaction the journalist will do. This scaffolding is apparent when considering "dialogic syntax".
Speakers
|
Media Professional
|
Mastery of public speaking
|
Types of interaction
|
Modes of Discourse
|
Journalists
|
+
|
+
|
dialogic / monologic (interaction management)
|
Narration / Description / Commentary
|
Reporters
|
+
|
+
|
monologic
|
Description (narration)
|
Witnesses
|
-
|
-
|
dialogic
|
Narration
|
Experts
|
-
|
+/-
|
dialogic
|
Commentary
|
Officials
|
+/-
|
+
|
monologic / dialogic
|
Commentary / Description
|
29Du Bois (2014), in his discourse-functional framework, tackles interaction from the point of view of resonance. This term refers to the repetition of some elements of an utterance, typically in the next speaker’s turn. According to Du Bois, what the notion brings to the table, that similar psycholinguistic notions do not, is engagement. Speakers not only coordinate and cooperate in dialogue as a cognitive activity; they take a stance. They mean something and orient to the constructed meaning as well. According to the stance triangle (Du Bois 2007) speakers not only orient to one another in dialogue, they also attend to the content of what is being said. This orientation implies monitoring but it crucially involves stancetaking. The following extract from the corpus features obvious resonance although stancetaking is less transparent:
(13) G2: well I was in the past train and there was a huge explosion sound everyone came out (looks at the building, gestures her description) / a large section of the building is blown out /
around like the 80th [floor]
M2: [didy-] was it hit by something or was it so[mething inside? ]
G2: (shakes her head) [it was inside]
[it was inside]
M2: [it was ins]ide
G2: (she nods) because it is- it looks -out everything was coming out //
M2: everything [was coming-]
G2: [all the window]s were coming out, all the papers wer[e coming (out)]
M2: [what is] on the sidewalk?
G2: // em (looks down to the sidewalk) I didn’t see anything I ran
30M2, a reporter, and G2, an eyewitness, both of them in the streets of Manhattan, construct collaboratively the interaction in which they are engaged. The cooperation is here maximal to the point that G2 in her second turn, first aligns gesturally (she shakes her head) before anticipating M2’s second candidate answer about the cause of the explosion. The second set of overlaps is remarkably in synch. Having rejected the idea of external factors (was it hit by something) G2 anticipates the alternate possibility that it was something inside. The timing is impeccable, exhibiting that the temporality of speech is “mastered” unconsciously by the speakers. The next set of overlaps can also be interpreted in terms of maximal cooperation: G2 repeats the information to make it perfectly clear that she assumes this piece of information: she does not only align with M2’s propositional content, she appropriates it as her own. M’s echoic utterance works as a confirmation token, acting as a conclusion of sorts, reinforced once again by G2’s gesture (she nods). G2 further elaborates on the description of the event for which a type-label has been collaboratively found. M2 interjects himself during her pause (less than 1 second long) but stops when she resumes elaborating. If the overlap has to do in this case with speaker selection, G2’s contribution still exhibits resonance but with her own previous turn. The predicate were coming out is reused with a more specific subject, all the windows. The second half of her turn echoes the structure a second time with the addition of the determiner phrase all the + plural noun which constitutes a second resonating structure. This double repetition in G2’s turn works as reinforcement of her descriptions as if arguments kept piling up in favor of her recounted perceptions. Overlap and resonance are two distinct mechanisms that can account respectively for the collaborative nature of the interaction as cospeakers orient to one another and for the stancetaking activity of the cospeakers oriented toward content and its management. As it turns out, these mechanisms can coincide. That is why we used the phrase "maximal alignment".
31The following extract can also be accounted for in these terms even if the target mechanisms apply in a slightly different manner. G1 is an eyewitness reached over the phone and interviewed by W, a female journalist in the studio, during the very first minutes of the broadcast.
