The position of the genitive in Old English prose: Intertextual differences and the role of Latin

1 Introduction

The variation of the genitive choice in English is a very-well researched case of syntactic alternation. It has been approached from a number of theoretical perspectives, with studies spanning different historical periods (see Rosenbach (2014) for a summary of relevant studies). Presently, the basic choice is that between the –‘s genitive in the prenominal position and the of-genitive in the postnominal position, but the latter structure only started to be regularly employed in Middle English (ME). In Old English (OE), the alternation saw the inflected genitive occur before or after the head noun, as illustrated in (1) and (2), respectively.

Rosenbach (2014: 240) remarks that “most of the factors governing genitive variation have probably been ‘discovered’”. Among internal variables, the semantics of the genitive and the weight of the possessor/possessum seem of particular importance for OE. Mitchell (1985: 551) points to an almost categorical preposition of proper name genitives in phrases with two unmodified nouns, adding that also “the names of persons, things, and qualities” are barely attested in postposition in such phrases. Allen (2008: 96–7) reports after McLagan (2004) that “animacy was a strong predictor of prenominal position”, while additional adjectival modification of either the genitive or the possessum was linked to postposition in over two thirds of cases in Ælfric’s Catholic Homilies. Sampson (2010: 60–1), analysing a larger collection of prose texts, confirms that the postnominal placement of the genitive is a) slightly more visible if the genitive is modified, b) almost as frequent as prenominal placement if the head noun is additionally modified, and c) overwhelmingly preferred if both nouns are modified.

Naturally, animacy and syntactic weight may overlap, in that animate nouns tend to be short, but Rosenbach (2005: 638) concludes that “the effect of animacy as a determinant of grammatical variation cannot be reduced to an effect of weight (and vice versa), despite the statistical correlation holding between the two factors”. This suggests that “[s]tudying the individual contribution of a single factor and establishing its relative impact vis-à-vis the many other factors that determine genitive choice has been one of the main challenges in research on genitive variation” (Rosenbach 2014: 217–8). Earlier studies, such as Altenberg (1982) on genitive variation in 17th-century English, address this issue by “conducting analyses on each individual factor within controlled subcorpora” (Rosenbach 2014: 217), while more recent studies, such as e.g. Rosenbach et al. (2014) on Late Modern English genitive constructions, employ complex multivariate models of statistical regression, a method which “quantifies the effect that individual explanatory factors have on a binary dependent variable, such as genitive outcomes” (Rosenbach et al. 2014: 140).

Language-external variables have been demonstrated to play a part too, including dialect or genre (Rosenbach 2014: 233). Another factor of potential significance in OE is the influence of the Latin source text in the case of translations. Overall, studies in OE syntax seem to underappreciate the possibility of Latin interference with the target structures. Cichosz et al. (2016: 32) remark that in many works “all clauses from translated texts are analysed, without any reference to the source texts, and the authors do not provide the equivalent Latin clause in examples used for analysis”. Meanwhile, a small-scale study by Taylor (2008) demonstrates that, aside from obvious syntactic calques, translation effect may be indirect, in that the overall high incidence of the source structure increases the production of the corresponding structure in the target text, even with no clausal counterparts in Latin. Likewise, Timofeeva (2010: 198) found that the most frequent translation strategies of source structures may spread to original texts produced in the target language and “become minor use patterns in text types that imitate corresponding Latin text types”. This multifaceted interference is barely surprising given that the best part of written Old English was produced by English–Latin bilinguals.

The complexity of the relations between Latin and Old English is further compounded by the fact that the translators were very conscious of the audience on the receiving end of their output and had to make their linguistic choices accordingly. Stanton (2002: 83) remarks that Alfred the Great’s translation enterprise “combines the priorities of preservation, textual devotion, and linguistic performance”, while Mugglestone (2012: 54) adds that the king paid attention to “championing the vernacular” so that the texts produced in the course of his scheme could be used in schools to teach English; to that end, the language of these translations tended toward the native idiom that “neither calls attention to itself nor adheres self-consciously to any Latin pattern” (Stanton 2002: 58). Ælfric, who was responsible for a number of scriptural translations, had misgivings about changing the word order in texts of divine provenance, only to conclude that “correct, idiomatic English is the best protection against misunderstanding” (Stanton 2002: 137). Therefore, the relations of source and target texts were typically not reducible to the basic formula of either “word for word” or “sense for sense”.

Since Latin permeated written Old English on multiple levels, it seems reasonable to consider it as another variable potentially influencing the choice between the investigated genitive variants. Mitchell (1985) raises this possibility to explain away some instances of genitival postposition. Crisma (2012: 205, fn. 11) observes that in “some texts, i.e. Cura Pastoralis and, to a lesser extent, Bede’s Ecclesiastical History, the rate of postnominal genitives approaches 50%”, while Yerkes (1982: §33b, in Allen 2008: 113) notices that the reviser of the later manuscript of Gregory’s Dialogues moves many postnominal genitives to a pre-head position, thus reversing the order dominating in the Latin original. On the other hand, Nunnally (1992: 363), based on a self-compiled parallel corpus, discovers that the overwhelming preponderance of postnominal genitives in the Latin version of the Gospel of Matthew “is not significantly influencing OE genitive placement”, since the translator uses the prenominal genitive 96.3% of the time. These results suggest that a larger-scale systematic comparison of source and target texts is called for to fully appreciate the nature of possible Latin interference in the structures under investigation.

