Cotextual Cues for the Annotation of Perspective in Livy’s Ab Urbe condita

Ðàçìåùåíî íà http://allbest.ru

Cotextual Cues for the Annotation of Perspective in Livy's Ab Urbe condita

Simon Aerts

Ghent University (UGent), Department of Linguistics,

Blandijnberg, Gent, Belgium

The current paper examines the statistical correlation between a number of cotextual cues and the latent variable of perspective in a corpus of episodes taken from Livy's narrative. The possible perspectives (external, internal, scenic camera-eye, immersive eyewitness and distanced eyewitness; cf. “Localization” in narratological studies) featured in Latin historiographical narratives are introduced and hypothesized to co-occur with a number of lexical (e.g. evaluative, descriptive), deictic (e.g. proximal, distal) and grammatical (e.g. voice) elements in the cotext of events and situations occurring in the main narrative tenses (especially the perfect, imperfect, and present indicative). During the analysis, statistical tests are used to ascertain the combined impact of these cues on (a) Livy's choice of verb tense and (b) the linguists annotation of perspective. While the former is objectively observable, the latter relies on the functional linguist's interpretation as a non-native speaker. Therefore, an additional test is used to show the correlations between the different narrative tenses and the perspectives to corroborate the results. This paper's findings pave the way for a more systematic and objective annotation method for the latent variable of perspective in Latin narratives. Its argumentation is built on the statistical tests performed on the data gathered in close readings on a larger scale. From those close readings, some examples are provided in the current paper to illustrate specific or challenging concepts.

Keywords: Latin tense system, tense and aspect, Systemic Functional Linguistics, perspective, text linguistics, corpus linguistics, multivariate statistics, linguistic proxies, Conditional Inference Tree, Livius.

1. Introduction: interpreting interpersonal meaning in Latin narratives

When investigating the Latin narrative tenses from a theoretical approach based in Systemic Functional Linguistics (SFL) See e.g. Halliday & Matthiessen (2014), Thompson (2014), and Bache (2008). and its three metafunctions (cf. Aerts 2018), In other papers (Aerts 2018, section 3; Aerts forthcoming, section 1), I proposed that in Latin, both tense and aspect were grammatical categories expressed by the morphology of the verb tenses, both with a semantic potential on three levels of meaning. Grammatical aspect revolves around the representation of states of affairs (SoAs) as (un)terminated on the ideational level, around their presentation with global view (as an indivisible whole) or partial view (as a background frame) on the textual level, and around a viewpoint (see below, section 2). Tense concerns the temporal structure of reality (i. e. establishing the link with the moment of speaking/writing (S) as ultimate ground) on the ideational level, the status of independent foreground event or secondary background event on the textual level, and a vantage point, from which the story-world is narrated, e.g. in the speech moment (looking back with hindsight) or in the story-world (as a virtual eyewitness). especially innovative but at the same time tricky is the interpretation of the interpersonal meaning of “perspective” (P) as conveyed by the language used by the author. The grammatical categories “tense” and “aspect” may both be considered to have an interpersonal dimension of meaning, i.e. the expression of a vantage point (e.g. looking back from the moment of speech/writing In this paper, the terms “speech moment” and “moment of speech/writing” (S) are always used to refer to the moment of communication, which is considered to include both the author's act of writing and their audience's act of listening/reading.) and the expression of a viewpoint (from without or from within), respectively (cf. Aerts 2018, 124-125; Aerts forthcoming, section 1). However, for Latin we cannot rely on judgements about the correlation between tense/aspect and the sense of perspective one may experience; nor can we rely on tense/aspect alone for our annotation of P, because the aforementioned correlation is exactly what we are trying to assess. Therefore, we must scan the author's language for “perspective cues” in the surrounding cotext. For this purpose, I have scrutinized eight narrative episodes from Livy's Ab urbe condita For our research, the relevant OCT-editions (Ogilvie 1974; Conway & Walters 1919; Briscoe 2016; Conway & Johnson 1935) were consulted. Of the 808 observations in our corpus, only 35 (4,33 %) occur either in potentially relevant alternative readings in the manuscripts, or in a cotext which contains elements with a potentially relevant alternative reading. Of these 35 data points, only 2 appear in examples cited in this paper; notes 46 and 49 explain our preferences as regards these examples. Nevertheless, all 35 data points that could potentially influence the statistics, have been left out of the calculations. in an attempt to find a statistically significant correlation between certain (combinations of) cotextual cues and the annotation of P The texts in that case study are “The rape of the Sabine virgins” (1. 9-13; N The symbol “N” refers to “sample size”, i. e. the number of observations within a certain corpus, subcorpus or sample. = 158), “Horatius Cocles” (2. 10; N = 39), “Defeat of the Hernici” (7. 7-8; N = 59), “The Battle at Sentinum” (10. 27-29; N = 128), “The Battle at Luceria” (10. 35-36; N = 104), “The Battle at Cannae” (22. 44-50; N = 147), “Hostile camps in North-Africa” (30. 3-6; N = 110), and “Roman fleet at Utica” (30. 9-10; N = 64), amounting to about 7.200 words and 489 observations of the indicative of the relevant narrative tenses, out of a total of 808 finite Some observations are morphologically not actually finite, e.g. the historic infinitives and participles used as such. verb forms annotated for P.

