Lability and the rigidification of word order: evidence from Early Middle English

1 Introduction

Case marking, agreement and constituent order overlap in their function to disambiguate clausal arguments. For this reason, they correlate in several ways. For instance, there is a tendency for languages with free word order to make use of case marking (Butt 2009: 28), whereas the absence of morphological marking (case inflections and agreement) is linked to a more rigid word order typologically (Siewierska 1998: 507).^[1]

In diachronic terms, it has long been observed that the loss of case marking often goes hand in hand with a shift towards a fixed word order (Primus 2001: 855; Sims-Williams and Baerman 2021: 22). As is well known, Present-day English marks grammatical relations mainly by rigid SVO constituent order and case morphology is confined to (some) pronouns. In contrast, nominal paradigms did show case, as well as gender and number distinctions in Old English (the first attested stage of the language, c. 700–1100), and constituent order was then much more flexible. Noun morphology was severely impoverished during the Middle English period (c. 1100–1500), when word order variability was reduced. The possibility of a causal link between the two processes based on functional arguments has been often addressed in the literature.^[2] Some scholars have proposed that the loss of case marking in Middle English called for the rigidification of word order patterns in order to distinguish clausal constituents (thus for instance, Fischer and van der Wurff 2006: 188; Marchand 1951; Pysz and Wiland 2012). Others have suggested that word order inflexibility made case marking redundant and hence prone to disappear (e.g., Lehnert 1957; Shores 1971: 217; implicitly Berndt 1989: 110–111). Traugott (1972: 111) affirmed that both processes feed into each other; so, too, Allen (2006: 220), who wrote that it “seems likely that the two developments worked hand in hand”. The functionalist proposal that there is a causal relationship at all between the loss of case marking and rigidification of word order in Old English has been questioned among others by Pintzuk (2002a, 2002b). She demonstrates that overt case morphology does not influence the position of the object in Old English.^[3]

One of the main goals of this study is to assess the effects of a so far unexplored instance of morphological syncretism that may have favored the rigidification of word order patterns in English, namely the absence of the valency alternation marking, which leads to ‘S/P labile’ or ‘ambitransitive’ verbs (see a succinct description of valency alternations in Malchukov and The Leipzig Valency Classes Project Team 2015: 33–34). The term S/P labile, or ‘labile’ for short, designates here the quality of those verbs that can alternate between an intransitive (more specifically unaccusative) and a transitive frame without morphological change, whereby the subject of the intransitive corresponds to the object of the transitive reading. Compare the use of burn in (1) and (2):

Note that the alternation illustrated in (1) and (2) is one between a non-causal and causal sense of the verb burn. In some studies, it has been called the inchoative-causative alternation (Haspelmath 1993; Levin 1993; Malchukov and Comrie 2015), but Haspelmath et al. (2014) provide convincing arguments for rejecting this term. When used in reference to a labile verb, ‘intransitive’ and ‘transitive’ also imply ‘non-causal’ and ‘causal’ in this article.

The number of labile verbs greatly increases in Early Middle English (Ingham 2020; McMillion 2006). This is likewise the period in which there is still optionality between VO and OV, but the frequency of OV is in rapid decline, until it becomes strongly structurally restricted by 1400 (Fischer et al. 2017: 195). Therefore, Early Middle English prose texts are the obvious choice for a study that examines the connection between these two linguistic developments.

A second goal is to quantify the effect, in Early Middle English, of the most salient morphosyntactic variables that previous studies on early English have already shown to have an impact on word order.^[4] The new contribution of the present article is that it aims to measure statistically the conditioning force of each of these well-known variables in a substantial corpus of Early Middle English prose, as well as to place them on a hierarchical scale of predictive power and to establish possible correlations and interactions between them.^[5]

The article is organized as follows: Section 2 introduces the main working hypothesis of the study: the possible relationship between verbal valency coding and syntax. Section 3 describes the corpus (Early Middle English prose) and presents all the variables whose effect on the position of objects has been assessed (object type, object length, clause type, type of verbal cluster and lability). Section 4 discusses the results and evaluates them qualitatively and quantitatively. Section 5 offers a summary and conclusion.

