coniuratione; Ver. Aeneid, selections from book 6; Prop. Elegies, book 1; Ov. Metamorphoses, book 1; Petr. Satyricon 26-78 (Cena Trimalchionis); Vulgata: ...
BETWEEN SEMANTICS AND SYNTAX: SPATIAL VERBS AND PREPOSITIONS IN LATIN LINDA MEINI, BARBARA MCGILLIVRAY (University of Pisa)
1. INTRODUCTION* Studies about Latin verba composita have been mostly devoted to analyzing the role of preverbs in relation to the actionality of Latin verbs (Lehmann 1974; García-Hernandez 1989, 2005; Haverling 2000, 2003; Romagno 2003).1 It has been argued that preverbs modify the actionality of Latin verbs through often unspecified processes: the event is atelic, for instance, in a simplex verb like faciō, but is telic in the prefixed form perficiō. The fact that in most cases these preverbs have a primarily spatial meaning is often considered a given, something so obvious that maybe it is not worth an in-depth analysis. Moreover, previous research on this topic has mainly followed a theoretical example-based approach, with a few exceptions (e.g. Viti 2008). No quantitative studies employing statistical techniques have been conducted in order to test theoretical hypotheses. Instead, we decided to focus our study on the spatial value of preverbs in Latin. In addition, we carried out a corpus-based statistical analysis on two Latin annotated corpora. This analysis allowed us to achieve quantitative results that corroborate our hypotheses, while at the same time showing how procedures of linguistic data annotation commonly used to process modern languages can be fruitfully extended to the study of a dead language like Latin.
* The article is the result of the close collaboration between the two authors; however, for academic purposes, Linda Meini is responsible for background and theoretical discussion (sections 2 and 4) and Barbara McGillivray for processing the corpus data and statistical analysis (section 3); both the authors are responsible for the introduction and conclusions (sections 1 and 5). We would like to thank prof. F. Kammerzell and prof. S. Kutscher and their excellence cluster TOPOI, who invited us to hold a seminar about the present topic at the Humbolt University in Berlin. Furthermore, we would like to thank Gard Jenset for his valuable comments about the final English version of the paper. 1 Scholars of historical grammars of IE languages, as Delbruck (1897), Meillet and Vendryès (1924) and van der Heide (1934) for instance, already pointed out that in Latin, and in all IE languages, preverbation modifies the actional value of predicates. For a slightly different interpretation, see Brucale (2008).
2. DIACHRONIC DEVELOPMENT OF ADPOSITIONS AND PREVERBS 2.1. IE preverbs and adpositions According to a widely accepted reconstruction (Pisani 1949; Kuryłowicz 1964; Lehman 1974), most of the Latin preverbs and adpositions derive from PIE particles and adverbial elements, constituting the category that Cuzzolin et al. (2006) labeled as ADV/ADP/PREV. These particles involved local or temporal notions and were characterized by a rather free position in the sentence, still displaying undifferentiated functions. This is exactly what we find in Vedic and, to a lesser extent, Classical Sanskrit, where there is not a full-blown class of prepositions, but adverbial words used in combination with nouns or verbs. Comparing Vedic with Classical Sanskrit we can witness the gradual evolution of the grammaticalization process of these particles: in Vedic, the particle used with a verb is often separated from it by one or more words (MacDonell 1990: 285), whereas in Classical Sanskrit the preverb is not separated from the simple verb (Whitney 1989: 414 ff.).2 Traces of the freedom of PIE particles are also found, for instance, in Homeric Greek, where it is often very difficult to decide whether the form belongs to the verb or to the noun (Chantraine 1953: 82)3. So, the translation of the verse in (1a.) can be like that in (1b.), if we consider ejk a preposition, or like the one in (1c.), if we consider ejk as a preverb in thmesis: (1)
a. b. c.
ejk de; Crushi;>ı nho;ı bh' pontopovroio [A, 439] “Chryseis came from the sea-crosser ship” “Chryseis disembarked from the sea-crosser ship”
From their unmarked position, which was before the verb, these forms could either be fused with the following verb, thus becoming bounded preverbs, or linked to the preceding noun, thus becoming postpositions. At first, they served as a sort of adverbial accompaniment to reinforce the information already given by the cases themselves; then they ended up governing the noun by specifying its case. In other words, the rise of two distinct grammatical categories, preverbs and adpositions, is traditionally explained by means of two different reanalyses of the basic structure NP + X + V, where X is the oldest category ADV/ADP/PREV. A reanalysis NP + [X + V] brought about bound 2 For a more detailed presentation of both Vedic and Classical Sanskrit data, see Ferrari and Mosca (this volume). 3 See also Pompei (this volume).