2. The interpersonal meaning of “perspective” in Latin narratives

The morphology of the Latin verb tenses and the combination of aspecto-temporal interpersonal meanings allow for five different perspectives potentially taken on the narrated events (cf. Aerts forthcoming).

On a preliminary note, the expression of interpersonal meaning in narratives should be conceived of as optional: if an author wants to focus on ideational (termination) and/ or textual meaning (foreground vs. background For the notions of foreground and background, see e.g. Fleischman (1990, Chapter 6, Part I).), we could say that the perspective (P) is “neutral”.

The actual marked Ps are, in terms of effect on the audience, quite comparable to the narratological idea of “focalization”, although there are important differences in terms of methodology and theoretical concepts. First, events narrated from an “external perspective” (P1) are not expected to have a high impact on the psyche of some conscience nor to appeal greatly to the senses, but rather to occur with observations, evaluation, commentary, comparisons, etc. which originate from a story-external source of knowledge. These characteristics are very similar to those of Adema's (2019, 28-30) report mode. The tense expected to occur most often with P1 is the perfect tense (pf. ind.), combining an aspectual interpersonal meaning of “viewpoint from without” with a temporal interpersonal meaning of “vantage point in the speech moment” (E < S The notation of temporal relations, in casu “event time before speech time”, draws on Reichenbach (1947).).

Second, the narrator may want to opt for the “internal perspective” (P2), which invites the reader/listener to experience the events from the point of view of a story-internal character. The language thus appeals to the perceptive senses as well as the emotions and thoughts of a character (or group of characters), who need to be explicitly “given” For the notion of “givenness”, see e.g. Gundel (2003). The literal sense is “event time simultaneous to reference point”. The speech time S is deliberately left out of the formula, as the speech moment plays no part in the interpersonal temporal meaning here. for us to be able to identify with them. The temporal interpersonal meaning is an “identification with a story-internal character” (E = R¹¹), combined with an aspectual “viewpoint from within”. These are the values associated with the imperfect indicative (impf. ind.).

Third, the impf. ind. may also be used when there is no character “given” to identify with. Its temporal interpersonal meaning is then “E = R < S”, retaining the link with the speech moment. The idea is that the narrator (“R”) is displaced to the story-world in the past (the part “R < S” in the formula), from where he transmits the events he is witnessing (the part “E = R” in the formula) to the audience in the speech moment “S.” Inspired by de Jong and Nunlist (2004, 64-65), I have called this the “scenic camera-eye perspective” (P3). In particular, the language is expected to appeal to the senses as well as the psyche of a more neutral but internal observer, and to contain comments only insofar as they can be made without story-external knowledge.

Fourth, in the “immersive eyewitness perspective” (P4), the present indicative (pr. ind.) combines the temporal interpersonal meaning of “temporal illusion of proximity” (E = S) with an aspectual interpersonal meaning of “viewpoint from within”. The idea is that the audience is “immersed” For the concept of “immersion”, see e.g. Allan, de Jong, and de Jonge (2017, 34-37). into the story-world while the narrator shows them the events without any mediation. He does not provide commentary or evaluations; the audience is left to draw their own conclusions, often by the mere shock of experiencing them themselves, i.e. virtually.

Fifth, and finally, the pr. ind. -- which is considered neutral in terms of aspect See Haverling (2010, 474). -- combines its only temporal interpersonal meaning with an aspectual interpersonal meaning of “viewpoint from without” in the “distanced eyewitness perspective” (P5). The audience and narrator are again witnessing the events, but they remain at the sideline (or more panoramically, on top of a hill, cf. de Jong and Nunlist 2004, 64-65), commenting on the events.

3. Cotextual cues as perspective proxies Examples referred to within this section, can be found cited in section 4 below.

Lexical cues

Tucker (1998, Chapter 7.3) discusses a number of adjective categories -- within an SFL framework -- that are relevant for the interpretation of P. “Comparative lexis” (COMP) offers a comparison between the events of the current story-world scene and events happening at a different location or time. “Evaluative lexis” (EVA) offers a certain evaluation of the events of the story-world on the part of a certain “evaluator”, which can be either story-internal or story-external. In addition, non-evaluative information that could not have been known to a story-internal evaluator is also categorized as external EVA. “Descriptive lexis” (DESC), i. e. words appealing to the perceptive senses of a potential observer, stimulates a mental visualization of the narrated events on the part of the audience. “Personal lexis” (PERS), i. e. words appealing to the emotions, mind or experience of an observer, stimulates a mental or emotional experience of the events.

The relevance of these lexis types for the interpretation of perspective is confirmed in literature on cognitive poetics (e.g. Stockwell 2002, Chapter 4). COMP and external EVA are expected to occur with a viewpoint from without, which is the aspectual interpersonal value contributing to P1 and P5 (cf. Stockwell 2002, 54). Because these Ps are expressed, by default, by the perfect and pr. ind. respectively, COMP and external EVA are also expected to occur especially with these two tenses. Accordingly, P2, P3 and P4 are expected to occur with a complete lack of external EVA and COMP, because of the viewpoint from within. An internal evaluation, however, is more likely to accompany P2 and P3 (Stock- well 2002, 44) rather than P4: in the latter, the cognitively immersed audience is on their own in forming an opinion on the narrated events (Aerts forthcoming, section 2). DESC is expected to occur with all perspectives except P1, and therefore especially with the impf. ind. (because of the viewpoint from within) and pr. ind. (because of the temporal illusion of proximity). PERS is expected to occur especially with P2 and P3 and the correlated impf. ind. (viewpoint from within, cf. Stockwell 2002, 53).