The emerging picture also hints at possible intertextual variation, which may have been a local reflection of a more global trend toward an increased dispreference of late OE for postnominal inflected genitives: Cura Pastoralis and Ecclesiastical History were originally composed earlier than the Gospels, which may imply that the genitive alternation displayed diachronic tendencies. Such explanation is suggested by Allen (2008: 118), who reports “a statistically highly significant trend towards favouring the prenominal position for structurally assigned genitive case” in later OE. Therefore, composition/manuscript date appears to be another candidate for inclusion in a detailed analysis of how a combination of language-internal and language-external factors motivated the position of the OE genitive.

The global aim of this study is to explore the degree of intertextual variation in genitive placement in the corpus of OE prose texts and assess to what extent translation effects may be responsible for the identified differences. To that end, the following study questions have been formulated:

1. How do the factors described in the relevant literature as influencing the position of the genitive (additional modification of the head noun, weight of the genitive phrase, animacy) operate in individual texts?

2. Is there a difference in genitive placement between translated and non-translated texts?

3. Does Latin exert a visible influence on genitive placement in OE translated texts?

Our study is not the first analysis of OE genitive placement where intertextual differences are noted, but it is the first one to explore them in detail and produce a comprehensive picture of intertextual variation in genitive position across OE prose texts on the basis of all available corpus data. What is more, even though in earlier studies differences between translations and non-translations were explained by the possible influence of Latin, this influence has never been systematically explored. Our study provides a detailed comparison of Latin phrases and their OE equivalents to show the extent to which foreign transfer may have influenced the genitive position in various early and late OE translations.

Section 2 describes the methodology of the study. Section 3 presents general results, with three subsections focusing on additional modification of the head noun (3.1), weight of the genitival modifier (3.2) and animacy (3.3) respectively. The relative strength of these factors is assessed in Section 3.4. The aim of Section 4 is to analyse the relation between the Latin source texts and their Old English translations as far as genitive placement is concerned. Section 5 concludes the study.

2 Methodology

The study is based on the data from the Toronto-York-Helsinki Parsed Corpus of Old English Prose (YCOE) (Taylor et al. 2003), searched by means of the CorpusSearch 2 application (Randall et al. 2005–2013). The data extracted from the corpus for a pilot study included all the noun phrases with a genitive modifying the head noun, placed before or after the head noun. Then, a set of 100 token phrases with a genitival modifier was sampled from the 10 longest texts in YCOE (1000 phrases altogether) and after a close inspection of the data we identified a set of 9 regularly recurrent constructions containing a genitival modifier (listed below). Then, we wrote queries to extract all of these recurrent constructions from YCOE (the queries are included in the appendix).

GN (a head noun modified only by a preposed genitive: query 1)

GAN (a head noun premodified by a genitive followed by an adjective: query 2)

DAGN (a head noun premodified by a demonstrative followed by an adjective and a genitive: query 3)

DGN (a head noun premodified by a demonstrative followed by a genitive: query 4)

AGN (a head noun premodified by an adjective followed by a genitive: query 5)

NG (a head noun modified only by a postposed genitive: query 6)

DNG (a head noun premodified by a demonstrative and postmodified by a genitive: query 7)

ANG (a head noun premodified by an adjective and postmodified by a genitive: query 8)

DANG (a head noun premodified by a demonstrative followed by an adjective and postmodified by a genitive: query 9)

It was also frequent for OE genitives to be dislocated, whereby the genitive was separated from the head noun, or split, whereby part of the genitive phrase came before and part after the head noun, as illustrated in (12) and (13) respectively. Since the study is focused on the variation between pre- and postnominal placement, such cases are not taken into consideration in the analysis.

Naturally, the list of patterns presented in (3)–(11)^[1] is not exhaustive and there are other, theoretically possible arrangements such as DGAN or GNA, but they were so infrequent in the samples extracted for the pilot study that we decided to ignore them and focus on the more productive patterns. All the queries used to extract the data were structured in such a way that there is no overlap between the results (i.e. DGN is not a part of GN since in the query extracting GN phrases G is the sole modifier of N). This gave us 9 initial output files, but we kept modifying the queries to obtain more detailed results, as described in the following paragraphs.

Weight of the genitive was measured in words, i.e. light genitives are bare nouns without modification as in (10), (12) or (14). These phrases were extracted by means of the iDomsOnly function (the line ‘NP-GEN* iDomsOnly N^G|NR^G’ was added to the respective query files). Longer phrases such as in (11) or (16) were extracted by means of iDomsTotal function (the line ‘NP-GEN* iDomsTotal> 1’ was added to the queries). This procedure rendered 18 output files in total (for each of the nine genitival constructions we got two output files, one with light and the other with heavy genitives).