preverbs, whereas a reanalysis [NP + X] + V gave rise to postpositions.4 As Cuzzolin et al. (2006: 11) state: «Apparently, the fortune of PREV and ADP does not depend on the basic word order, but it is rather the syntactic position that assigns the functional value of ADV or ADP». The typological shift from SOV to SVO word order, which most IE languages underwent in their diachronic evolution, caused a further development: most postpositions became prepositions (Lehmann 1974). Although there is general agreement on the common adverbial origin of IE preverbs and adpositions, the different stages of their evolution are not uncontroversial (Kuryłowicz 1964; Lehman 1974; Baldi 1979; Oniga 2005). For reasons of space, these hypotheses will not be discussed in detail here. We will make some general considerations during the discussion on the outcomes of our analysis (§ 4). 2.2. Latin data Latin written documents testify a quite recent stage of the Latin language, especially when compared to many other IE languages (Vineis 1993 among others).5 Still in Early and even Classical Latin we can find some traces of the freedom that preverbs and adpositions must have had in older stages. For instance, in Festus we find two very interesting passages where the preverb is detached from the verb (Cuzzolin 1995, where the Latin author is quoted from Lindsay’s Oxoniensis edition): (2)
a. b.
ob vos sacro = obsecrō vos “I beseech you” sub vos placo = supplicō vos “I implore you”
In Festus we also find endo (= in) as preverb (3a.), whereas the Duenos inscription has endo in a postpositional function (3b.):
4 Obviously the linguistic change at issue must have taken decades to stabilize and get fixed, allowing for overlapping transitional stages (Heine et al. 1991; Magni 2008). We believe that at the beginning compound verbs were likely to be created by speakers through synchronic rules, like German trennbare Präfixe (e.g. Er steigt in den Bus ein ‘he gets on the bus’); then preverbs got fixed to the verb through a grammaticalization process, like German untrennbare Präfixe (e.g. In wenigen Minuten erreichen wir München ‘in few minutes we will reach Munich’). In German there are also some prefixes that can be both separable and unseparable depending on whether they are stressed or not: e.g. ü;berlaufen ‘to go over’ (Er läuft zum Feind über ‘he goes over to the enemy’) vs überschrèiten ‘to cross’ (Die Tzruppen überschreiten die Grenze ‘the troops cross the border’). 5 Here we will focus mostly on literary documents (Poccetti 1999; Santini 1999).
(3)
a. b.
endoque plorato = imploratoque “et implore!” nei tēd endō cosmis vircō siēd = ni in te comis virgo sit “lest the girl be not kind towards you”
Lindsay (1936) draws a parallel between Festus’ examples (2a., b.) and Plautus’ passage in (4): (4)
distraxissent disque tulissent satellites tui me [Trin. 833] “your attendants would have torn me in pieces and rent me asunder”
There would be a lot to say about the latter example. For instance, one could object that dis- functions only as suffix in compounds (unlike sub and ob in Festus’ passages) and is never employed as preposition or adverb. Another objection could be that Plautus’ style is linguistically very creative and it does not reflect exactly the spoken Latin. Here it will suffice to say that Plautus’ passage testifies the possibility to detach a prefix from the verb, keeping intact the meaning of the compound verb both for the author and for the audience. In addition to these few examples of Early Latin,6 throughout the history of Latin we find a more consistent syntactic construction: local complements occurring with some prefixed verbs of motion can be encoded both by the bare cases and by prepositional phrases, where the preposition is the same as the preverb, as in the following examples: (5)
a. b.
(6)
a. b.
