2.1 Loan word accommodation

During the Old and Middle English periods, the English language borrowed words from various languages, such as Latin, French and Old Norse (e.g. Ingham Reference Ingham2012; Pons-Sanz Reference Pons-Sanz2013; Durkin Reference Durkin2014). Borrowing is taken to mean the integration of new linguistic material into a language, often called ‘replica language’, on the model of linguistic material from another language, often called ‘model language’, with which the replica language has been in contact in some way (Weinreich Reference Weinreich1953). Frequently investigated examples of model languages are Old Norse (ON) and French (Finkenstaedt & Wolff Reference Finkenstaedt and Wolff1973; Grant Reference Grant, Haspelmath and Tadmor2009; Durkin Reference Durkin2014), which have added to the English language with verbs such as ME taken (‘to take’) and travailen (‘to travail, work’) respectively.

The number and nature of borrowings resulting from a contact situation depend on the intensity of contact (Thomason & Kaufman Reference Thomason and Kaufman1991), but also on, for instance, the morphological complexity of the borrowable categories (Matras Reference Matras2020: 191f.). Therefore, linguistic closeness of the languages in contact may favour the borrowing of more complex categories (e.g. Meillet Reference Meillet1921; Moravcsik Reference Moravcsik1975; Johanson Reference Johanson, Jones and Esch2002; Winford Reference Winford2003: 51ff.).

The present study concentrates on loan verbs.Footnote ⁵ The morphosyntactic implications of loan integration of verbs are generally focused on less in models of borrowability and loan integration (cf. Thomason & Kaufman Reference Thomason and Kaufman1991; Matras Reference Matras2020) than the formal or semantic implications of the loan integration of other lexical categories, or they are presented as a constraint on lexical borrowing (cf. Winford Reference Winford2003). Wohlgemuth's (Reference Wohlgemuth2009) work on loan verb accommodation is seminal and identifies four accommodation strategies based on his typological research: ‘direct insertion’, ‘indirect insertion’, the ‘light verb strategy’ and ‘paradigm insertion’. Direct insertion describes a loan verb integration pattern where replica language inflections are added directly onto the borrowed stem, while in indirect insertion an additional verbalising affix is added to the word stem of the copy before it can be inflected in the replica language (Wohlgemuth Reference Wohlgemuth2009: 87, 94). Under the light verb strategy, Wohlgemuth (Reference Wohlgemuth2009: 102) classifies all patterns where a copied verb is integrated as part of a complex predicate in combination with a dedicated light verb which carries inflections. Finally, paradigm insertion shows copied verbs continuing to carry their source language inflections in the replica language (Wohlgemuth Reference Wohlgemuth2009: 118, 119). Direct insertion is the most frequent strategy cross-linguistically and has, moreover, been identified as the most prominent strategy for loan verb insertion into English in both the Scandinavian and French contact situations (Wohlgemuth Reference Wohlgemuth2009: 338). Examples of direct insertion are Old French comander and Old Norse reisa, which are implemented in Middle English as commaund-en and reis-en respectively (cf. Lewis et al. Reference Lewis, McSparran, Kerr, Moske, Powell, Price-Wilkin and Schaffner1952–2001). Both loan verbs are used with the native English infinitival -en marker (cf. native English find-en). Since inflection cannot be avoided under direct insertion, Wohlgemuth's (Reference Wohlgemuth2009: 291) subsequent argument is that loan verb integration is not as constrained by inflection as much as often assumed (e.g. Harris & Campbell Reference Harris and Campbell1995: 135; Sijs Reference Sijs2005: 56–7).

When entering a language through direct insertion, loan verbs are integrated grammatically into their replica language system (Poplack, Sankoff & Miller Reference Poplack, Sankoff and Miller1988; Muysken Reference Muysken2000), which means that they adopt replica language inflections and can be used in all morphosyntactic categories in which model language verbs can be used.Footnote ⁶ In the case of Norse-derived ME taken and French-origin ME travailen in examples (1)–(2), both taken from The Penn–Helsinki Parsed Corpus of Middle English (PPCME2) (Kroch, Taylor & Santorini Reference Kroch, Taylor and Santorini2000–), the verbs are inflected as past third-person singular forms. Taken is used in the past strong form tooke (‘took’), and travailen in the past weak form trauaylde (‘worked’).

That loan verbs taken and travailen can be used in inflected forms such as past forms points to them functioning as fully integrated verbs in ME.