Animacy could not be checked automatically because there is no semantic annotation in YCOE. However, we decided to use the division into proper and common nouns as a rough guide since most proper nouns are names and demonyms, so there are more animate nouns among proper nouns than among common nouns. This is of course a technical solution which is only an approximation of the semantic criterion, but it does yield some meaningful results, as shown in Section 3.3. To divide the genitives into proper and common nouns, we searched the output files from stage 2 (all the constructions with a subdivision into light and heavy genitives, 18 output files), using the iDoms function (‘NP-GEN* iDoms NR^G’, i.e. the genitive phrase containing a proper noun, or ‘NP-GEN* iDoms N^G’, i.e. the genitive phrase containing a common noun). This gave us 36 output files altogether and all the results reported in the following section come from this procedure.

The data obtained in this way were aggregated for the purposes of statistical analysis^[2] in the following way:

1. without additional modification of the head noun: GN and NG

2. with additional modification of the head noun: GAN, DAGN, DGN, AGN (for premodification) and DNG, ANG and DANG (for postmodification)

This allowed us to construct a contingency table including three independent variables (additional modification of the head noun [with or without], weight of the genitive phrase [light or heavy] and noun type [proper vs. common]), with position of the genitive (pre- or postnominal) as the dependent variable. Data in this format were used to construct a logistic regression model in R (R Core Team 2014) for the general data in YCOE. For the data from individual texts, we treated text type as the next independent variable. On the basis of the general results presented in Section 3, we grouped the texts into those authored by Ælfric, biblical translations and early translations, leaving Vercelli Homilies as a separate type since they do not pattern clearly with any other text. This file was the basis for another logistic regression model. Finally, we also fed the data into the Weka implementation (Hall et al. 2009) of the C4.5 decision tree algorithm (Quinlan 1993). The tree shows which of the variables are most useful to model the choice between pre- and postnominal placement of the genitive. The visualisation was created in yEd Graph Editor.

In the absence of parallel corpora, the analysis of Latin influence required manual identification of equivalent Latin phrases in the source texts. Therefore, we resorted to sampling in order to measure the degree of dependence between source and target text order. For each of the analysed translations we took a sample of the first 50 prenominal and the first 50 postnominal genitives and manually searched for their equivalents in the source texts. The Latin source texts of the translations were selected on the basis of their availability in the electronic form since the exact manuscripts on which the translations were based are unknown. According to Marsden (1995: 396), “for the first 29 chapters of Genesis no Latin witness survives at all”, and Petersen (1994: 256) admits that “the Latin archetype from which the West Saxon Gospels were translated is lost”. Thus, it is not possible to determine which manuscript of the Latin Vulgate is closest to the perished source text. For Bede’s Historia Ecclesiastica, the situation is similar: there are several manuscripts of the Latin text and four complete manuscripts of the OE translation, but the original OE translation has not survived (Wallis 2013), and “none of the manuscripts available seem to carry the exact text which served as copy for the translation” (Lemke 2015: 43). For the Latin text of Cura Pastoralis, the surviving manuscripts are a mixture of two versions, only one of which was the basis for the Alfredian translation (Pratt 2007: 142) and it is “impossible to reconstruct Wærferth’s original text” for the Old English translation of Gregory’s Dialogues (Thijs 2006: 273). Thus, our choice of manuscript (Migne 1849a, Migne 1849b, Plummer 1896, Weber & Gryson 1994) was based on technical criteria.

3 Results

The queries described in Section 2 yielded 20,202 noun phrases with a genitival modifier for YCOE in general. Ten texts with the greatest amount of genitives selected for detailed analysis (listed in Table 1 and arranged according to the frequency of the queried structure) contain 11,606 phrases altogether, which means that they represent the majority of the whole YCOE corpus.

As can be seen in Table 1, in most texts the most frequent pattern is GN (with the prenominal genitive as the sole modification of the head noun); this is true about all the analysed texts except Cura Pastoralis (where DNG prevails), though in Gregory’s Dialogues and Bede the dominance of the GN pattern is rather limited. The general frequency of GN ranges from 92.5% in the West Saxon Gospels to only 36.8% in Gregory’s Dialogues, so intertextual differences are substantial and the general result for the whole YCOE corpus (69.4% for GN) does not reflect the proportions between the analysed patterns in any of the longer texts.

When all the patterns with a prenominal genitive and all the patterns with a postnominal genitive are put together, some more differences come to light (see Table 2).

It turns out that the general result for YOCE (76% of prenominal genitives) does not reflect the whole range of possibilities either. There are seven texts which clearly prefer premodification: Ælfric’s Supplemental Homilies, Catholic Homilies I and II and Lives of Saints (i.e. all the Ælfric’s texts analysed in this study), biblical translations as well as the Vercelli Homilies, while in Bede, Gregory’s Dialogues and Cura Pastoralis the proportions between pre- and postnominal genitives are quite balanced. All in all, it is possible to group the texts and the most simple division would be:

1. Ælfric’s texts (ca. 80% of GN, other patterns attested but rare)

2. Biblical translations (ca. 90% of GN, other patterns hardly present)

3. Early translations, i.e. Bede, Gregory’s Dialogues and Cura Pastoralis (40–50% of GN, strong presence of other patterns)

4. Vercelli Homilies (70% of GN, resembles the aggregate result for YCOE).

The differences between individual texts are rather striking and call for further investigation. Interestingly, all the texts in which the proportion of postnominal genitives is relatively high are translations, which suggests that Latin may have had an impact on the use of the postnominal pattern. However, not all the translations follow this tendency since the biblical ones (West-Saxon Gospels and the Heptateuch) resemble non-translated prose in their clear preference for the prenominal placement of the genitive, so the possibility of foreign transfer is not straightforward and other factors also need to be taken into account.

