ATTRIBUTION RELATIONS IN RHETORICAL STRUCTURE ANNOTATION – DATA FROM JAPANESE FICTION

It may be said that an adequate representation of Attribution relations represents a noticeable stumbling block for annotation techniques. Annotation in itself is absolutely indispensable as a practical way of testing any text structure theory. However, a reference to the authorship of a saying, or a thought, working as an intrusion on a meta-level, poses a challenge to a clear-cut hierarchy of rhetorical relations. Fragments evoking Attribution tend to jut out from linearly developing texture of discourse units, creating loops and crossing ties. Moreover, interpositions of revive the age-old issue of the admissibility of discontinuous units in text – see the general move from = ‘no’ to a shy ‘yes’ there in

Below, I will introduce one concrete case of such troubleshooting in the example of a Japanese fiction text annotated. It is the same short piece of fiction that I already used in my paper

I use many of the lists of available rhetorical relations in theory basics and annotation manuals

Attribution, nevertheless, becomes more vital with some shift to viewing wider discourse events than predominantly newspaper texts that made the departure point for the PRT. It was supported for the case of discourse by

Another important contribution here is the division between types of Attributional relations, see

In the bulk of annotated text (692 elementary discourse units) Attribution relations are by far not the most frequent, though considerable stretches of text are polyphonic, including exchange of direct quotes. Only a limited number of options in markers are available, mostly the quotation construction. For a left-branching, loosely S-O-V/O-S-V structure of Japanese the canonical way of attributing quotes would be an optional Source postposition for the Cue, comprising, beside a speech verb, a grammatical quotation marker TO. If one counts the (marked) borders of attributed speech in dialogue (and not the, say, graphic limits of each utterance recorded dialogue, which would be times bigger numbers), we find the prevailing of Content preceding – in 27 cases, of which 14 with quote marker TO (indirect speech – 3 of them), with Content following at 14, of which 1 with TO.

TO becomes indispensable in cases of so-called indirect speech, roughly conveying Content, though in the case of Japanese not only keeping but more often than not highlighting its grammar. Apparently, in fiction indirect speech is preferred in instances of exchange that the author means to defocus, mainly with secondary participants (Content as indirect quote in 9-11): ⑥ 8 BOKUNIN WA ODOROKI, 9 SORE WA IKENAI, 10 KOCHIRA NI WA IMADA NAN NO SHITAKU MO DEKITE INAI, 11 BUDOO NO KISETSU MADE MATTE KURE, 12 TO KOTAETA. = 8 Shepherds were taken aback 12 and replied that, 9 it just wouldn’t do, 10 they weren’t prepared yet, 11 and whether he could wait till grape harvest. Indeed, ‘the Greek choir is treated in the same way once again: ㉑23 GUNSHUU WA, DOYOMEITA. 24 APPARE. 25 YURUSE, TO 26 KUCHIGUCHI-NI WAMEITA. = 23 The crowd came into motion. 24 Excellent! 25 Free him! – 26 shouted hundreds of mouths.

Graphically an intermediate option exists in Japanese, of quotes closer to direct (that is, retaining features of ‘real’ speech, such as, among other, addressive affixation) but not marked with quotation brackets. That is the case for Melos’s friend’s speech rendered by a messenger. Here we have indeed a rather advanced gradation for those slightly more or less worthy of (de)focusing:⑲ 35 OOSAMA GA, SANZAN ANO KATA O KARAKATTE MO, 36 MEROSU WA KIMASU, 37 TO DAKE KOTAE… = 35 The king continually ridicules him 37 but he only goes on replying – 36 Melos will return!.. Quite interestingly, here the friend is defocused not so much for his insignificance, but for the fact that his speech is in fact rendered by a third speaker. When he is given the chance to speak in the final par. 24, he is bestowed with regular quotation brackets alright.

There are, roughly half as frequent, entries with precedent Source/Cue, accompanied by little or no marking: ② 6 ROOJI WA, ATARI O HABAKARU HIKUGOE DE, 7 WAZUKA KOTAETA. 8 "OO SAMA WA, HITO O KOROSHIMASU" = 6The old man, in a low voice fearful of the surroundings, 7answered shortly. 8 ‘The king is killing folks.’