2.2 Loan word accommodation biases

More recent research confirms that loan verbs can be used just like native verbs. However, even under direct insertion, loan verbs are subject to constraints on inflection and are biased towards some morphosyntactic usage categories, a phenomenon referred to as loan word accommodation biases (De Smet Reference De Smet, Van de Velde, Smessaert, Van Eynde and Verbrugge2014; Shaw & De Smet Reference Shaw2022). Specifically, French loans in late ME have been found to occur disproportionately more frequently in uninflected forms than in inflected forms when compared to native verbs (Shaw & De Smet Reference Shaw2022). Additionally, loan verbs are disproportionately more frequent in non-finite forms (i.e. infinitive, past participle, present participle) than in finite forms (i.e. imperative, past, present) when compared to native verbs. An example of the non-finite usage of French loan verb maintenen (‘to maintain’) is provided in (3), where the verb is used in its past participle form, meigtened (‘maintained’). The non-finite usage of this French loan verb is contrasted with the finite usage of native English verb komen (‘to come’) in (4), where kom (‘came’) is a third-person singular form in the past tense. Both examples have been taken from the same text sample from The Helsinki Corpus of English texts (Rissanen et al. Reference Rissanen, Kytö, Kahlas-Tarkka, Kilpiö, Nevanlinna, Taavitsainen, Nevalainen and Raumolin-Brunberg1991).

The usage of finiteness in (3) and (4) above exemplifies the dominant distribution of French loan verbs used in non-finite forms as compared to native English verbs, which are more prevalent in finite forms.

Biases of loan words towards specific morphosyntactic categories have been found not just for verbs, but also for adjectives, both in the historical French contact outcome and for Modern English loan words in Dutch (Shaw Reference Shaw and Smet2022; Shaw & De Smet Reference Shaw2022). However, this article will focus on loan verb accommodation biases in historical contact situations. As suggested by Shaw (Reference Shaw and Smet2022: 241), the morphosyntactic integration of loan verbs requires further investigation, particularly across different contact situations, as the research by Shaw & De Smet (Reference Shaw2022) has only focused on French loans in Middle English and Modern English loans in Modern Dutch. Therefore, the present article deepens our understanding by comparing the accommodation of French and Norse-derived loan verbs into ME. The factors of temporal distance to the period of direct contact and typological relation to ME are the main points of comparison. We compare loan verbs from both French and Norse regarding their morphosyntactic accommodation into the English replica language system by operationalising accommodation biases as a measure of loan verb integration.

5.1 Data and operationalisation

To address these questions, a corpus study on Norse-derived and French loan verbs entering ME was conducted, comparing them to a baseline of native English verbs. Their overall usage as well as the nature and course of their morphosyntactic integration were compared. In this study, accommodation biases served as a measure for the degree of integration of loan verbs, meaning that stronger accommodation biases imply less complete morphosyntactic integration. We operationalised accommodation biases as the difference in relation of non-finite and finite uses of the verbs between foreign etymology verbs (French and ON) and English verbs.

5.2 Data extraction

Data were extracted from The Penn–Helsinki Parsed Corpus of Middle English (PPCME2) (Kroch, Taylor & Santorini Reference Kroch, Taylor and Santorini2000–) and A Parsed Linguistic Atlas of Early Middle English (PLAEME) (Truswell et al. Reference Truswell, Alcorn, Donaldson and Wallenberg2018). The PPCME2 corpus is mostly based on the Middle English section of the diachronic part of The Helsinki Corpus of English Texts (Rissanen et al. Reference Rissanen, Kytö, Kahlas-Tarkka, Kilpiö, Nevanlinna, Taavitsainen, Nevalainen and Raumolin-Brunberg1991). It encompasses 56 text samples, totalling around 1.2 million words. It is subdivided into four time periods: M1 (1150–1250 CE), M2 (1250–1350 CE), M3 (1350–1420 CE) and M4 (1420–1500 CE), following the Helsinki Corpus classification.Footnote ⁸ The PLAEME corpus includes 68 text samples from the Linguistic Atlas of Early Middle English (Laing Reference Laing2023–) which total roughly 173,000 words.Footnote ⁹ Both corpora include syntactic annotations (cf. Truswell et al. Reference Truswell, Alcorn, Donaldson, Wallenberg, Alcorn, Kopaczyk, Los and Molineaux2019) following the Penn Parsed Corpora of Historical English.Footnote ¹⁰ Together, these parsed diachronic corpora span the time between 1150 CE and 1500 CE and include prose of different genres as well as some poetry. However, there is approximately two-thirds less data for the M2 subperiod in the PPCME2 corpus than for the other subperiods in this corpus (Percillier & Trips Reference Percillier and Trips2020: 7172f.). This is why the PLAEME data were used as a supplement to make the data more balanced diachronically (Truswell et al. Reference Truswell, Alcorn, Donaldson, Wallenberg, Alcorn, Kopaczyk, Los and Molineaux2019).