(7)
a. b.
exercitum Ligerim traducit [Caes., B. G. VII, 11, 9] “he leads the army over the Loire” vexillum trans vallum traiecit [Liv., XV, 14, 4] “he threw the sign over the enemy’s earthwork” castris egressi [Caes., B. G. II, 11, 1] “having marched out of the camp” e castris Helvetiorum egressi [Caes., B. G. I, 27, 4] “having marched out of the Helvetians’ camp” muro turribusque deiecti [Caes., B. G. VII, 28, 1] “being dislodged from the wall and towers” de muro se deiecerunt [Caes., B. C. I, 18, 3] “they threw themselves down the city wall”
6 We can find more examples concerning adpositions, especially with relative pronouns, with a following or a preceding adjective and in fixed expression: Plaut. (Cist. 677) loca haec circiter excidit mi; Plaut. (Cas. 815) sensim super attolle limen pedes; Ter. (And. 834) per ego te deos oro; Cic. (Verr. 3, 92) contraque legem; Plin. (N. H. II, 67) Gades usque pervectum; Cic. (Verr. 3, 50) socii [...] quos inter res communicata est; Plaut. (Pseud. 174) viris cum summis; Caes. (B. G. VI, 14, 4) duabus de causis; mecum, tecum ...; quoad; quocirca; quapropter.
In examples a. of (5), (6) and (7) we have compound verbs that are equivalent to the prepositional phrases in the corresponding examples b. In other words the preverb assigns a case to its complement noun, that is, the complement of the compound verbs in a. is inflected in the same grammatical case as would be expected if a preposition were the head of such a phrase (Horrocks 1981; Viti 2008). Horrocks (1981: 23 ff.) calls preverbs like those in examples a. “prepositional preverbs” and the resulting compound verbs “prepositional verbs”. The syntactic construction of prepositional verbs with bare cases is traditionally said to be a relic of an older stage of the language, where prepositions were not obligatory and preverbs were still keeping a certain degree of semantic and syntactic autonomy from the verb. In order to verify to what extent spatial preverbs have kept their original semantic and syntactic properties throughout the evolution of Latin, we analyzed occurrences of local complements with no spatial preposition occurring with prefixed verbs of motion or rest in two syntactically annotated Latin corpora. We chose this particular syntactic structure for various reasons. First of all because Latin preverbs usually tend to lose their original semantic content and, as we saw at the beginning, add actionality values to the verbum simplex. On account of the loss of semantic content and syntactic function, our hypothesis is that the frequency of this structure will decrease diachronically in texts of Late Latin and in texts where there are hints of spoken Latin. Moreover, this construction is an interesting case of univerbation that will allow us to suggest some considerations on grammaticalization and lexicalization processes. 3. A CORPUS-BASED ANALYSIS 3.1. The corpora The corpus-based statistical analysis we conducted is based on two Latin corpora: the Latin Dependency Treebank (Bamman and Crane 2006) and the Index Thomisticus Treebank (Passarotti 2007). These corpora consist of over 53000 word tokens each and collect selections of the works by a number of Latin authors, ranging from the Late Republic to the Medieval Age: Caes. De bello gallico, selections from book 2; Cic. In Catilinam, 1.1-2.11; Sall. De Catilinae coniuratione; Ver. Aeneid, selections from book 6; Prop. Elegies, book 1; Ov. Metamorphoses, book 1; Petr. Satyricon 26-78 (Cena Trimalchionis); Vulgata:
Apocalypse; Th.: selections from Scriptum super Sententiis Magistri Petri Lombardi and Summa contra Gentiles.7 A pragmatic reason motivated the selection of the corpora: they are the only ones that are available with a rich annotation, which is very relevant to our study. The corpora are syntactically annotated, which implies that each sentence has been manually analyzed by marking the syntactic function of its words. For each sentence, the verbal arguments have been annotated and were automatically extracted to form two valency lexicons (McGillivray and Passarotti 2009): for each verbal occurrence in the treebank, these lexicons contain the list of its arguments. These lexicons constitute the data for our analysis. 3.2. The analysis We selected 58 prefixed verbal lemmas occurring in the corpora that met the following requirements: first, the preverb has a spatial meaning (ab-, ad-, ante-, circum-, de-, ex-, in-, inter-, intra-, per-, prae-, pro-, sub-, trans-); second, the verbs are verbs of motion and verbs of rest allowing a local complement without preposition (“prepositional verbs” in Horrock’s terms). To verify the latter point we used the Thesaurus Linguae Latinae, the Lewis and Short Dictionary and the Conte-Pianezzola-Ranucci Dictionary. We then automatically extracted the argument structure of all occurrences of these 58 verbs as recorded in the valency lexicons, focusing on the syntactic realization of their spatial complements, and in particular on the alternation between the two following constructions: A) arguments expressed as noun phrases whose case coincides with that required by the preverb, like examples (5a.), (6a.) and (7a.); the variable considered is “preposition”, its value ABSENT; B) arguments expressed as prepositional phrases whose preposition coincides with the preverb, like examples (5b.), (6b.) and (7b.);8 the variable considered is “preposition” ”, its value PRESENT.9 Since in A) we aimed at detecting the cases where the use of the bare case could be ascribed to the role of the prepositional preverb rather than directly to the verb’s argument structure, we excluded the transitive verbs where the object 7 One could object to including Thomas Aquinas (13th century AD) in the list of Latin authors, since he certainly did not speak Latin in his everyday life. Still we think that someone who was been considered an authority on Latin by his contemporaries can provide linguistic material relevant to our analysis. 8 We took into account also verbs of rest like intersum. For instance Cic. Cat. (I, 5): inter me atque te murus intersit “there is a wall between me and you”. 9 The construction where a preposition different from the preverb is used would give more evidence for the non-transparency of the preverb. For an analysis of these cases, see McGillivray (2010).