3.4 Summary

In order to assess the degree of correlation between the factors analysed in the previous section, we used the chi-square test to determine whether the categorical variables are independent. Table 6 shows the overlap between noun type (proper vs. common) and weight of the genitival modifier.

Table 6:

Overlap between weight and noun type for genitival modifiers.

Text	Proper nouns			Common nouns
	Light G	Heavy G	Total	Light G	Heavy	Total
Non-translations
Ælfric’s Supp. Hom.	248 (87.9%)	34 (12.1%)	282	89 (27.3%)	237 (72.7%)	326
Catholic Homilies I	556 (87.3%)	81 (12.7%)	637	285 (29.9%)	667 (70.1%)	952
Catholic Homilies II	486 (90.5%)	51 (9.5%)	537	244 (28.5%)	613 (71.5%)	857
Lives of Saints	390 (90.9%)	39 (9.1%)	429	122 (28.4%)	307 (71.6%)	429
Vercelli Homilies	142 (75.9%)	45 (24.1%)	187	237 (40.4%)	349 (59.6%)	586
Translations
Bede	538 (72.8%)	201 (27.2%)	739	275 (27.8%)	713 (72.2%)	988
Cura Pastoralis	125 (82.8%)	26 (17.2%)	151	99 (11.3%)	780 (88.7%)	879
Gregory’s Dialogues (C)	372 (72.1%)	144 (27.9%)	516	138 (10.8%)	1143 (89.2%)	1281
Heptateuch	612 (95.9%)	26 (4.1%)	638	74 (21.6%)	269 (78.4%)	343
West-Saxon Gospels	284 (90.4%)	30 (9.6%)	314	268 (50.1%)	267 (49.9%)	535
YCOE	6282 (83.8%)	1212 (16.2%)	7494	4053 (31.9%)	8655 (68.1%)	12,708

It turns out that there is a statistically significant difference in weight between proper and common nouns, i.e. proper nouns are generally lighter than common nouns in all the texts,^[6] so the impression that these two groups of nouns behave differently may be to some extent a side effect of their relative length. Of course, this does not exclude animacy as a meaningful factor since common nouns may also be animate, but a further study involving semantic analysis of the data is needed to determine whether animacy is indeed an independent variable influencing the OE data.

Table 7 shows the overlap between additional modification of the noun and type of the genitival modifier. It turns out that in some texts these variables are strongly independent (Catholic Homilies II, Vercelli Homilies and West Saxon Gospels) or rather independent (Catholic Homilies I, Gregory’s Dialogues), while in others there is a statistically significant dependence between the factors: head nouns take additional modification less frequently when the genitival modifier is a proper noun.^[7] There is also one exception: in Lives of Saints it is actually the other way round and proper noun genitives co-occur with additional modification more often than common nouns. Thus, the overlap in the case of these two variables is generally much weaker and the factors seem rather independent.

Table 7:

Overlap between additional modification of the noun and type of genitival modifier.

Text	Proper nouns			Common nouns
	Modified N	Non-modified N	Total	Modified N	Non-modified N	Total
Non-translations
Ælfric’s Supp. Hom.	33 (11.7%)	249 (88.3%)	282	76 (23.3%)	250 (76.7%)	326
Catholic Homilies I	90 (14.1%)	547 (85.9%)	637	173 (18.2%)	779 (81.8%)	952
Catholic Homilies II	79 (14.7%)	458 (85.3%)	537	152 (17.7%)	705 (82.3%)	857
Lives of Saints	94 (21.9%)	335 (78.1%)	429	50 (11.7%)	379 (88.3%)	429
Vercelli Homilies	48 (25.7%)	139 (74.3%)	187	128 (21.8%)	458 (78.2%)	586
Translations
Bede	213 (28.8%)	526 (71.2%)	739	413 (41.8%)	575 (58.2%)	988
Cura Pastoralis	53 (35.1%)	98 (64.9%)	151	546 (62.1%)	333 (37.9%)	879
Gregory’s Dialogues (C)	328 (63.6%)	188 (36.4%)	516	717 (56.0%)	564 (44.0%)	1281
Heptateuch	22 (3.2%)	656 (96.8%)	678	37 (12.5%)	258 (87.5%)	295
West-Saxon Gospels	9 (2.9%)	305 (97.1%)	314	29 (5.4%)	506 (94.6%)	535
YCOE	1429 (19.1%)	6065 (80.9%)	7494	3678 (28.9%)	9030 (71.1%)	12,708

Finally, Table 8 shows the overlap between additional modification of the head noun and weight of the genitive. The factors are independent in Gregory’s Dialogues (no statistical significance) and Lives of Saints (weak significance); in all the other texts heavy genitives are more likely to co-occur with other modifiers of the head noun, while light genitives are usually the only modifiers of the noun.^[8] Thus, the overlap is rather considerable.

Table 8:

Overlap between additional modification of the noun and weight of the genitive.