As in many languages, a cataphoric deixic item is possible, but absent here. I would deem it incompatible with the modifier WAZUKA, turning ②7 into a full-fledged discourse unit, as compared to a more technical remark-like KOO ITTA (cf. “…” – fit-il, typically inverted, in French). The lexical Cue KOTAETA is typical but not indispensable, even without graphic Cues such as inverted commas here, a certain breach of relevance relations would convince the hearer that the content of the following phrase belongs to a character’s (internal) speech. In fact, KOO ITTA does occur once. Other speech verbs marking a probability of a following quotation include AIGAN-SHITA – implored,

It remains an untrivial question whether to count mental activity in Attribution relations. Alongside passages centered on dialogue, this short piece of fiction also contains stretches of reflection, which divide into dual prose – the character’s thoughts and ‘objective’, ‘from-the-outside’ descriptions of what is taking place. This may fit less into some schemas proposed for dialogic discourse, but the general setting remains the same – these form two separate planes of narration, disserved by a manifest interface between, some amount of marking indispensable. Relevant proposals have been sounded before, such as dividing Attribution relations into Quotes and Thought Incorporation

A difference with attributed speech is a practically equal preceding Content at 15 (6 with TO) and following at 14, since the text is basically arranged in such a way that conjoining phrases tend to be somehow logically connected. That holds particularly for the cases of the unmarked ‘inner speech’, but not limited to it. A switch from mental monologue to reality, as was mentioned, is still in some cases marked with TO, and accompanied by mental (attitude) verbs, mostly of the more technical type - as OMOU = think, NEGAU = wish, KANGAERU = ponder. Preceding Cue verbs are somewhat less frequent but semantically richer – as KAKUGO SURU = be aware.

Mind-world direction of the same interface, on the other hand, mostly passes unmarked, except for the breach in tense consequence (mental reality always in the present, and includes hypothetic and other modal expressions) – see the return in 16-17 below: ⑧ 14 CHOTTO HITONEMURI SHITE, 15 SOREKARA SUGU-NI SHUPPATSU-SHIYOO, 16 TO KANGAETA. 17 SONO KORO NI WA, AME MO KOFURI NI NATTE IYOO. = 14 I could take some sleep 15 and then leave at once, 16 he thought. 17 By then the rain’s probably going to stop. Generally, a far richer grammatical realization of mental reality, uninhibited by temporal, modal, or deixical constraints in the course of the narration, is indeed in itself a clear sign of this switch: ⑫ 4 MEROSU NO ASHI WA, HATATO, TOMATTA. 5 MIYO, ZENPOO NO KAWA O. = 4 Melos stopped in his track. 5 Just take a look at this river! Modality includes evidentiality ⑯ 3 …IKI O NONDE MIMI WO SUMASHITA. 4 SUGU ASHIMOTO DE, MIZU GA NAGARETE IRU RASHII. = 3 He held his breath pricking his ears. 4 Just under [his] feet there seems to be running water. See for deixis, being the only marker of a switch to mental reality in 20-21, until some grammatical disorderliness, characteristic of spontaneous utterances, and final present tense take up some of the job – ⑪ 19 MEROSU WA HITAI NO ASE O KOBUSHI DE HARAI, 20 KOKO MADE KUREBA DAIJOOBU, 21 MOHAYA KOKYOO HE NO MIREN WA NAI = 19 Melos wiped his forehead with his fist – 20 coming to this point is already something, 21 there are no regrets left behind in the village. ⑬ 22 MIGOTO = wonderful! and 24 ARIGATAI = (I am) grateful as lone assessment markers intersperse author’s description of the hero’s actions, creating a two-dimensional narrational space.