2 Labile verbs and syntax

The following quote from Los (2015) summarizes one of the linguistic assumptions that suggest a connection between lability and word order:

Syntax creates slots for certain kinds of information and provides routines for lining this information up word by word. The verbs are particularly helpful for the hearers or readers of the message because they show them what information to expect [emphasis mine]: AGENTS, PATIENTS, attributes, etc. (Los 2015: 3)

That is, the valency of a verb, its combining power, creates expectations that are part of the communication process. The following examples let intuit rather plainly the workings of expectation and the conflicting expectations that labile verbs may generate:

Example (3) is a complete sentence. The addressee does not necessarily expect anything after was sleeping. In (4) the verbal cluster was making, which can only be transitive, arouses expectations that are not fulfilled: the utterance is incomplete. However, (5) can be interpreted in two ways: (i). As an intransitive clause with the children as subject-patient and were burning as an unaccusative verb (alarming news in most cases), or (ii). As an incomplete utterance, where the subject, the children, is an agent and were burning a transitive causal verb. Labile verbs (burn in this case) are literally ambivalent, ambiguous with respect to their valency, and therefore create multiple expectations. This may interfere with communication and influence the way we use them in discourse, as the interpretation of their valency depends solely on contextual cues. In particular, lability might affect word order patterns. This connection can be glimpsed by comparing two English sentences with their German and Spanish translation in (6), (7) and (8); the door and its equivalent in German and Spanish is the subject in (6a), (7a) and (8a) and object in (6b), (7b) and (8b).

The English sentences contain a labile verb, close, which is unaccusative in (6a) and transitive in (6b). In these English sentences, word order alone shows whether door is object or subject, and hence whether closed is used in its unaccusative or transitive sense, which is the relevant point here. In contrast, the transitive and unaccusative senses are formally distinguished in the corresponding Spanish and German verbs. In Spanish, the unaccusative sense has a pseudo-reflexive pronoun se; in German two different verbs are used for the unaccusative and transitive senses (ging zu and machte zu). In both languages, the nouns puerta and Tür follow the verb, but can be correctly interpreted as subject or object because the verbs are unambiguous with respect to valency. In these examples, there seems to be a correlation between word order (in)flexibility and (un)ambiguous valency: ambiguous valency is associated with a fixed word order in (6) and unambiguous valency with flexible word order, at least in terms of argument placement, in (7) and (8). Whether this correlation can be detected in a specific language, English, at a particular stage, Early Middle English, is what this study sets out to discover.

3 Corpus and methodology

The Early Middle English period (conventionally 1150–1300) is pivotal for both the processes with which this article is concerned. At this stage of the language, the number of labile verbs was rapidly increasing (Ingham 2020), while the optionality between VO and OV was dwindling in favor of the first. The corpus for this study on Early Middle English is comprised of the earliest prose texts in the second edition of the Penn-Helsinki Parsed Corpus of Middle English (PPCME2, 2000), namely those composed or whose manuscript dates from between 1,150 and 1,250 (periods M1 and MX1 respectively in PPCME2). All the texts are from the South and Midlands areas, since no northern texts from this period have been preserved. Some philological information about the texts in the corpus is provided for descriptive purposes only.

The Kentish Homilies (MS Cotton Vespasian D xiv) are written in the Kentish dialect. The East Midlands (EM) texts are the Peterborough Chronicle, First and Final Continuations, in MS Laud Misc. 636 (MS. E of Anglo-Saxon Chronicle), the Trinity Homilies (MS Trinity 335, B.14.52) and Vices and Virtues (MS Stowe 34). Finally, three texts are from the West Midlands (WM): the Ancrene Riwle (MS Cotton Cleopatra C vi), the Lambeth Homilies (MS Lambeth 487) and the five texts assembled in the so-called Katherine Group (MS Bodley 34), Hali Meiðhad, St Julian, St Katherine, St Margaret and Sawles Warde.

Table 1 below visualizes the texts that make up the corpus of the study, as they are named in PPCME2 (Text ID), with an indication of their date of composition (CDate) or the manuscript date (MDate), if the former is unknown, dialect and genre, all based on PPCME2. The total word count is 210,769 words.

The first main goal of the study is to assess the effect of a verb’s valency (labile or non-labile transitive) on the position of the object with respect to the verb. Given that transitive instances of labile verbs in the corpus are not that numerous, it was deemed more effective to start by finding as many verbs which were labile in Early Middle English as possible and extract all the transitive clauses in which they appear in the corpus. After that, some non-labile transitive verbs were selected, and all their examples collected.

The identification of the labile verbs in the corpus was a complex undertaking, as determining a verb’s valency at a particular stage in the past is not straightforward. Dictionaries and corpora of earlier stages of English, such as the Dictionary of Old English (Cameron et al. 2018), the Middle English Dictionary (MED) or PPCME2 do not provide that information at all, or not consistently. Moreover, the Early Middle English period is of course not dealt with separately in MED. Since the definition of lability is crucial in this study to ensure reliable results, it seemed reasonable to err on the side of caution and follow the most restrictive view of the term.