Text	Light G			Heavy G
	Modified N	Non-modified N	Total	Modified N	Non-modified N	Total
Non-translations
Ælfric’s Supp. Hom.	33 (9.8%)	304 (90.2%)	337	76 (28.0%)	195 (72.0%)	271
Catholic Homilies I	76 (9.0%)	765 (91.0%)	841	187 (25.0%)	561 (75.0%)	748
Catholic Homilies II	81 (11.1%)	649 (88.9%)	730	150 (22.6%)	514 (77.4%)	664
Lives of Saints	98 (19.1%)	414 (80.9%)	512	46 (13.3%)	300 (86.7%)	346
Vercelli Homilies	56 (14.8%)	323 (85.2%)	379	120 (30.5%)	274 (69.5%)	394
Translations
Bede	180 (22.1%)	633 (77.9%)	813	446 (48.8%)	468 (51.2%)	914
Cura Pastoralis	56 (25.0%)	168 (75.0%)	224	543 (67.4%)	263 (32.6%)	806
Gregory’s Dialogues (C)	284 (55.7%)	226 (44.3%)	510	761 (59.1%)	526 (40.9%)	1287
Heptateuch	30 (4.4%)	656 (95.6%)	686	37 (12.5%)	258 (87.5%)	295
West-Saxon Gospels	17 (3.1%)	535 (96.9%)	552	21 (7.1%)	276 (92.9%)	297
YCOE	1510 (14.6%)	8825 (85.4%)	10,335	3597 (36.5%)	6270 (63.5%)	9867

When all these factors are put together (see Table 9) to assess their combined impact on the variation between pre- and postnominal position of the genitive, it allows us to formulate some interesting observations. First of all, regardless of genitive type (proper or common noun), when there is no additional modification of the head noun and the genitive is light, the genitive is placed before the head noun (for proper nouns the result for the whole corpus is 99% and there are hardly any exceptions, for common nouns some exceptions are attested but the pattern is very strong). In addition, regardless of genitive type (proper or common noun), when there is additional modification of the head noun and the genitive itself is also modified (i.e. heavy), postmodification is dominant. Interestingly, intertextual differences come into play when only one of the factors (additional modification of the head noun or additional modification of the genitive) is involved. When the noun is not additionally modified but the genitive is heavy, in most texts it is usually placed before the noun. The only clear exception to this rule is Bede, where postposition is more frequent with proper nouns (62%) and relatively frequent with common nouns (37%). Also in Gregory’s Dialogues the proportion of postposed genitives is quite high (40%), but only with proper noun genitives. When the noun is additionally modified but the genitive is light, the differences between texts are most substantial. Premodification dominates for both proper and common noun genitives in Ælfric’s Supplemental Homilies, Catholic Homilies I and II, Gregory’s Dialogues, the Hepteteuch, Lives of Saints and Vercelli Homilies, but the degree of dominance differs from 100% to ca. 75%. West-Saxon Gospels should probably also be included here though due to the low number of proper nouns in this category they do not pattern well with the others. In Bede and Cura Pastoralis premodification is more common with proper noun genitives (though only at 67.5% and 73% respectively) but postmodification dominates with common noun genitives (72% and 74%).

Table 9:

The combined influence of all the factors on the position of the genitive.