This combination also leads to a more balanced representation of dialect areas (cf. Truswell et al. Reference Truswell, Alcorn, Donaldson, Wallenberg, Alcorn, Kopaczyk, Los and Molineaux2019: 6), as the PPCME2 contains more texts from the east and west Midlands overall and texts from the M2 subperiod only represent the southeast of England, while the smaller PLAEME corpus contains relatively more northern and southern texts (cf. Percillier & Trips Reference Percillier and Trips2020: 7173). As Scandinavian contact spread from the northeast in late OE and French contact spread from the south starting in 1066 CE (see section 3), the contact situations under investigation were most intense in different dialects at different times. Therefore, we controlled for the varying intensity and topological spread of linguistic contact in different regions, operationalised as four broad dialect areas, namely Northern, East Midlands, West Midlands and Southern. Herein, the latter three are operationalised as in the Penn Parsed Corpora of Historical English and Southern combines the Southern and Kentish dialect classifications for the PPCME2 data.Footnote ¹¹ As the dialect text metadata for the PLAEME data is more fine-grained, its broad localisation subcategories South East, South Central, South West and Essex and London were collapsed into one Southern category while the categories North West Midlands and South West Midlands were collapsed into a general West Midlands category to make them congruent with the PPCME2 dialect groups.Footnote ¹² The PLAEME dialect groupings for Northern and East Midlands largely correspond to those of the PPCME2. This way, patterns of loan verb accommodation biases for verbs from either contact situation can be compared between high- and low-intensity contact areas for each contact respectively across both corpora. However, these four dialect areas are not represented equally in the combined data overall, with Northern accounting for 9.77 per cent of the combined data, East Midlands for 45.61 per cent, West Midlands for 26.23 per cent and Southern for 18.39 per cent of the data, and neither are they diachronically balanced, as table 1 shows (cf. also Percillier & Trips Reference Percillier and Trips2020: 7173, figure 3).

Table 1. Percentual distribution of text words per dialect area and time period in the combined PPCME2 and PLAEME data (n = 1,365,624)

Note that 29.21 per cent of text from the aforementioned corpora is based on French or Latin originals with varying degrees of literality. Following Shaw (Reference Shaw and Smet2022), we did not exclude texts on the basis of the language of the original text. From a diachronic perspective, the ME corpora represent an ongoing contact situation with French, while contact with ON had subsided by the end of the Old English period (see section 3). This directly reflects the factor of temporal distance to the period of contact, which possibly affects accommodation biases. This makes the ME data a fitting basis for this comparative analysis of loan verb accommodation.