Halliday and Matthiessen (2014, Chapter 5) and Thompson (2014, Chapter 5.2) provide a categorization of verb types within SFL. Material processes (MAT) are verbs of “doing” or “happening”, mental processes (MEN) are verbs of “thinking”, “feeling”, “desiring”, and “perceiving”, verbal processes (VER) are verbs of “saying”, relational processes (REL) are verbs of “being” and “having”, behavioural processes (BEH) are verbs of “(physiologically or psychological) behavior”, and existential processes (EXI) are verbs of “existence” (see notes 35 to 40). Material and verbal processes are “perceptible” par excellence and are expected to be correlated to P4 and P5 and therefore to the pr. ind. (cf. Allan, de Jong & de Jonge 2017, 41). Mental processes are usually quite internal, so they are predicted to occur especially with P2 and P3 and therefore in the impf. ind. (Stockwell 2002, 53).

Another lexical phenomenon is the use of temporal adverbials (e.g. temporal location, duration) in the cotext. When events are narrated from a viewpoint from within, mediation of the narrative with temporal adverbials is unlikely (cf. Fleischman 1990, 124). The presence of these is therefore expected to be negatively correlated to P2 and P3 (and therefore the impf. ind.) and with P4.

Deictic cues

Concerning deixis, Fillmore (1997) and Anderson and Keenan (1985) discuss a range of (English) words which express a relation to the speech situation or “deictic center”: proximal indicators (PROX), i. e. words such as now, here, tomorrow and to come, are usually in opposition with distal indicators (DIST) such as then, there, the next day and to go (see also McIntyre 2006, Chapter 4.4). Whereas DIST, used in a narrative set in a certain story-world, assert the distance between that story-world scene and the world of speech/ writing, PROX suggest an assimilation between those two worlds. As Anderson & Keenan put it: “[N]otions such as “near to the speaker” may be interpreted not only in the literal, physical sense, but also by extension to “psychological proximity”, i. e., vividness to the mind of the speaker” (1985, 278).

Starting from the cognitive concepts materialized in the English words mentioned in the literature cited above, I have been looking out for similar words and constructions in the Latin texts during my close reading. Some of the more frequently occurring PROX were iam (in the sense of “now” (see example (14), section 4.2), as opposed to the more text-structuring meaning “already” (see example (7), section 4.1), dexter/laeuus, hinc (see example (6), section 4.1), (con)uenire (see example (14)), andproximus (see example (14)). These are expected to be a sign of either a temporal illusion of proximity or an identification with a story-internal character (i. e. P4 and P5, and P2; cf. Stockwell 2002, 53-54). Therefore, they are expected to occur especially with states of affairs (SoAs) in the pr. ind. and impf. ind. Frequently attested DIST werepostero die “the next day” (see examples (7)), tum (in its temporal-locative sense “at that moment”, as opposed to its text-structuring sense “next, and then” as in example (6)), ibi “there”, and eo “thither”. These are expected to occur with a vantage point in the speech moment (i. e. P1), and therefore with the pf. ind. (cf. Stockwell 2002, 53-54). A special case is P3, which is hypothesized to potentially occur with both PROX and DIST because of its temporal interpersonal meaning (E = R < S): the “displaced narrator R” element could evoke PROX, whereas the “audience in the speech moment S” element would allow for DIST as well.

A final deictic cue is the use of verb tenses to link story-time events explicitly to the speech moment, especially by means of a pf. sub). or an actual or generic use of the pr. ind. Such deictic tense usage may indicate that the story-world SoAs in their textual vicinity are to be viewed from a vantage point in the speech moment, i. e. from an external perspective (P1) (cf. Stockwell 2002, 55).

Grammatical cues

With regard to grammatical phenomena disclosing a certain value of P, especially the combination of the passive voice with subject animacy seems relevant. As Pinkster (1985, 116-118) already noted, the use of the passive with a (grammatical) Subject whose perspective may be interesting to view the events from, can serve to continue that particular perspective; it can be inferred, then, that the use of the passive with a less interesting Subject may lead to discontinuity of such a P2. Importantly, however, as Pinkster also concludes, the author is free to apply this idea: such an association may be simply a tendency.

Aims and purposes of this paper

As previously mentioned, our aim is to ascertain whether the correlations, hypothesized above, between the mentioned cotextual cues and the annotation of perspective are statistically significant. It should be explicitly noted, however, that the annotation of perspective during our close reading of the corpus was already performed on the basis of the available linguistic material in the clause and its immediate, relevant cotext. Therefore, the main aim of this paper is not to provide convincing proof of the worth of the individual cotextual cues for the annotation of interpersonal meanings: the effects of these cognitive concepts on human readership have been discussed in the cognitive literature cited above. Within a cognitive-functional paradigm, the effects of these cues are not language-specific: the cognitive concepts expressed by the English words in the examples cited in the secondary literature, as well as their cognitive effects on the readership, are considered to be -- at least to a considerable extent -- transferable to many other languages. Rather, this paper attempts to provide a descriptive, quantitative analysis -- supported by statistical tests -- of the impact of such cues on the annotation of P, and of the conflicts that may (or may not) arise among them.