Thirdly, on a certain microlevel of structural analysis, a much wider scope of perceptual expressions may be seen in this way, as not only what we see or think, but also what we see indeed bears a stamp of the seer’s individuality, see (3) in the list of Attribution relations. It is seen through the eyes of a hero that reality is depicted in prose of any psychological depth. A relevant Cue is, naturally, an expression of perception (Source), and a construction of consequence with a cognitive effect: ⑯ 5 YOROYORO OKIAGATTE, 6 MIRU TO, 7 IWA NO SAKEME KARA KONKONTO, 8 NANIKA CHIISAKU SASAYAKINAGARA, 9 SHIMIZU GA WAKIDETE IRU NO DE ARU. = 5 He rose unsteadily – 6 to look at it, [and from here on we look at the thing with his eyes, in a way] 7 from a fissure in the rocks there’s a fountain 8 quietly gurgling (only finally the perceptive ‘inlay’ is corroborated by the mirative NO). 

Now, breaking this relation down to units of discourse may be a challenging task: ⑲ 25 MEROSU WA 26 MUNE NO HATISAKERU OMOI DE, 27 AKAKU OOKII 25a YUUHI BAKARI O MITSUMETE ITA = 26 His heart breaking in his chest, 25 Melos kept his eyes on the sun, 27 red and huge. In this piece, it is not the grandeur of the scenery that the author is trying to impress us with, but its effect on Melos. ‘Red and huge’, in this contraption, becomes a quasi-quotation highlighting a complex emotional attitude of the hero, and enlivening the narration by the sheer sense of presence. It looks as if the whole subtlety of the effect, though, is balanced on the minimal amount of stressing it. Any more direct marking would surely destroy it – to an inquisitive annotator’s distress. 

Prose at its best volume, then, is achieved in masterly interweaving all the dimensions mentioned – descriptions, subjective perception, internal speech, and dialogue. In fact, straight after the fragment just quoted we can see a shift to Melos’s internal speech, unmarked but for grammar and the present tense, and then starts a dialogue, marked purely graphically, in quotation brackets! ⑲ 28 HASHIRU YORI HOKA WA NAI. 29 ‘HASHIRU NO O YAMETE KUDASAI… = 28 There’s no choice but to run. 29 ‘Please stop running… We couldn’t be unable to call such mixture of layers and levels a sort of multimedia writing.

All cases considered, a significant difficulty is orienting Attribution by the rhetoric weights of its components – a sine qua non for placing it in context both within a broader stretch of text and in the unified range of rhetorical relations. According to the Rhetoric Structure Theory, mononuclear relations comprise a more central and a more peripheral element. With Attributions, if one is led by the analogy with the Elaboration relation, then the attributor is the Nucleus, and the attributed content its Satellite. To tackle the problem from the point of view of the salience in discourse, this may be completely conformant to cases of the quotation illustrating an already clear type of reaction, such as ‘he thanked’ or ‘she was furious’. In this, Attribution relations may well almost completely merge with the Elaboration kind, providing a ready argument for their expulsion from the annotator’s list.

Another option how attribution constructions may be treated, is Generalization, the exact opposite of Elaboration. That is more appropriate in cases when remarks following lines of dialogue have a crucial input in the understanding of the scene, as is the case when a shift to a subjunctive mood in the Cue material actually transfers the whole utterance from spoken speech to the category of internal speech: ㉑ 6 TO OOGOE DE KEIJOO NO GUNSHUU NI MUKATTE SAKENDA TSUMORI DE ATTA GA 7 NODO GA TSUBURETE … = 6 all of that he was about to shout to the throng at the execution grounds, 7 but his throat became stuck… Now, as far as a more macro-level outlook is concerned, it would be logical to ascribe prevalence to that more salient fragment which is directly linked to the higher level of the text organization, that is, to the upper node in the discourse tree representing the speaker’s plan. The problem is that in fiction prose conceiving such a plan is not by a long chalk so easy as in more practically oriented texts such as an editorial or even a love letter.