We queried the dataset for all occurrences of lexical verbs in three etymological groups, namely English, French, and Norse-derived using CorpusSearch (Randall Reference Randall2010). To fulfil this aim, we used versions of the PPCME2 and PLAEME enriched with verb lemmatisations from the BASICS project (cf. Percillier & Trips Reference Percillier and Trips2020).Footnote ¹³ Etymological origins of verbs for these three groups were operationalised as follows: French verbs were queried following the BASICS etymology annotations for French origin verbs. The queried list of Norse-derived verbs was based on the Gersum project database (Dance, Pons-Sanz & Schorn Reference Dance, Pons-Sanz and Schorn2019) as well as the Oxford English Dictionary (Proffitt Reference Proffitt2000–.) and Middle English Dictionary (Lewis et al. Reference Lewis, McSparran, Kerr, Moske, Powell, Price-Wilkin and Schaffner1952–2001). We restricted the set of verbs to lemmata with strong phonological and morphological evidence supporting ON influence for which no cognates are attested in OE (Gersum category A1–A3, Dance, Pons-Sanz & Schorn Reference Dance, Pons-Sanz and Schorn2019, e.g. casten), or where they are, they are neither formally nor functionally equivalent (Gersum category A1*–A3*, Dance, Pons-Sanz & Schorn Reference Dance, Pons-Sanz and Schorn2019, e.g. raise vs rear). A number of verbs which were not classified in the Gersum database were added to the set of Norse-derived verb lemmata under investigation. These verbs were listed in the Oxford English Dictionary (Proffitt Reference Proffitt2000–), Middle English Dictionary (Lewis et al. Reference Lewis, McSparran, Kerr, Moske, Powell, Price-Wilkin and Schaffner1952–2001) or other current research on Norse-derived lexis in ME (Pons-Sanz Reference Pons-Sanz2007, Reference Pons-Sanz2013; Dance Reference Dance2003, Reference Dance, Jefferson and Putter2012) as being of early Scandinavian origin based on sufficient formal evidence. Of these we only included verbs listing no or contrasting native West Germanic cognates to match the conditions of the set of verbs extracted from the Gersum database (e.g. liten ‘to dye’). By extension, the set of English verbs serving as a baseline contained all verbs annotated as ‘non-French’ in the BASICS annotations, also excluding non-contrasting close cognates between English and ON. This way, we reduced the overlap of the etymological verb sets between English and Norse-derived verbs in the extensive domain of cognates between these languages. We eliminated overlap between the etymological sets by excluding all instances ambiguously lemmatised (Percillier Reference Percillier2016: 210; Percillier & Trips Reference Percillier and Trips2020) between French and non-French lexemes (e.g. orthographic type comyn lemmatised as either comen ‘to come’ from OE cuman or as communen ‘to share, commune’ from OF com(m)uniier) and all verbs ambiguously lemmatised between non-contrasting close Old Norse and Old English cognates or other formally close lemmata (e.g. orthographic type lythe lemmatised as either lithen ‘to sail, travel’ from OE līþan or as lithen ‘to alleviate’ from OE līþigian or as lithen ‘to listen’ from ON hlýða). This way, we excluded lemmata that are of mixed Old Norse and Old English influence that could easily be integrated by directly mapping them onto the inflectional paradigm of the native cognate by way of identification between lexemes. Such copies would likely not show accommodation biases of the nature investigated in Shaw & De Smet (Reference Shaw2022) and would not serve to answer our research questions. Including these in our data would have conflated the effects of accommodation processes at work in mutually intelligible communication between high cognate languages and the long-term structural accommodation of loan verbs without identifiable cognates in a language contact situation (see section 3). Our query also excluded be and have, modal verbs and gerunds, the former for their status as auxiliary verbs and the latter for their status as nominalisations. Fixed expressions like according to (ME: accorden < OF) and that is to say (ME: seien) were manually excluded as they are no longer actively generated structures, but lexicalised, and thus do not require active inflection in usage (Shaw Reference Shaw and Smet2022: 78).

5.3 Data analysis

We automatically annotated all retrieved instances of verbs concerning their verb form, lemma, etymological origin and finiteness of the morphosyntactic realisation, drawing on the extracted corpus data and annotations. This resulted in a total of 124,308 attestations. For our operationalised diagnostic of finiteness, we distinguished between non-finite (infinitive, present participle, past participle, passive participle) and finite (inflected present, past, imperative). Additionally, we extracted text metadata concerning dialect (i.e. Southern, Northern, East Midlands, West Midlands; see above for categorisation) and Helsinki time period (i.e. M1, M2, M3, M4; see above for categorisation). For these annotations we followed the PPCME2 classifications (Kroch, Taylor & Santorini Reference Kroch, Taylor and Santorini2000–) to retain comparability across the two corpora.

On this dataset we ran basic quantitative analyses, relating the variables of etymology and finiteness of morphosyntactic realisation generally and across the variable values of time period and dialect area. Chi-square test was used to obtain p-values for the differences in proportion of finite and non-finite forms of each of the two foreign etymology sets, comparing them to the English baseline. Yates’ correction was used as a measure to prevent overestimation of statistical significance of the data.Footnote ¹⁴ For subset analyses like lemma and frequency effects (see section 7.1.1), Fisher's exact test was applied, as this type of test is typically used for smaller sample sizes than Chi-square test (Levshina Reference Levshina2015: 214). The imbalanced nature of the data across the four variables did not allow for valid application of regression analysis. Therefore, we only conducted pairwise comparisons in this study to test the probability of differences being significant.

6.1 Disentangling typological and temporal distance effects

To disentangle the effects of typological distance and temporal distance to the time of direct contact, we take a diachronic perspective on the ME data. Figure 2 shows the proportion of non-finite usage for verbs of all three etymologies split up by Helsinki subperiods, from M1 to M4.