4. Case study: methodology and results

The eight narrative episodes which constitute the corpus for the current case study (see section 1), were read closely with the three-dimensional meaning potential of the verb tenses and the theoretical observations on cotextual cues in mind. The total number of SoAs annotated positively for P (i. e. the corpus with the name data) is 760, of which 511 tokens are indicative forms. These numbers include the exclusion of 35 data points for which there were potentially relevant alternative readings in the manuscripts (see note 4).

The pf. ind. (“A”, N = 177), impf. ind. (“B”, N = 120), and pr. ind. (“C”, N = 143) constitute the outcome (or “dependent”) variables in the first part of the analysis (section 4.1). In addition, the indicatives in the corpus data include 55 pluperfect tense forms -- including ictus fuerat (Liv. 22. 49. 1) as an early attestation of the innovation in the passive analytic forms of the perfectum stem -- and 16 tense forms that could -- morphologically -- be both pr. ind. and pf. ind. (“CA”). The semantics of the pluperfect tense is not expected to be correlated independently to perspective and the cotextual cues that are investigated in this paper. The “CA” forms are excluded from the tests in the first part of our analysis as well, because these would influence the results in an undesired way (“noise”).

Other than indicatives, data includes (i) 158 subjunctives, (ii) 67 past participles and 22 (“historic”) present infinitives used as finite verb forms, and (iii) 2 forms that are morphologically ambiguous between the (“shortened”) pf. ind. and the present infinitive: avertere (Liv. 10. 28. 8) and ostendere (Liv. 10. 36. 10).

The gathered data was subjected to a number of statistical tests. For multivariate impact analyses, the model of the Conditional Inference Tree (CIT) was used; for monovariate impact analyses, the X² (Chi-squared) Test of Independence was considered more appropriate. The software used for the statistical analysis is R Studio²⁰ with the additional packages rattle, R Core Team. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria, 2018. URL https://www.R-project.org/. Williams, G. J. Data Mining with Rattle and R: The Art of Excavating Data for Knowledge Discovery. New York, Springer, 2011. Hothorn, T. and Zeileis, A. Partykit: A Modular Toolkit for Recursive Partytioning in R. Journal of Machine Learning Research, 16, 3905-3909, 2015. partykit²² and corrplot. Wei, T. and Simko, V R package “corrplot”: Visualization of a Correlation Matrix (Version 0.84). 2017. Available from https://github.com/taiyun/corrplot. In the examples quoted in this paper, forms in the pf. ind. will be in boldface, forms in the impf. ind. in small caps, and forms in the pr. ind. underlined. Forms that could be both pr. ind. and pf. ind. are

bold and underlined. Peruentum est “they reached” is an impersonal passive without a grammatical Subject and has no equivalent translation in English. In my view, this construction could be used to superimpose inanimacy on the SoA if an animate logical subject (agmen “the column”) was not meant to be in focus. The translations of the examples quoted in this paper draw heavily on the Loeb editions (Foster 1919; Foster 1924; Foster 1929; Moore 2002); they were adapted whenever necessary to support the argumentation.

The correlations between cotextual cues and verb tense selection

The first part of the analysis consisted of evaluating the hypothesized correlations between the observable variables of cotextual cues described in Section 3, and the observable variable of verb tense. For this part, a first subcorpus was made (datal, N = 440) which contains only the observations with the verb in the pf. ind. (A), impf. ind. (B) and pr. ind. (C).

First, the combined (i. e. multivariate) impact of “voice” and “subject animacy” on verb tense selection was analyzed (see section 3.3). Example (1) illustrates why this correlation is hypothesized to be statistically significant:

(1) Ita ut imperauerat signa sub occasum solis efferri sunt coepta; ad primam ferme uigiliam ag- men explicauerunt; media nocte--septem enim milia itineriserant-- modico gradu ad castra hostiumperuentum est. (Liv. 30. 5. 3)²⁴

“In accordance with the orders he had given the standards began to be set in motion just before sunset; at about the first watch they deployed the column. At midnight -- for it was a march of seven miles -- proceeding at a moderate speed they reached²⁵ the enemy's camp.”²⁶

During the close reading of our corpus, the use of the morphologically passive form with an inanimate grammatical subject (e.g. sunt coepta “began” and peruentum est “they reached”; see note 25) seemed to occur especially in the pf. ind. (A). The Conditional Inference Tree (CIT) investigating the statistical significance of the combined impact of “voice” and “subject animacy” is represented in Figure 1.