The selection of the labile verbs for the data sample was performed in two separate stages. First, a provisional list was drawn up based on existing secondary literature. In particular, the Middle English reflexes of the Old English labile verbs in García García (2020), Hermodsson (1952), and Visser (1963) were included, as well as the Middle English labile verbs in García García and Ingham (forthcoming) and Ingham (2020). The second step was to confirm that the verbs were in fact used as labile in the early decades of Middle English and not only later in the period. It was considered that a verb was labile in Early Middle English if attested more than once in both an intransitive and transitive sense in the relevant texts in PPMCE2 (M1 and MX1) or in quotes provided by the MED antedating 1,250. This condition wasn’t met by some of the Middle English verbs listed as labile in the previously mentioned García García and Ingham (forthcoming) and Ingham (2020), such as adiliʒen, biburien, setten, ripen, shrinken, sinken, strecchen, thauen, colen, dimmen, drien, narwen, shitten or souren, which were therefore culled.

The following verbs were tagged as labile in Early Middle English and all their transitive instances in the corpus registered: anginnen ‘to begin’, aquenchen ‘to extinguish, to die out’, awaken ‘to wake (sb.) up, to be awake’, baken ‘to bake, to undergo baking’, beien ‘to bend’, brennen ‘to burn’, breken ‘to break’, forbrennen ‘to burn up’, formelten ‘to melt (sth.) completely’, gladen ‘to gladden, to be gladdened’, goden ‘to improve’, hongen ‘to hang’, heien ‘to raise, to go up’, heten ‘to heat, to become hot’, lighten ‘to light, to emit light’, melten ‘to melt’, alighten ‘to light, to emit light’, openen ‘to open’, quenchen ‘to extinguish, to die out’, sheden ‘to divide, to separate’, stiren ‘to set in motion, to move’, tobreken ‘to break into pieces’, turnen ‘to cause to rotate, to rotate’, warmen ‘to warm’, wenden ‘to turn’.

The fragments below illustrate the intransitive (a) and transitive (b) argument frames of some of openen, quenchen and breken:

All the clauses in the corpus in which one of the above listed verbs was used with a direct object, as in (10b), (11b) and (12b) were collected for the database.

Further, several non-labile transitive verbs were selected, and the clauses in which they are embedded included in the data sample. The verbs chosen were don, finden, maken, nimen, quellen, taken, slen, sechen, thurhsechen, underfon, undernimen. Notice that all of them are mono-transitive verbs. With ditransitive verbs such as bringen or given the presence of a second object might affect the position of the first and interfere with the results of the study. To which extent this might be the case is not elucidated in recent literature on the position of objects in double-object constructions, as it focuses on the position of the arguments in relation to each other, and not to the verb (e.g., Allen 2006; Koopman and van der Wurff 2000; Yáñez-Bouza 2016). Therefore, to avoid unpredictable interferences, no ditransitive verbs were included in the data sample.

Examples had to be extracted manually using an external search engine, as verbs are not tagged for valency and the corpus is not lemmatized. The relevant fragments were loosely translated and the parsing in PPCME2 was examined and occasionally modified. For instance, the argument in bold in (13) is tagged as direct object in my data sample, but not in PPCME2:

Quite a large number of the initially selected clauses had to be removed from the sample. For example, relative clauses with an object relative pronoun had to be discarded. They are obviously not diagnostic of word order, since the relative pronoun (in bold in [14]) must be the first clausal constituent. An example is:

Imperative clauses were also excluded, as they almost never show OV order in the corpus, and given their abundance, their inclusion would have inflated the number VO instances out of proportion. One of the very scarce examples of an object (in bold) preceding the imperative is:

The total number of tokens is 961, 212 of them with labile-transitive and 749 with non-labile-transitive verbs. The clauses containing the verbs were collected in a spreadsheet and tagged for six variables: (i) object type (pronoun, NP or quantified); (ii) object length (short objects consisting of three words or less, long objects with more than three words); (iii) clause type (main, subordinate, conjunct and infinitive); (iv) type of verbal cluster (with or without an auxiliary); (v) valency of the verb (labile or non-labile transitive) and (vi) object-verb order (VO or OV), which is the independent variable in this study.