(14) W: Sean, what kind of plane was it?--was it small plane, [a jet -]
G1: [it was a:] erm it was a jet // erm look(ed) like a two engine jet // erm maybe a 737
W: you’re talking about a large passenger co[mmercial jet]
G1: [a large pas]senger commercial jet. /// (1 s)
32Under discussion here is the type of plane involved in the crash, which of course has crucial consequences for the interpretation of the event, its nature and magnitude. The extract opens (again) on a proposed alternative in W’s turn. G1 takes up W’s proposed description qualifying the plane (a generic and rather vague term) as a jet, implying passenger transportation. The target word jet is the object of W’s second turn explicating the salient aspect of the definition with a metadiscursive introduction segment (you’re talking about...). The witness does not wait for W’s turn to be over: large and passenger are indeed the key aspects that are under consideration. Remarkably, G1 will keep processing W’s turn as he starts producing his own, voluntarily repeating W’s full noun phrase extensively. G1’s following turn is thus more than a simple confirmation: G1 takes a clear stance making a strong claim about the type of plane and consequently the type of event. The debate here is not so much referential as it is categorical. The label jet is then repeated four times in two pairs of turns - the first pair establishes the reference, the second one qualifies it. The passage is collaborative but exhibits the dissymmetry between W and G1: if G1 is the source of the information and must consequently assume full responsibility of the chosen description, it is W’s responsibility in turn to act as an information filter within the broadcast for a silent cospeaker, the viewer (since breaking news can also be characterized as any television broadcast as a parasocial interaction [Horton and Wohl 1956]).
33The study of dysfluencies has received increased attention from such different areas of research as psycholinguistics, conversation analysis or phonology. The reason why they can be tackled from such diverse perspectives has to do in our opinion with the fact that dysfluencies can be explained in terms of management and maintenance problems related to language production (e.g. Schnadt and Corley 2006), interaction organization (e.g. Heldner and Edlund 2010) and / or speech segmentation (Zellner 1994). They are multifaceted phenomena the study of which logically benefits from varied approaches. Pauses could be another way to test our hypothesis that prosody might have more to do with information processing than structuring.
34Our transcription aimed at reproducing any dysfluency such as "stuttering" and false starts:
(15) G37: [no people] I I I was em / and I’ve I was there right under the: / right under / right in the in the building that got / hit by the d- / by the debris //
35This paper however focuses on pauses, categorized as follows:
Symbol
|
Signification
|
. (between two spaces)
|
silent pauses
|
silent micropause (between 0.12 and 0,19 second, included)
|
/
|
short silent pause (between 0.2 and 0,5 s, not included)
|
//
|
median silent pause (between 0,5 and 1 second, not included)
|
///
|
long silent pause and silences (over 1 second)
|
em
|
filled pauses
|
short filled pause
|
erm
|
long filled pause (visible plateau-shaped intonational contour)
|
36We determined the different thresholds by looking at consensual measures in the literature in such domains as conversation analysis, discourse functional approaches and psycholinguistic. 0,2 second for example is the audibility threshold within interaction, speech perception, and is usually associated with the mean length of pronunciation for one syllable (in English mostly). After Jefferson (1989), 1 second is widely accepted to be considered a silence in conversation. The intermediate threshold (//) was established to refine the categorization, corresponding to more than two beats (0,2s x2) but less than a silence (Danino 2014 for a complete presentation of this classification).
37If pauses are associated with speech fluency, related to conceptual fluency, that is, with the relative ease the speakers experience (or not), then there should be different “pause profiles” corresponding to the different sociodiscursive profiles described above.
38In order to assess this hypothesis, we calculated the ratios number of words / number of pauses for each category of speakers in order to get an estimate frequency of pauses within their speech. These ratios are really just quantitative and we do not claim that they bear any other significance.
- 3 Corpus: circa 43230 words. Reporters produce 38,5% of total words, witnesses only utter 7,6%.