The CIT shows that 84 of the 440 perfect (A), imperfect (B) and present (C) indicatives in the corpus For each test in section 4.1, some data points in datal may have been excluded simply because they were annotated with a value “NA” (i. e. no answer) and/or “NR” (i. e. not relevant) for the input (i. e. “independent”) variables in the test in question, in this case “voice” and “subject animacy”. The resulting datasets with relevant exclusions (if any) are provided with an additional suffix related to the number of the test and Figure in this paper (e.g. data1_1). only have an active form and occur most often in the impf. ind. (node 5; e.g. erat “he was”). Since most of these tokens are forms of esse, this should not surprise us: the impf. ind. of esse is often used simply (and primarily) to depict the background. The label “active only” was used for the voice of verbs which have no passive forms (e.g. esse “to be”). These “animate” subjects also include nouns which refer to more abstract collectives of animate beings (e.g. legatio “embassy”). For the concept of metafunctional hierarchy in narratives, see Aerts (forthcoming, section 3).

Fig. 1. The combined impact of voice (“active”, “active_only”,²⁸ “passive”, “deponent”) and subject animacy (”yes”,²⁹ “no”) on verb tense selection.

In addition, verbs which do have a passive conjugation (N = 356) occur far more often with the pf. ind. (N = 48) than with the impf. ind. (N = 24) or the pr. ind. (N = 13) if the subject is inanimate (N = 85 For reasons of incomplexity, this number contains data points in the impersonal passive; the reasoning behind such an aggregation is that the effect of both constructions on tense selection and perspective is hypothesized to be the same (see note 25).); if the subject is animate (N = 271), however, the pr. ind. gains considerable ground (N = 111) and even trumps the pf. ind. (N = 106), with the impf. ind. being used for 54 out of 271 observations. However, there is no clear indication in the CIT about the role of voice in all this.

Second, the correlation between the variable “process type” and verb tense selection was analyzed. Examples (2) and (3) illustrate the two hypotheses concerning process types put forward in section 3.1, i. e. a positive correlation between MEN and the impf. ind. (B), and positive correlations between MAT and VER on the one hand, and the pr. ind. (C) on the other hand.

(2) Nusquam benigne legatio audita est: adeo simul SPERNEBANT, simul tantam in medio crescen- tem molem sibi ac posteris suis metuebant. (Liv. 1. 9. 5)

''Nowhere was the embassy heard willingly. In fact men spurned, at the same time that they feared, both for themselves and their descendants, that great power which was then growing up in their midst.”

(3) Quingenti ferme Numidae, (...) repente ex equis desiliunt, parmisque et iaculis ante pedes hostium proiectis in mediam aciem accepti ductique ad ultimos considere ab tergo iubentur. Ac dum proe- lium ab omni parte conseritur, quieti manserunt; postquam omnium animos oculosque occupau- erat certamen, (...) auersam adoriunturRomanam aciem, tergaque ferientes acpoplites caedentes stragem ingentem ac maiorem aliquanto pauorem ac tumultum fecerunt. (Liv. 22. 48. 2-4)

''About five hundred Numidians, (...) suddenly jumped down from their horses; having thrown down their bucklers and javelins at the feet of their enemies, and having been accepted in the midst of the battle line and led to the rear, they were ordered to sit down at the Romans' backs. And as long as the battle was initiated at every point, they kept still; after the strife had taken hold of everyone's minds and eyes, they then (...) attacked the Roman battle line from behind, and striking at their backs and hamstrings, effected a great slaughter and a terror and confusion that were even greater”

For this (monovariate) impact test, an X² Test of Independence was performed (N = 414; X² = 53.92, df= NA, p < 0.00 1 For a X² Test of Independence, a p-value of less than 0.05 signifies that the results of the test -- in this case that the two variables are dependent within our corpus -- are statistically significant. In a CIT, only the nodes that are in the same way statistically significant are rendered as output by the software.). For this particular X² test, the required assumptions were not met. The reported p-value is the outcome of an adapted X² test with a simulation of the p-value (for further reading, see Hope 1968, 583 ff.). The correlation plot in Figure 2 represents the standardized residuals of these tests. The correlation plots in this paper are set to display the exact standardized residuals only when they are significant, i. e. lower than -2 for a negative correlation (black), and higher than +2 for a positive correlation (grey); however, if they are not significantly low or high, the size of the circle may still help the reader to grasp the strength of the correlation.

Fig. 2. The impact of process type ('material”,³⁵ “mental”,³⁶ “relational”,³⁷ “existential”⁴⁰) on verb tense selection

Based on the sample datal_2 (N = 414), the pf. ind. (“A”, N = 172) seems to be significantly positively correlated to MAT: 108 out of 231 MAT are found in the pf. ind. A significantly negative correlation is the one with REL. The impf. ind. (“B”, N = 118), then, appears to be significantly linked to REL and BEH: 51 out of 101 REL and 5 out of 8 BEH occur in the impf. ind. A strong negative correlation, however, seems to exist with MAT. Finally, the pr. ind. (“C”, N = 124) appears to be correlated in a significant positive way to VER and MAT: 15 out of 26 VER and 80 out of 231 MAT occur in the pr. ind. Slightly more significant, however, appears to be the negative correlation between the pr. ind. and REL.