Figure 2. Diachronic distribution of verbs of Norse-derived and French origin compared to the baseline of English verbs in finite and non-finite forms in combined data from PPCME2 and PLAEME (n = 124,308)

Each subperiod of ME shows different trends concerning the proportions of non-finite versus finite usage in the three etymological sets. Diachronically, the non-finite bias for Norse-derived verbs steadily decreases throughout ME. In M1 Norse-derived verbs show a significant non-finite bias (p = 0.0003, Chi-square test) but this is no longer significant (p = 0.6330, Chi-square test) by the M4 period. This points to Norse-derived verbs not yet being well integrated at the end of the direct contact situation between Old Norse and English. Biases for French verbs, however, persist throughout ME and do not show a clear trend of decrease when the temporal distance to the start of the period of direct contact increases. This finding of persistent biases for French verbs even at the end of the direct linguistic contact situation is parallel with the significant bias attested for Norse-derived verbs in M1. What is more, the non-finite bias for French loans initially increases throughout earlier ME (e.g. from M1 to M2). This may coincide with an increase in texts translated from French and the peak of newly attested French loans, which are reflected in the data.

6.2 Dialect areas

The starting locations and speeds of dispersion of the respective linguistic contact settings differ among the dialectal areas. This means that the areas were affected differently by language contact. For example, whereas French found its way into medieval England through the southern dialects (cf. Rothwell Reference Rothwell1983: 259–60), the Old Norse language entered the country through the northern dialects (cf. Pons-Sanz Reference Pons-Sanz2013: 6f.). Dialectal distribution of non-finite biases for the French and Norse-derived etymological sets may reveal more about the diachronic development of accommodation biases, as dialect areas relate to areas of longest and most intense contact. From this, we hypothesise that biases would be least strong in areas where contact originated or was most pervasive. A comparison of biases across different dialects (figure 3) will reveal any diatopic trends.

Figure 3. Finiteness distribution per dialect area for verbs of Norse-derived and French origin compared to the baseline of English verbs in combined data from PPCME2 and PLAEME (n = 124,308)

In figure 3, the dashed black line and the associated percentage given at its lower end show the baseline of non-finite usage of native English verbs for the data from each dialect area. Like in the analyses above, this is the comparandum to which the proportion of non-finite usage of loan verbs is compared for each dialect.

As figure 3 shows, our hypothesis holds for French loans to some degree, as the non-finite bias is least strong in Southern texts (at 43.73 per cent non-finite usage compared to 37.04 per cent for English verbs), where French initially entered English (see section 3). However, next lowest are the biases in Northern texts (56.45 per cent compared to 47.07 per cent for English verbs) and East Midlands texts (50.40 per cent compared to 40.19 per cent for English verbs), while West Midlands texts show the strongest non-finite bias for French loans (47.54 per cent non-finite usage) as compared to English verbs (36.13 per cent). For Norse-derived verbs, a similar trend presents itself in the data, albeit with contrasting implications. Biases are weakest in Southern texts (at 37.25 per cent non-finite usage compared to 37.04 per cent for English verbs), followed in strength by biases in Northern texts (at 50.53 per cent non-finite usage compared to 47.07 per cent for English verbs) and West Midland texts (at 41.97 per cent non-finite usage compared to 36.13 per cent for English verbs), with East Midland texts showing the strongest non-finite bias for Norse-derived verbs with 50.68 per cent non-finite usage (compared to 40.19 per cent for English verbs), as figure 3 shows.

East Midlands dialect texts make up the largest share of data overall in the corpora used (see section 3.2, table 1) and originate from the Danelaw area, where Scandinavian influence was most intense and long-lived. Therefore, this finding is somewhat unexpected, as integration of loan verbs is hypothesised to be more advanced in high-contact areas. However, more than half of texts from the M1 subperiod, during which we would expect the highest biases for Norse verbs diachronically, are from the East Midlands dialect (52.94 per cent of M1 texts). Therefore, we might expect this higher relative bias. Moreover, the Northern and Southern dialect texts mostly stem from later Middle English (M3 and M4; see section 3.2, table 1) and there are no Northern texts from the M1 subperiod at all. Hence, the low accommodation biases for Norse-derived verbs in the Northern and Southern dialects may be a reflection of the diachronic distribution of texts rather than dialect alone.