Now, with Content following Source/Cue material and ending with the quotation marker TO, we seem to face an assured case of elaboration, that is, an N-S sequence, the preceding phrase providing most of the semantic impact. So much indeed, that far more precisely it should be annotated as a case of Equivalence (to be exact, Illustration, perhaps). It is necessary to note that such reversal, the parcellated bit ending in TO, never occurs with author’s remarks to direct speech, but inside speech of a character, depicting someone else’s speech: ⑤ 24 MURA NO HITOTACHI NI SHIRASETE KOI. 25 KEKKONSHIKI WA, ASU DA TO. = 24 Go tell the villagers. 25 The wedding’s tomorrow, that is.

Illustration would be the best way to describe a relation with preceding, and informationally dominant Source phrase even without the TO: ㉓ 12 DOTTO GUNSHUU NO AIDA NI, KANSEI GA OKOTTA. 13 BANZAI, OOSAMA BANZAI’. = 12 Suddenly, a cry broke out from among the crowds. 13 Glory, glory to the king! No quotation marks, again – one does not indeed record verbatim cries of the crowds, does one? Here, however, an element of superstructure is inmixed in the purely semantic sequence, ‘banzai’ being one of those phrases that are conventionally expected at the end of a text or a large period of it – as is here with the end of an episode (what follows is actually more of a coda than continuation of the story).

A contrary example is that when the Source phrase contains Cue information that can be classified as Level-of-detail. As typically this fragment precedes the Content, a certain likelihood is noticed with introductionary (always preceding) Elaborations as Commentaries (or Evaluations) – collapsible to evaluative adverbials (such as luckily, to my surprise etc): ⑱ 11 ICHIDAN NO TABIBITO TO SATTO SURECHIGATTA SHUNKAN, 12 FUKITSU-NA KAIWA O KOMINI NI HASANDA. 13 ‘IMAGORO WA, ANO OTOKO MO, HARITSUKE NI KAKATTE IRU YO’. 14 AA, SONO OTOKO, SONO OTOKO NO TAME NI 15 WATASHI WA, IMA KONNA-NI HASHITTE IRU NO DA. = 11 Passing a group of travellers, 12 he overheard a piece of ominous conversation. 13 ‘By now that fellow would also be up crucified’ 14 Ah! But it is for the sake of that fellow 15 that I have ran all this way! Representing 12 as a satellite introduction to nucleus in 13 could yield either Contrast, or Conclusion (Cognitive Result, also tractable as Reformulation) 14-15, while temporal connection in 11-12 would still hold as a subsidiary. It is only the temporal sequence of ‘passed-heard-realized’ that would grow invisible for the overall representation, thus weakening the narrative structure of the text. And this is not the only example when positing a hierarchical relation in Attribution undermines an intuitively sequential nature of the narration.

Also, to get down to the microlevel of in-utterance structure instead, it becomes hard to bypass the impression that the speech verb in Clue position is not solely – nor even mainly – a generalization of elaborated Content. Rather, it fulfills a twofold function. Firstly, it marks a turn in the sequence of remarks by the character, thus fulfilling the structural task of setting up Participation Structure of text

③ 9 “KONO TANTOO DE NANI O SURU TSUMORI DE ATTA KA. 10IE!” 11BOOKUN DEIONISU WA SHIZUKA-NI, 12 KEREDOMO IGAN O MOTTE TOITSUMETA. = 9 ‘What did you mean to do with that knife? 10 Do tell!’ 11Tyrant Dionysos inquired quietly 12 but threateningly. // 11Tyrant Dionysos inquired 12quietly 13 but threateningly.

What kind of Level-of-detail information can be added in this way? Let us look at some examples - ③ 17 TO MEROSU WA WAZUBIREZUNI KOTAETA = Melos answered fearlessly; 23TO MEROSU WA IKINARI TATTE HAMPATSU SHITA = Melos suddenly rose up in protest, 41 OO WA SATTO KAO O AGETE HOOJITA = the king notified swiftly raising his face, ⑫ 6 HASHIRINAGARA = while running,㉑ 20 KASURETA KOE DE SEIIPPAI-NI = at the top of his lungs, and so on. None of these additions do not open new vistas in the narrative, remaining, however useful, secondary cul-de-sacs, and therefore representable as Satellites. But then, another possibility represents a juncture to a more subtle or more decisive shift in setting, where the comment opens up a hub to a new dimension in the exchange, as in 53 TO IIKAKETE 54 MEROSU WA ASHIMOTO NI SHISEN O OTOSHI SHUNJI TAMERAI = 53 speaking so, 54 Melos dropped his gaze and hesitated for a moment. How would that be representable enlarging on a Satellite?