The statistical analysis of the remaining lexical cues (COMP, EVA, PERS and DESC) and deictic cues (PROX and DIST and the deictic use of tenses), encounters two issues. Firstly, the annotation process of many of these variables is somewhat problematic, since the author is not obliged to make a choice whether or not to insert a word of each category in every clause. Therefore, a positive value (i. e. the presence of such a cue), is often much more relevant in the analysis than a negative one (i. e. its absence). Secondly, these cues are very specifically expected to show a correlation to the P taken on the events, i. e. interpersonal meaning. Working within the three-dimensional framework of meaning described above, we should always take into account the possibility that ideational and textual meanings could also influence the process of verb tense selection; these considerations on the part of the author would have been independent from the perspective cues investigated in this paper (see also note 30).

The third and fourth tests, then, analyzed the hypothesized correlations between verb tense selection on the one hand, and COMP (illustrated in example (4) for the pf. ind.) and EVA (illustrated in example (5) for the impf. ind.) on the other hand. E. g. facere “to make, to do”, dare “to give”, capere “to take”, opprimere “to overwhelm”, parare “to prepare”, oriri “to (a)rise”, uenire “to come”, descendere “to come down”, fugare “to rout”, proficisci “to set out”. E. g. cogitare “to think”, desiderare “to wish”, timere “to fear”, uidere “to see”. E. g. esse “to be”, appellare “to call”, habere “to have, to consider”, posse “to be able”, tenere “to hold”. E. g. pugnare “to fight”, opperiri “to (a)wait”. E. g. dicere “to say”, referre “to report”. E. g. erat “there was”, stetere “there stood”. Personal lexis (N = 439; X² = 3.98 df= 2, p = 0.14) and DESC (N = 439; X² = 0.10, df= 2, p = 0.95) did not pass the tests of independence, i. e. they do not appear to be correlated to verb tense selection within our corpus. For this reason, the correlation plots of these X² tests were not included at this point.

(4) Sub equestris finem certaminis coorta estpeditum pugna, primo et uiribus et animis par dum constabant ordines Gallis Hispanisque. (Liv. 22. 47. 4).

“Toward the end of the cavalry engagement the fight of the infantry began, at first equal in strength and spirit, as long as the Gallic and Spanish ranks stood firm.”

(5) Legatis quos mitteret ad Syphacem calonum loco primos ordines spectatae uirtutis atque pru- dentiae seruili habitu mittebat, qui dum in conloquio legati essent (...) specularentur mo- remque simul noscerent stationum uigiliarumque (...). (Liv. 30. 4. 1-3)

“With the legates whom [Scipio] kept sending to Syphax he would send some first centurions of attested courage and discretion as servants and garbed as slaves, that while the legates were in conference they might take note (...) and discover their practice as regards outposts and sentries (...).”

In example (4), the switch from the end of the cavalry battle, narrated just before, to a previous point in time and a different point in space where the infantry battle began, indicates the ability of an external observer to narrate the events with a certain overview (i. e. viewpoint from without). In example (5), the extra finality in the relative clause containing the (prospective) subjunctive forms specularentur “take note” and noscerent “discover” indicates Scipio's intentions or purposes for mittebat “he sent [repeatedly]”.

For the correlation between the tenses and COMP (N = 439; X² = 6.42, df= 2, p = 0.04) and EVA (N = 431; X² = 36.96, df = 4, p < 0.001), the correlation plots in Figure 3 and Figure 4 represent the standardized residuals of the performed tests.

Based on the relevant samples, the presence of COMP and the presence of external EVA are positively correlated to the pf. ind. (A), whereas the absence of COMP and the presence of internal EVA is positively correlated to the impf. ind. (B).

In the fifth and sixth tests, the hypothesized correlations were evaluated between verb tense selection on the one hand, and deictic indications (illustrated for PROX E. g. hinc aut illinc “on this side or that side”. in example (6) and for DIST E. g. eo die “on that day”, postero die “on the following day”, eius diei “of that day” in example (7)) and deictic tense usage E. g. tenuerit “withheld” and fuerit “was”. Although past participles like uisum “it appeared” are, in my close readings, usually considered merely as a past participle rather than a pf. ind. with ellipsis of a form of esse, such a use of the participle almost exclusively occurs with the function of stating events in a much more simple and economic way in Livy's narrative (“participium economicum”, cf. Aerts submitted). In this passage, therefore, a case could yet be made for the interpretation of uisum as uisum (est). Briscoe (2016) mentions an alternative reading of fuit for fuerit. In that case, the effect of tenuerit would not change, nor would the data or the outcome of the statistical tests. (also illustrated in example (7)) on the other hand (see section 3.2).

Fig. 3. The impact of the presence (“yes”) or absence (“no”) of comparative lexis on verb tense selection

Fig. 4. The impact of evaluative lexis (“internal” or “external”) or its absence (“no”) on verb tense selection

(6) Tunc inter primores duorum populorum res geritur; quidquid hinc aut illinc communis Mars belli aufert, multiplex quam pro numero damnum est. Volgus aliud armatorum, uelut delegata primoribus pugna, euentum suum in uirtute alienaponit. Multi utrimque cadunt, plures uol- nera accipiunt. (Liv. 7. 8. 1).

“The struggle then lay between the best men of both nations, and whatever losses the chance of war inflicted on this side or that side were serious out of all proportion to their number. The common herd of soldiers, as though they had made over the battle to their betters, rested their future on the bravery of thers. Many on both sides fell and more took injuries.”