|
Number of words / number of silent pauses
|
Number of words / number of all pauses
|
Journalists
|
10,79
|
7,80
|
Reporters
|
12,20
|
8,33
|
Witnesses
|
15,34
|
9,39
|
Experts
|
11,4
|
7,60
|
CORPUS3
|
11,75
|
8,22
|
39Officials were excluded from this computation for two reasons: audio quality in some of their interventions does not allow for such a count, and their interventions are interactionally varied. If Bush’s allocution is clearly monologic, the interview of New York’s mayor by M shares a lot of features with witness interview. They do not pattern as a type from the point of view of interaction or speech production (authority and accountability clearly sets them aside though). For the ratios that we did obtain, it is no surprise that reporters set a sense of average. More surprising was that witnesses tend to produce fewer pauses than the other categories. (Our ratios, as they were calculated, point to the number of words between two pauses; the higher the number, the more words produced in one interpausal segment, the fewer pauses in the overall speech). Pauses do correlate the categories of speakers in the sense 1) that they distinguish between professionals and non professionals (of public speech) and 2) because they point to differences between journalists and reporters. This difference, along with the numbers for witnesses, was unexpected. We expected, quite naively indeed, that witnesses, untrained and shocked, often interviewed onsite after they escaped, still panting, would produce more pauses. In the same line of ideas, journalists have the highest number of pauses, with experts being close second, even if they are trained public speakers with a sense of discursive mastery. To illustrate the discrepancy in speech fluency, we present the following examples, G11 being a witness recounting what he saw, and M, the main anchor, commenting on live areal images of the south end of Manhattan as he discovers them:
(16) G11: E:rm well // we got erm I heard on the radio that one of the towers was on fire and we went to a high point on building which is on the 25th floor and you had a clear view both world trade centers and the one tower erm smoking ?hard and erm there was another plane that was flying low and we just looked at it and before we know (it?) it kamakazed boom right into the other tower (inspires) and mass explosion windows flying it was just horrible erm still distraught xxx
(17) M: you can see in this em shot our viewers can em / helicopter shot coming across the harbor / erm / the statue of liberty /// (1,42 s) prominent in the foreground and // smoke // and devastation // and tragedy // in the background / a tragedy that continues to unfold // and one that still has many many / unanswered questions
40The best explanation of the results is the scripted/unscripted speech distinction. No category of speakers has notes. But reporters and witnesses recount from memory: they know what they want to say and even if language (and sentence) planning is just minimal it is sufficient to produce quite fluent turns. On the other hand, experts (interviewed on the spot by journalists on quite specific issues) and journalists constantly managing the broadcast on the fly exhibit a far less fluent speech due to an increased informational pressure. Put bluntly, they do not always anticipate the end of their sentences or turn when they start speaking. If information structuring can account for the increased cost of utterance design, processing could still be at stakes. Indeed, witnesses’ accounts follow a basic narrative pattern with a chronological ordering of events: processing is made easier for the cospeaker and viewer because these turns follow a well-known pattern or frame. Journalists’ interventions on the other hand cannot be anticipated so much in terms of contents or structure, however minimal. We listen to them precisely to learn something we do not already know. This would tend to suggest that pauses are linked to information structuring for the speaker and to information processing for the cospeaker.
41These considerations only begin to sketch the type of reflections and analyses that pauses (among other dysfluencies) can contribute to speech semantics. If they correlate quite nicely with types of speakers, pauses can also shed light on the use of a particular construction. If they factor in sentence production, then they should interact with information structure.
42In order to examine pauses as cognitive and semantic indicators in naturally occurring speech, we considered a specific construction namely the existential / presentative construction (there + (be)agr + NP ± complement). Here are some instances of this construction:
(18) em department has arrived in force on the scene / there’s a park police helicopter overhead / em every car that arrives at the gate /
(19) / em emergency sirens and rescue personnel / erm erm / fanning out across Washington / there is a gigantic black billowing cloud of smoke // ?that is - that is rising over the pentagon-
(20) erm e-e- there was a: // a convoy --I can’t think of a better word - a convoy of fire and police trucks / racing down the westside highway
(21) there has to be a much / greater degree of cooperation // between nations to deal with this ///
43These instances present a variety of cases: the verb can be in the present tense (18, 19), in the past tense (20), or introduced by a modal phrase (21); the NP can refer to a concrete (18, 19) or abstract entity (21). It can refer to singular entity or a collective one (20). The structure can be expanded by a locative complement (18), by a relative clause (19), by a participial clause (20) or by a prepositional group expressing metaphoric location (21). They have in common to express existence or presence (which is existence in a particular location) of an entity, static or dynamic. They are logically quite important in our corpus where they act as scene builders. We found 270 occurrences of this construction.
44From the point of view of information structure (after Lambrecht 1996), it can be conceived as introducing a new referent in discourse (in all examples given above, the NP is indefinite). It allows however for secondary predication since the NP can be expanded by a participial or a relative clause typically. The question incidental to prosody is one of semantic integration, or semantic segmentation depending on the prosodic pattern (after such work as Gillespie et al.’s [2012]).