Noun	Proper								Common
Modified N	No				Yes				No				Yes
Weight of G	Light		Heavy		Light		Heavy		Light		Heavy		Light		Heavy
Order	Pre-N	Post-N	Pre-N	Post-N	Pre-N	Post-N	Pre-N	Post-N	Pre-N	Post-N	Pre-N	Post-N	Pre-N	Post-N	Pre-N	Post-N
Ælfric’s Supp. Homilies	223 (100%)	0 (0%)	22 (84.6%)	4 (15.4%)	25 (100%)	0 (0%)	0 (0%)	8 (100%)	78 (96.3%)	3 (3.7%)	160 (94.7%)	9 (5.3%)	7 (87.5%)	1 (12.5%)	1 (1.5%)	67 (98.5%)
Bede	409 (97.8%)	9 (2.2%)	41 (38.0%)	67 (62.0%)	81 (67.5%)	39 (32.5%)	0 (0%)	93 (100%)	199 (92.6%)	16 (7.4%)	227 (63.1%)	133 (36.9%)	17 (28.3%)	43 (71.7%)	4 (1.1%)	349 (98.9%)
Catholic Homilies I	493 (99.4%)	3 (0.6%)	33 (64.7%)	18 (35.3%)	44 (73.3%)	16 (26.7%)	0 (0%)	30 (100%)	260 (96.7%)	9 (3.3%)	426 (83.5%)	84 (16.5%)	13 (81.3%)	3 (18.8%)	5 (3.2%)	152 (96.8%)
Catholic Homilies II	423 (98.1%)	8 (1.9%)	20 (74.1%)	7 (25.9%)	50 (90.9%)	5 (91.5%)	2 (8.3%)	22 (91.7%)	211 (96.8%)	7 (3.2%)	417 (85.6%)	70 (14.4%)	22 (84.6%)	4 (15.4%)	3 (2.4%)	123 (97.6%)
Cura Pastoralis	92 (100%)	0 (0%)	6 (100%)	0 (0%)	24 (72.7%)	9 (27.3%)	1 (5.0%)	19 (95.0%)	73 (96.1%)	3 (3.9%)	235 (91.4%)	22 (8.6%)	6 (26.1%)	17 (73.9%)	8 (1.5%)	515 (98.5%)
Gregory’s Dialogues (C)	118 (100%)	0 (0%)	42 (60.0%)	28 (40.0%)	231 (90.9%)	23 (9.1%)	0 (0%)	74 (100%)	102 (94.4%)	6 (5.6%)	400 (87.7%)	56 (12.3%)	24 (80.0%)	6 (20%)	3 (0.4%)	684 (99.6%)
Heptateuch	590 (100%)	0 (0%)	22 (84.6%)	4 (15.4%)	19 (86.4%)	3 (13.6%)	0 (-)	0 (-)	63 (95.5%)	3 (4.5%)	221 (95.3%)	11 (4.7%)	6 (75.0%)	2 (25.0%)	2 (5.4%)	35 (94.6%)
Lives of Saints	304 (99.7%)	1 (0.3%)	26 (86.7%)	4 (13.3%)	83 (97.6%)	2 (2.4%)	1 (11.1%)	8 (88.9%)	100 (91.7%)	9 (8.3%)	253 (93.7%)	17 (6.3%)	13 (100%)	0 (0%)	7 (18.9%)	30 (81.1%)
Vercelli Homilies	109 (96.5%)	4 (3.5%)	20 (76.9%)	6 (23.1%)	26 (89.7%)	3 (10.3%)	1 (5.3%)	18 (94.7%)	204 (97.1%)	6 (2.9%)	212 (85.5%)	36 (14.5%)	20 (74.1%)	7 (25.9%)	3 (3.0%)	98 (97.0%)
West-Saxon Gospels	279 (99.6%)	1 (0.4%)	17 (68.0%)	8 (32.0%)	2 (50.0%)	2 (50.0%)	0 (0%)	5 (100%)	251 (98.4%)	4 (1.6%)	236 (94.0%)	15 (6.0%)	12 (92.3%)	1 (7.7%)	4 (20.%)	12 (75.0%)
YCOE	5194 (99.0%)	50 (1.0%)	615 (74.9%)	206 (25.1%)	886 (85.4%)	152 (14.6%)	17 (4.3%)	374 (95.7%)	3438 (96.0%)	143 (4.0%)	4772 (87.6%)	677 (12.4%)	313 (66.3%)	159 (33.7%)	84 (2.6%)	3122 (97.4%)

A logistic regression model built for the data from the whole YCOE corpus identifies all the three factors as significant predictors of the genitive position: proper noun genitives increase the probability of prenominal placement, while additional modification of the head noun and heavy genitives increase the probability of postnominal placement in a statistically significant way. A logistic regression model built for the 10 selected texts confirms the significance of weight and additional modification, fails to confirm the impact of genitive type, and identifies text type as a significant variable: biblical translations favour prenominal placement, while early translations^[9] favour postnominal placement in a statistically significant way (summary of both models is included in Appendix 2). The decision tree (shown in Figure 1) generated on the basis of the same data confirms these results and visualises their relative strength.^[10]

Figure 1:

Decision tree modelling the choice between pre- and postnominal placement of the genitive.

The tree shows that lack or presence of additional modification is the strongest predictor of word order since it causes the primary split. When no modification is present, the prenominal placement is predominant. When the noun is further modified, weight of the genitive comes into play and when the genitive is heavy, the postnominal placement prevails. However, when the noun is modified, but the genitive is light, there is another split: proper noun genitives are placed before the head noun, while the behaviour of common noun genitives depends on the text, with early OE translations preferring the postnominal pattern, and all the other texts favouring the prenominal placement of the genitive. In short, statistics confirms that even though additional modification, weight of the genitive and (possibly) its animacy influence the genitive’s position, intertextual differences exist and early OE translations stand out from the other texts in a significant way. There are two possible explanations for this: diachrony (Bede, Cura Pastoralis and Gregory’s Dialogues are early OE texts) and foreign transfer (the texts are all translations). It is difficult to measure the strength of both factors independently because there is no long non-translated early OE text that we could use for comparison, but the possibility of foreign transfer may be checked by looking at translation strategies of early and late OE translators, which is the topic of the next section of this paper.

4 Influence of Latin on translated texts

The influence of Latin on the constructions investigated has been checked on an individual-text basis, since the “Latins” used by particular authors may differ significantly (see e.g. Cichosz et al. 2016). With respect to the placement of the genitive, postposition was initially uncommon in Latin, but its incidence increased in time and the alternation depended on pragmatic factors (Magni 2009: 238–9). It is therefore more accurate to speak of the potential influence of individual source texts. Table 10 presents the arrangement of the Latin NPs corresponding to the analyzed OE phrases (100 in each text, as explained in Section 2). Some NPs were free translations, i.e. did not correspond to Latin NPs consisting of the genitive and the head noun, while Table 10 only considers those examples where these elements were present so that the percentages would be meaningful, i.e. reflect whether the Latin text tended toward preposing or postposing the genitive. These relative values are then compared with those holding for the corresponding OE translations.