(7) Id uero indignum uisum⁴⁵ ab tumultuario auxilio iam etiam castra Romana terreri, ut ea modo una causa ne extemplo transirent flumen derigerentque aciem tenuerit Romanos quod summa imperii eo die penes Paullum fuerit.⁴⁶ Itaque postero die Varro, cui sors eius diei imperii erat, nihil consulto collega signumproposuit instructasque copias flumen traduxit, sequente Paullo quia magis non probare quam non adiuuare consilium poterat. (Liv. 22. 45. 4-5)

“This came across as truly outrageous, that now even a Roman camp should be terrorized by irregular auxiliaries, so that only that one reason withheld the Romans from immediately crossing the river and forming up the line, namely that the highest command was with Paulus that day. The next day, therefore, Varro, who had the lot of that day's command, having consulted his colleague about nothing, put forward the standard and brought his formed up troops across the river, while Paulus followed, because he could more easily not approve than not help the plan.”

In example (7), tenuerit and fuerit indicate that a vantage point is taken in the moment of speech, i. e. Livy's (and his audience's) present: these perfect subjunctive forms with past time reference should be interpreted as explicitly anterior to the speech moment. See note 3.

The X² Tests of Independence indicate that deictic indications, if present at all (N = 90; X² = 16.865, df= 2, p < 0.001), and deictic tense usage (N = 415; X² = 24.15, df= 2, p < 0.001) are highly significantly correlated to verb tense selection. The correlation plots in Figure 5 and Figure 6 represent the standardized residuals of these tests.

Based on the relevant samples, the presence of PROX appears to be significantly positively correlated to the use of the pr. ind. (C), while the same seems true for DIST and the pf. ind. (A). In addition, deictically used tense forms are very likely to occur in the cotext of SoAs in the pf. ind., but quite unlikely to occur near an impf. ind. (B) or pr. ind. (C). Note that all of the actual and generic present tense (C) forms, which refer to e.g. geographical facts (N = 24), have been left out of the count of “C” within data1_6 because they are considered “noise”: they are a cotextual cue, they are not influenced by one. Both Conway & Johnson (1935) and Walsh (1986) mention conuehebat, conuehebant, and conuehe- bantur as alternatives to conuehit in the manuscripts; however, I follow -- and attempt to support with my analysis -- Walsh (1986) in his preference for conuehebat over conuehit.

⁵⁰ E. g. iam “already”, non iam “not anymore”, nondum “not yet”.

⁵¹ E. g. per multos annos “for many years”, diu “for a long time”.

⁵² E. g. postero die “the next day”, ad primam uigiliam “by the first watch”.

⁵³ E. g. tunc, inde “(and) then, next”, primo ... deinde ... denique “first ..., then ..., finally ...”.

⁵⁴ See Mittwoch, Huddleston & Collins (2002, sections 6-9) for this classification of temporal adverbials.

Fig. 5. The impact of deictic indications (“proximal” and “distal”) on verb tense selection

Fig. 6. The impact of the absence (“no”) or presence (“yes”) of deictic tense usage in the cotext on verb tense selection

As the last test of the first part of our analysis, the hypothesized correlation between the presence of a temporal adverbial and tense selection was evaluated (see section 3.1). Example (8) illustrates the hypothesis put forth above:

(8) Carthagini erat quidem ingens terror, et circumferentem arma Scipionem omnibus finitimis raptim perdomitis ipsam Carthaginem repente adgressurum credebant. Itaque et muri refi- ciebantur propugnaculisque armabantur, et pro se quisque quae diutinae obsidionis tole- randae sunt ex agris conuehit.⁴⁹ (Liv. 30. 9. 3-4)

“At Carthage there was a veritable panic, and they believed that Scipio, whose forces were circling about them, after swiftly vanquishing all their neighbours would suddenly assail Carthage itself. Accordingly the walls were repaired and armed with battlements; and people each for themselves brought in from the country what was needed in order to endure a long siege.”

In example (8), the temporal structure of the timeline made up of reficiebantur “were repaired”, armabantur “were armed” and conuehebat “brought in” (see note 49) is not mediated by an external narrator; rather, these events were many and all around an internal observer on the spot.

The X² Test of Independence indicates that the use of a temporal adverbial is significantly correlated to verb tense selection (N = 434; X² = 47.52, df = 8; p < 0.001; see note 33). The correlation plot in Figure 7 represents the standardized residuals of the test.

Fig. 7. The impact of the absence (“no”) or presence of a temporal adverbial (“aspectual”,⁵⁰ “extent”,⁵¹ “location”,⁵² “relative”⁵³)⁵⁴ on verb tense selection

Based on the sample data1_7 (N = 434), a significant positive correlation seems to exist between the pf. ind. (N = 174) and all but aspectual temporal adverbials: 6 out of 8 adverbials of temporal extent (“duration”) and 57 out of 107 adverbials of temporal location are found with a verb form in the pf. ind. (cf. also the strong negative correlation to the absence of a temporal adverbial). The impf. ind. (“B”, N = 118), then, appears to be significantly linked to either the use of aspectual adverbials or the absence of any temporal adverbial: 7 out of 8 aspectual adverbials and 92 out of 273 SoAs without a temporal adverbial selected the impf. ind. Importantly for our hypothesis, however, a significant negative correlation seems to exist between the impf. ind. and the presence of an adverb of temporal location.