is the GOAL of the movement, such as insequor “to follow” and invado “to assail”. Table 1 shows the occurrences of constructions A) and B) in the different authors attested in the corpora. Author Cicero Caesar Sallust Vergil Propertius Ovid Petronius Jerome Thomas
Construction A (preposition.ABSENT) 2 4 4 4 6 2 10 3 2
Construction B (preposition.PRESENT) 6 4 9 0 2 0 11 7 28
Table 1. Frequency counts of construction A and B by author in the corpora.
Figure 1 shows a barplot of the distribution of constructions A) and B) by author.
Figure 1. Barplot of the distribution of construction A (light grey) and B (dark grey) by author.
3.3. Hypothesis testing From figure 1 it is already possible to detect a general trend in these data. An inspection of this barplot shows that the relative frequencies of construction A) are higher in the poets (Vergil, Propertius and Ovid) and tend
to decrease diachronically. However, we decided to carry out a formal analysis through a statistical test of independence, the Pearson’s Chi-square test.10 In our case, this test compared two patterns of frequencies — construction A) and construction B) by author — to see if the variable “author” and the binary variable “preposition” are significantly independent from each other. Given the small amount of corpus evidence for the poets, we applied the Chi-square test to a table similar to table 1 obtained by merging together the frequencies of Vergil, Propertius and Ovid.11 The result from the test is that the two variables are not independent at a 0.01 level of significance.12 The strength of this association is strong according to the classification in Cohen (1988: 224). Therefore, we can state that the preference for one construction over the other is not independent from the particular author we are analyzing, i.e. the different authors behave differently in this particular respect. In addition to knowing that somewhere in our data there is a significant difference in the distribution of the two constructions, we can find out where this difference is concentrated by inspecting an association plot. Figure 2 is a Cohen-Friendly association plot showing the contribution made by each author-construction pair to the total Chi-square statistic. In this plot, the dotted lines refer to the expected frequencies, each rectangle corresponds to an author-construction pair and its area is proportional to the difference between observed and expected frequencies: the height of the rectangles is related to significance and the width to the amount of data supporting the result. From the size of the rectangles in the association plot in figure 2 we can see that, in the data for Thomas, construction A) (preposition.ABSENT) occurs significantly less frequently than expected, whereas construction B) (preposition.PRESENT) occurs more frequently than expected. On the other hand, the merged data for Propertius, Vergil and Ovid display construction A more often than expected. The low frequencies for Caesar, Cicero, Jerome, Petronius and Sallust contribute very little to the overall association between “author” and “preposition”. 10 The Chi-square test compares the actual frequencies observed in the data with the expected frequencies we would get from a random distribution (Hinton 2004). This test gives two kinds of output: first, the level of significance, which says if the result is significant or not, i.e. if the two variables are significantly independent or not; second, the effect size, which accounts for the size of the detected independence. 11 The Chi-square test gives unreliable results when applied to a table whose expected frequencies are lower than 5. Table 1 produces 9 out of 18 expected cell frequencies below this recommended threshold. For this reason, we decided to merge together the frequencies for the poets: this way, 5 out 18 expected cell frequencies are lower than 5. 12 These are the results from a Pearson’s Chi-square test for independence with Yates’ correction for continuity: 2(6)=29.00, p-value