Furthermore, it is not impossible to imagine cases where the elaborated nucleus would be semantically quite detached from the Content of the quotation – the more so, the less of a speech verb in a strict sense it would be. Thus, a verb signifying psychological reaction would yield an ambiguous interpretation – either as a generalizer for the elaborated Content in 31-32 below, or else a Cause for a non-verbalized speech verb which only, in turn, would comprise the quote as its elaboration: ②30 KIITE, MEROSU WA GEKIDO-SHITA [SOSHITE KOO ITTA]. 31 “AKIRETA OODA. 32 IKASHITE OKENU” = 30On hearing that, Melos was furious [and so he said]. 31 ‘Miserable king! 32 He does not deserve to live!’

Generally, in cases of this type, one could claim verbs of psychological states and reactions can be construed as metonym (enlarging upon verbal realization of such state). This stance is corroborated by further examples when non-verbal reactions, described in author’s remarks, fill slots in the in-dialogue relations of Eliciting Response-Response, such as Melos’s sister’s sole reaction to his speech in ⑤20 IMOOTO WA KAO O AKARAMETA = His sister blushed. In ⑧40 again she merely nods silently after a long stretch of Melos’ speech (22-39). Obviously, in that environment, girls weren’t allowed much more as ways of answering. In a way, they are revenged by an ironical episode when Melos himself in the very final lines of the short novel made to turn purple in response to his nakedness being pointed out (action-against-speech pattern again, in the Exchange structure).

Just to get away from the elaboration/generalization dichotomy, I can report cases where attributing fragments are actually inserted between fragments of the Content, notably by the self and same speaker. In this case more clearly, but technically in the equal sense as well, beside relation between attributed and attributing matter, relations within the Content stand to mind (below – between 18 and 20-21), - just as they do in dialogue in complete accordance with Shiffrin’s Exchange structure: ③ 18 ‘OMAE GA KA?’ 19 OO WA, BINSHOO-SHITA. 20 ‘SHIKATA NO NAI YATSU JA. 21OMAE NI WA, WASHI NO KODOKU GA WAKARANU.’ = 18 ‘You?!’ 19 The king grinned in disdain. 20 ‘What tedium. 21 Have you any idea of my solitude?’ 

In this, Exchange structure – relations counting as adjacency pairs, such as Question-Answer, or Eliciting Response and Response, can hold over even longer spans of commenting material (Source/Cue), e.g. 5 units, which is not at all a limit for a strongly anticipated continuation. With other less acutely tense pairs the situation may be different, calling for more supports or more frequent interventions – as with Cause, or Reformulating.  

Exchange structure, needless to say, holds even before non-verbal reactions, which in turn need to be construed from this perspective, and not otherwise, as in 103-104 here: ③ 100 ‘HAHA. 101 INOCHI GA DAIJI DATTARA, 102 OKURETE KOI. 103 OMAE NO KOKORO WA, WAKATTE IRU ZO’. 104 MEROSU WA KUCHIOSHIKU, JIDANDA FUNDA. 105 MONO O IITAKU NAKATTA. = 100 Ha-ha. 101 If life’s dear to you, 102 come late. 103 I can see through you.’ 104 Melos stood his ground mute.  105 He had no wish to speak. No sooner than with 105 (offered here as Justification for 104) it becomes possibly to posit a Cause relation with a messing Generalization of what the king was saying in 100-103, connecting the circuit. 

Seen in this light, attribution relations and their markers, including speech verbs, can be seen as part and parcel of corresponding taxonomy pertaining to rhetoric structure on the whole. Whatever the ultimate balance of their representation inside the taxonomy, whether or not there can be found a unified account of them positioned often contrarily in the tree structure – they do indeed contribute differently by each element to the communicational exchange, step-by-step-wise, and therefore are eligible for annotating and reviewing among other rhetorical relations. On the other hand, they seem, after all, to constitute a slightly separate plane, and there are limits to how they can reasonably be made consistent in terms of representing attribution relations linearly, or, for that matter, in a one-way branching tree.