The correlation between the narrative tenses and perspectives

The second step of the analysis was to show the correlation between the observable variable of verb tense and the latent variable of perspective. For this, a X² Test of Independence was conducted on the subcorpus data2, which includes only those data points in the pf. ind., impf. ind. or pr. ind. and with an unambiguous annotation for one of the (non-neutral) perspectives (N = 378, X² = 121.3, df = NA, p < 0.001; see note 33). Figure 8 plots the standardized residuals of this test.

Fig. 8. The correlation between verb tense and perspective

In those cases where a marked choice of perspective was recognized unambiguously during our close reading of the corpus, significantly strong positive correlations can be found, in line with the hypotheses set forth in section 2, between the pf. ind. (A) and P1, between the impf. ind. (B) and P2 and P3, and between the pr. ind. (C) and P4 and P5. Significant strong negative correlations appear to exist especially between the pf. ind. and P3 and between the impf. ind. and pr. ind. on the one hand and P1 on the other hand.

The correlations between cotextual cues and the annotation of perspective

In the third and final part of our analysis, we will repeat the tests performed in section 4.1 in the same order, but this time with the annotation of perspective as dependent (“outcome”) variable instead of verb tense selection. See section 4.1 for all the information regarding the rationale and methodology of the related statistic tests, in case such information was considered unnecessary to be repeated in this section. The subcorpus that is used in this section (data3) contains all data points for which our close reading resulted in an unambiguous annotation of a marked perspective (N = 596). Note that the data points in data3 also contain (historic) infinitives and participles used as finite verbs, as well as subjunctive forms (see section 4.) In addition, observations which are morphologically ambiguous (e.g. uenit “he comes/came”) were readmitted. The rationale is based on the hypothesis that the correlation between the cotextual cues and “perspective” would exist independently of the morphology of the verb form (cf. note 30).

First, the CIT in Figure 9 shows the combined (i. e. multivariate) impact of “voice” and “subject animacy” on the annotation of P (see section 3.3).

Node 3 in the CIT suggests that only voice contributes significantly here: SoAs in the passive voice and verb forms that do not have a passive conjugation seem to prefer P1 (122 out of 221 observations, i. e. about 55 % of both passive SoAs (N = 122) and SoAs with verbs that have no passive conjugation (N = 99)).

Fig. 9. The combined impact of voice (“active”, “active_only”, “passive”, “deponent”) and subject animacy (“yes”, “no”) on the annotation of perspective

However, note that Node 1 (N = 596) shows that P1 was already used in 43 % (i. e. 256 out of 596) of the cases in the sample data3_L Second, the correlation plot in Figure 10 shows the standardized residuals of the X² Test of Independence that evaluated the correlation between process type and the annotation of perspective (N = 593; X² = 90.92, df= NA, p < 0.001; see note 33).

Based on the sample data3_2 (N = 593), P1 (n = 260) and P2 (N = 20) appear to correlate in a significant positive way to REL: out of 128 REL, 72 are narrated from a P1 and 8 from a P2; the latter count also constitutes 40 % of all internally perspectivized events counted unambiguously in our corpus. A significant negative correlation was observed between P1 and MEN. P3 (N = 107), then, appears to be significantly linked to VER and especially MEN: 21 out of 59 VER and 22 out of 50 MEN occur in P3. Significant negative correlations, however, seem to exist with MAT and REL. Finally, P5 (N = 144) appears to be correlated in a significant positive way to MAT, as hypothesized in section 3.1: 99 out of 318 MAT occur in P5. However, the similar positive correlation that we had hypothesized to exist between P5 and VER, appears to be, in fact, strongly negative: VER appear to prefer P4 (N = 66) instead, of which they constitute 14 observations.

The third and fourth tests of the second part of our analysis evaluate the hypothesized correlations between the annotation of P on the one hand, and COMP (illustrated in example (4) for P1 and in example (9) for P5) and evaluative (illustrated in example (5) for P2 and in examples (10) and (11) for P5 and P1) on the other hand (see section 3.1).

Fig. 10. The impact of process type (“material”, “mental”, “relational”, “behavioural”, “verbal”, “existential”) on the annotation of perspective

(9) [Dum altercationibus magis quam consiliis tempus teritur,] The use of square brackets in the examples is meant to indicate that the bracketed clause should not be regarded as a part of the example which helps to illustrate the use of the perspective in question; such clauses are merely added to provide the reader with the full cotext of passage cited to illustrate the perspective. Hannibal ex acie, quam ad multum diei tenuerat instructam, cum in castra ceteras reciperet copias, Numidas ad inuadendos ex minoribus castris Romanorum aquatores trans flumen mittit. (Liv. 22. 45. 1-2)

“[While the time was spent with discussions rather than with counsel,] Hannibal sent from the battle line, which he had held formed up till far on in the day, as he withdrew the rest of the troops into the camp, the Numidians across the river to attack the water fetchers of the Romans from the smaller camp.”