However, comparing the mere rates at which both patterns occur in source and target texts could be potentially misleading, or at least may not be grounds enough for stating that a given text is directly affected by the source syntax or decidedly free of any such influence. In order for the comparison to provide as fine-grained a picture as possible, each OE noun phrase was individually checked against its Latin counterpart from the sample to determine if the ordering of elements (GN or NG) in the translated version follows the original or modifies it. The results are presented in Table 11, where “following Latin” means that the target passage replicated the relative order of the genitive and the noun in the source, while “modifying Latin” indicates that the translation either reverses the original order or renders a Latin passage where no genitive phrase was used, thus being an instance of free translation.^[12]

Clearly standing out are the biblical texts, i.e. the Heptateuch and the West Saxon Gospels, whose Latin originals are most consistently NG, with the relative numbers for this pattern standing at 94% and 96% respectively. Interestingly, the scriptural translations show by far the greatest dispreference for postposing genitives, instead putting them before the noun: the Gospels and the Heptateuch are GN 94% of the time. Table 11 indicates that when either text places the genitive in postposition, they do tend to follow the Latin original, but this is barely surprising given the very high ratios of postnominal genitives in the Latin originals.

Indeed, plotting the OE phrases and their Latin counterparts (i.e. phrases where Latin used either GN or NG; free translations are thus not considered) on a contingency table proved that the choice of the target structure did not depend on the source (the difference in the distribution of the target variants was not statistically significant at p < 0.05^[13]). The exact numbers behind these relations in both scriptural translations are shown in Table 12.

All in all, the arrangement of the genitives and their head nouns in the biblical translations is independent of the original texts: in the analyzed sample, postnominal genitives do follow the source as in (22), but almost all prenominal genitives modify it, being moved to what seems to be their preferred position in OE, as in (23).

If the global proportions of the analyzed target structures are compared with those emerging from the source text samples (Table 10), it turns out that the OE and Latin versions of the scriptures are on the opposite ends of the scale with respect to the placement of the genitive, which is consistent with what Nunnally (1992) observed for the one Biblical text he analysed, namely the Gospel of Matthew. One possible explanation of this tendency is proposed in Stanton (2002), who explains that the educational purpose which the scriptures served might have paradoxically prompted the translator to change the original word order. Under this proposition, native syntax is the best protection against misunderstanding, so using the everyday idiom actually stands the readers in good stead since they are not distracted by the text’s foreign feel, which would likely be the case were the original syntax to be copied word for word. In this particular syntactic context, too high an incidence of postnominal genitives – which clearly dominated in the original compositions – might have been such a distraction, so the translator reverted to a more natural order.

Bede is another text which prefers postposing the genitive in Latin: the NG arrangement is observed 79% of the time. Unlike in the case of the scriptural translations, the OE version does not disfavour this order that visibly, with the rate of genitival postposition reaching 43%. Here, too, postnominal genitives frequently reproduce the source, doing so 82% of the time. However, this is not just a result of the high incidence of NG in the Latin text: comparing the target structures with those passages in the source which employed a genitive phrase revealed a statistically significant difference between the distribution of the prenominal and postnominal variants.^[14] These relations are demonstrated in Table 13.

This means that the translator was more likely to postpose the genitive in response to a similarly arranged NP in Latin as in (24). This is in line with Lemke (2015: 187), who, similarly to the YCOE compilers, uses Miller’s edition of the OE text, commenting that the translation “runs word for word rather than sense by sense and does not significantly deviate from the Latin source”, a style which is “sometimes at the brink of being gloss-like”. Lemke (2015) hypothesizes that this faithfulness may have reflected the translator’s recognition of Bede’s authority, a view shared by Rowley (2015), who remarks that “[i]n following Bede closely throughout the translation, the OE Bede signals that Bede himself is authority enough for a history of Anglo-Saxon England”. However, it should be noted that at the same time it was not uncommon for the translator to use GN where the source used NG as in (25).

King Alfred’s OE renditions of Cura Pastoralis and Dialogi are said to be close translations, often criticized for their slavishness. Bately (1988: 118) remarks that the Dialogues are “occasionally so faithful to their originals that the result is unidiomatic”, while according to Brown (1969: 667), the Pastoral Care is “characterized (…) by a literalness and a rather timid dependence upon the Latin text”. Since both Latin originals only slightly disfavor genitival postposition (45% and 44%, respectively), a higher-than-average rate of genitival postposition in the target is not surprising, and calques should be theoretically expected. Even if the OE translations behave somewhat differently from each other, with the Pastoral Care being NG 57% of the time, and the Dialogues using this pattern in 49% of cases, it is striking that in the analyzed sample both display the exact same rates of following and modifying the source structures, be it GN or NG. But when free translations are excluded, interesting differences come to light: it turns out that the OE version of the Pastoral Care is significantly more likely to postpose the genitive when the Latin order is also NG,^[15] as in (26).

These relations are presented in Table 14.

Meanwhile, no such dependence obtains in the Dialogues, where the differences in the distribution of the target variants, illustrated in Table 15, are not statistically significant.^[16]

In addition, it is noteworthy that – with free translations included – both these texts were quite ready to postpose genitives where the Latin versions had them preposed or did not use a genitive at all as in (27) (62% of the time in the analyzed sample), while Bede, on the other hand, was visibly less inclined to do so (such reversals happen 18% of the time).