Actually, annotating this text led me to picture its entire continuity as divisible into several layers. Continuous in themselves, they do not need to be taken up in the reader’s field of vision in every single sentence. However, as in the mind of the characters, they do stretch unbroken. Say, for the entire exchange in the 3rd paragraph, the outline is as follows:

And so on. It can be seen as two simultaneously developing interfaces, shifts between which can, but do not necessarily need to be marked. The need for such marking arises out of the necessity to convey a nuance, a change in the ambience of the exchange. Note that the author may intervene by far not only to mark the changing turn, but also inside continuing elocution by one speaker. In any event, from the layout schematically ordered here, it stands out vividly that orientation between layers and dimensions of narrative may well be crucial for significant stretches of rhetoric structure in dialogue-rich fiction.

Break-offs between dialogue content and continuation of narration, such as in the end of Melos’s talk with his sister in par. 5, are not marked in any way, but this is indeed a sort of connection too – as a default interface between planes of narration and dialogue. No Attribution is present here as such, nor needs to be, at all times. All in all, the proportion between author’s intervention (pushing forward the narrative drive) and dialogue (following its own logic) vary greatly according to the type of situation described. Thus, in the opening of par. 19 dialogue unfolds in an unforeseen setting that demands much more comment (Cue) than a ‘stable’ exchange between, say, Melos and king previously mentioned. To keep every and each of such episodes on a double ‘payroll’ of dialogue-internal dialogue-external relations would hardly meet the need to differentiate between the types, wouldn’t it? Besides, even this relatively high level of informativity in Source and Cue material can in fact be kept down or omitted, thus confirming the prevalence of Content in a stretched-out Attribution interface. For instance, no background information would be truly necessary in a radio show, where the hearers would be guided uniquely by the internal rhetoric relations in dialogue (underlined – Greeting-Question-Reply):

1 ‘AA, MEROSU SAMA’ 2 UMEKU YOO-NA KOE GA 3 KAZE TO TOMO NI 2A KIKOETA. 4 ‘DARE DA’. 5 MEROSU WA 6 HASHIRINAGARA 5A TAZUNETA. 7 ‘FUIROSUTORATOSU DE GOZAIMASU… = 1 ‘Hey Melos’ – 2 a groaning voice came 3 carried by the wind. 4 ‘Who’s there?’ 5 Melos asked 6still running. 7 ‘I am Philostratos…

A competition for salience in the minds of the communicators, thus, is an important feature of this double-, triple-decked structure that I seem to fathom here. At times, it appears, the whole succession of Content phrases can ‘go under’ and Source/Cue ‘come above’, contrary to the example just shown. A prerequisite condition for this would be extensive ‘abstract’ description of the scenario in the speech verbs. A more natural option, though, would be the Content driving the narration forward, and remaining central to its development, garnished by author’s remarks ‘strung’ to each successive element in the keynote structure of the dialogue.

In this tentative study I looked at a specific genre of text, which has so far been getting limited attention from the point of view of rhetorical structure study, at least in language of a structure similar to Japanese. It was all the more obvious how the issue of attributing significant fragments of text is difficult to solve along the lines of canonical methods in rhetorical structure annotation. Indeed, clearly establishing Attribution relations plays a vital role in various types of discourse, including not only narrative (as was our object), but also expositional and even argumentative, with an obvious exemption for a more flat-structured descriptive

As long as human beings carry mental reality in themselves, there is supposed to always be such biplaneity in this or that proportion in any depiction and report of the world in speech – the external and internal, with spoken speech serving only as a medium. Various means of expressing that interface should all be taken in consideration and re-ordered according to their contribution and value. For instance, Speech verbs in themselves, at par with grammatical and graphical markers act as connectives marking attribution relations, among other things conveying procedural meaning as to whose turn in a conversation is recorded. But that already constitutes an area of further research.