45Preliminary analyses of the 270 occurrences led us to a typology of four cases as far as pauses are concerned: 1) there can be no pause, when the construction is produced as one interpausal segment (18); 2) there can be a pause after the NP, usually separating it from its expansion as in (19); 3) there can be a pause before the NP indicating lexical search (20), with this instance explicitly referring to the conscious activity of lexical selection; 4) there can be a pause, or several pauses within the NP and its expansion indicating that meaning construction processes are here more difficult.
46It is to be noted that we did not find any example of pauses within the introduction segment there + be. On the contrary, it seems to be highly integrated by prosody, the copula being contracted most of the time, reduced systematically. This tends to suggest that the construction itself is highly routinized (deserving the label "construction") whereas most of the cognitive and attentional resources (of both the speaker and the hearer then) are allotted to the NP and its possible extension. The planning process concerns the referential effort at the heart of this construction. The lack of an observed (in our corpus) unified prosodic pattern apart from chunking of the introduction segment tends to confirm that pauses in this context are due to local constraints of word retrieval, sentence production planning and possibly information structuring (such as emphatic stress where pauses are used as a contrast marker). If this account explains the contribution of prosody to information structuring, it points to its possible function in information processing. Indeed, the introduction segment is prosodically reduced and not salient: this would indicate to the cospeaker that it presents no particular processing stakes - thus directing the cospeaker’s attention and effort toward what is introduced. We can also consider that the introduction segment is purely functional and does not represent processing activity on the part of the speaker himself or herself. Indeed, in scene description, information processing goes with the perception of the situation. Perceptual information retrieval and processing also have a cost: the use of highly routinized and cost-effective linguistic constructions (after Goldberg 1995) could factor in explaining prosodic patterning.
47These elements call for further research in diversified corpora and for systematic analysis of constructions where similar phenomena could take place. If we willingly recognize the exploratory nature of this analysis, we nonetheless hope to have shown that prosody factors in the semantic set up of an utterance and that the study of speech dysfluencies could allow linguists (and not only psycholinguists) to study language production in corpus data (and not only in experimental settings).
48Spoken media exhibit the expected characteristics of any spoken linguistic productions in which prosody factors in meaning construction mechanisms. It also influences interaction management, which is itself a parameter of meaning construction. If the oral dimension of spoken media did receive the attention of phonetics and phonology in determining for example phonostyles and other professional speech practices, it is worth considering media corpora from the point of view of semantics. Semantics needs to deepen its understanding of speech. And because the media present highly coded formats and interactions, it is our contention that they offer a research material that is somewhat naturally controlled.
49We tried to describe the breaking news format, as a specific format of spoken media. The BNF can be understood as a technical and sociocultural set-up aimed at dealing with urgent situations. These situations, and among them typically the major ongoing event, put two different pressures on speech production: temporal pressure and informational pressure. They affect both the materiality of speech, such as prosody, rhythm, tempo, quality of voice, and the content of the spoken word. If the thematic content (rescue personnel actions, evacuations underway, description of scenes of panic, etc.) are obvious contributors to the format, it became apparent that modes of discourse offer a finer-grained picture of what breaking news is about: different speakers with different trainings, backgrounds and responsibilities collectively construct an emerging and changing representation of a given situation which constitutes the one constantly share attentional focus. Modes of discourse are worth transposing to the spoken word where they prove particularly efficient in analyzing the different types of information conveyed within the interaction.
50The sociocognitive approach is best exemplified by the study of pauses which allowed us to consider at once, with one type of markers 1) the cognitive dimension of language production, building on works from psycholinguistics; 2) the interactive and intersubjective dimension of speech segmentation and 3) the sociocultural importance of speech styles.
51It is our conviction that oral semantics is a worthwhile endeavor for researchers interested in language production whether from the point of view of orality, cognition or sociocultural interaction (Di Cristo 2002 on prosody and cognition, Missire (dir.) 2014 on approaches of oral semantics). In this endeavor, prosody proved to be an indicator relevant to the different dimensions of meaning making. It also appeared that prosody might have more to do with information processing than information structuring.