On the whole, it appears that the overall high incidence of NG in the Latin Bede had a direct translation effect on its OE rendition, where postnominal genitives tend to be calques. The Pastoral Care and Gregory’s Dialogues appear to behave in line with a mechanism well-described by Taylor (2008): the translation effect appears to be indirect, meaning that postnominal genitives were not necessarily a response to Latin phrases employing this order (which appear with an above-average frequency, but are relatively rarer than in Historia Ecclesiastica). Still, it would be an oversimplification to conclude that the OE Bede was slavish on the score of genitival postposition: note that among the 36 cases of GN which have their Latin counterparts, as many as 23 actually reverse the Latin order. In short, while the majority of postnominal genitives in Bede were calques, the majority of prenominal genitives were reversals of the Latin order, meaning that the translator decided against calquing the original.

Overall, we are looking at texts which increased their use of postnominal genitives somewhat independently of their Latin originals (Pastoral Care and the Dialogues), possibly to achieve a more global effect of similarity, as well as texts that may have been more reliant on the source when they did choose postposition, but at the same time regularly avoided it where calquing would have been an option (Bede). In addition, there is another group of texts, which substitutes almost categorical genitival postposition for almost categorical genitival preposition (the Heptateuch and the West Saxon Gospels). All this suggests that non-native transfer does not override the intra-linguistic constraints on the genitive placement identified earlier in the paper. Indeed, looking at the postnominal genitives in OE that copy the relative arrangement in Latin reveals that in most cases they occur in phrases where both the genitive and the head noun are modified. This is shown in Table 16.

The scriptural translations somewhat stand out, in that a good deal of their calques from Latin in the sample studied are phrases where only one of the genitive or the head noun is additionally modified, whereas the modification of either element was shown to be a less stable predictor of postposition than the heaviness of both. However, recall that the Heptateuch and the West Saxon Gospels are texts where the rates of pre-position are close to 95%, so even the erratic behavior of the relatively few instances of postposition is not enough to conclude that these translations are in any way suspect in terms of allowing foreign transfer. In the case of the Pastoral Care and Gregory’s Dialogues, it is safe to assume, based on the findings of the previous sections, that the overwhelming majority of their postnominal genitives calquing Latin would have been perfectly idiomatic anyway, given that they occur in phrases where both the genitive and the noun are additionally modified. This leaves the Ecclesiastical History, where over a quarter of postnominal genitives copying the Latin source would not have been that unexpected in pre-position. In sum, Bede’s text seems to be the only translation considered in this study which may be somewhat suspicious with regard to the potential influence of Latin on the position of the genitive.

5 Conclusions

The study has demonstrated major intertextual variation with respect to genitive placement across a representative corpus of OE. However, the differences are not categorical: the variables tested, i.e. additional modification of the head noun, weight of the genitive, or its animacy impact the choices in each text, but the degree of this influence varies. Sometimes, greater syntactic weight of either element slightly reduces the incidence of the prenominal placement, which continues to prevail nevertheless (as in the scriptural translations), and sometimes it results in a visibly increased rate of postposition (as in Bede). The most consistent behaviour has been observed when the factors typically associated with the postnominal placement of the genitive operated concurrently. This lends support to the general assumption underlying other studies in genitive variation and summarized by Rosenbach (2014), namely, that factors contributing to the eventual choice of the variant should be studied vis-à-vis each other.

In addition, the study hints at a possible diachronic trend, having discovered (similarly to Allen 2008) statistically significant differences between early and late OE texts, with the former tolerating postnominal genitives to a greater extent than the latter. However, the early texts in question are translations, and, in the absence of more or less independent contemporary compositions of sufficient length which could be used for comparison, it is difficult to verify the hypothesis of change in time. The growing reluctance of later OE texts to postpose genitives may have been a result of the increased confidence of OE scribes, who were becoming less and less dependent on Latin, the influence of which (be it direct or indirect) may have been stronger when the Anglo-Saxon written tradition was still in its infancy. What is certain is that some of the early translations were somewhat reliant on their sources with regard to genitive placement. A case in point is the OE version of Bede’s Historia Ecclesiastica: some aspects of its clause-level syntax have been demonstrated to depend on Latin (see e.g. Cichosz 2017; Cichosz et al. 2016), and it appears that the text may not be entirely free of foreign transfer on the phrase level either. In any case, the findings of this study strongly suggest that the influence of Latin on the syntax of OE should be by no means disregarded and that translation effect should be included among the variables studied in the context of syntactic alternation.

Finally, it turns out that translation status on its own is not enough to explain the variation in genitive placement as the differences between individual OE translations are rather far-reaching. The study highlights the importance of an individual approach to every text included in the YCOE corpus: a prori generalizations of any kind and grouping texts according to date of composition, translation status, genre or dialect area will conceal significant differences between individual compositions. YCOE is not a balanced corpus and it has never aspired to be one so a bottom-to-top approach seems the most reasonable way of studying the degree of intertextual variation and the impact of extralinguistic factors on various aspects of OE syntax.