Unlike variable (v), the effect of variables (i) to (iv) on the position of objects relative to the verb the in early stages of English (Old and Middle English) has been examined in numerous previous works. As has been pointed out in the Introduction, the reason to include them in the present analysis is threefold: to quantify their effect in the selected corpus of Early Middle English prose, to determine whether there are any interactions among them or between them and lability (for example, whether the effect of lability on the position of the object is canceled with object pronouns) and to establish a hierarchy of conditioning power among them, including the fifth one, lability.

Regarding variable (i), it has long been acknowledged that OV is much more likely with pronominal than with nominal objects in early English (see Denison 1993: 40; Fischer et al. 2000: 141–142, 2017: 195; Los 2015: 173; Mitchell 1985: II, 965–966; Traugott 1992: 276).^[11] The following example illustrates the frequent OV pattern with a personal pronoun as object, in bold:

Example (17) records the postverbal position favored by NP objects:

Among nominal objects, quantified NPs have been shown to precede the verb more frequently than non-quantified ones in Late Middle English (van der Wurff 1999). The trend seems to be operative in Early Middle English, too, according to Pintzuk et al. (2000) and Kroch and Taylor (2000: 156), who deduce from an admittedly small database of clauses with two objects, one of which is a postverbal pronominal object, that “non-quantificational noun phrases [unlike quantificational ones] do not prepose in Early Middle English [from their assumed position in underlying structure]”. The sentence below shows a quantified preverbal object in our sample:

With respect to variable (ii), VO is more frequent with long objects for reasons of processing ease (see for instance Kohonen 1978: 123–132; Pintzuk 2002b: 384; Pintzuk and Taylor 2006: 252, 254). Weight may overlap with object type (variable i), as pronouns are usually shorter than NPs, but not fully, as will be argued in Section 3.^[12] In the following example, a long NP object follows the verb, as expected:

Variable (iii), clause type, clearly affected the position of verb and object in Old English, as main clauses had a tendency to verb second, whereas verb-final was favored in subordinate clauses (see e.g., Denison 1993: 29–30; Fischer et al. 2017: 188–189; Mitchell 1985: II, 958–959). It logically follows that OV is more frequent in the latter than in the former.^[13] On the other hand, conjunct clauses – those introduced by and or ac ‘but’ – have been tagged separately because of their special behavior in Old English, where they may display verb-late and verb-final order (and hence OV), even when coordinated to a main clause (Bean 1983: 88; Bech 2017; Los 2015: 167; Mitchell 1985: Vol. I, 694, II 958). Infinitive clauses were also included with an independent tag, as they show distinctive word order patterns in the Continuations of the Peterborough Chronicle (henceforth PeterCont). The category includes infinitival purpose clauses and infinitival complements to non-auxiliary verbs, as in (20) and (21) respectively:

Variable (iv) distinguishes between clauses with verbal clusters consisting of a finite auxiliary verb and a non-finite main verb and those with a single finite verb. In the tagging for this work, the category auxiliary verb includes (pre)modal verbs (connen, moten, mouen, ouen, shulen, willen), as well as let + infinitive and have and be + past participle. An example with willen is:

This variable is crucial for much of the recent research on verb-object placement in Old English and Early Middle English, such as Kohonen (1978), Kroch and Taylor (2000), Pintzuk (1999, 2002a, 2002b), and Pintzuk and Taylor (2006). The authors of the three latter studies restrict their corpus to clauses with a verb phrase containing an auxiliary and an infinitive complement “where the order of the (non-finite) main verb and its complements is not affected by finite verb movement” (Pintzuk 2002b). They exclude from their database clauses with a single tensed verb as not diagnostic of VO because they assume that their element order is affected by such movement (see further the explanation in Kroch and Taylor 2000: 144–145). Whether this is the case is immaterial to the present study, whose main goal is to assess the effect of lability on word order and is based solely on surface phenomena. Nevertheless, clauses with and without an auxiliary have been tagged differently to quantify the divergence from one another.

Variable v) addresses the valency of the main verb in the clause. Its effect on word order has not been taken into consideration in the literature so far, except in the aforementioned study on the Continuations of the Peterborough Chronicle. Example (9) above illustrates the assumed workings of this variable.

Once the tagging was completed, the impact of variables i) to v) (object type, object length, clause type, type of verb phrase and lability) on the position of the object relative to the verb was assessed. First, Pearson’s chi-squared test (Pearson 1900) was performed on each factor separately in order to measure the significance of their effect. Further, some of the factors were cross-tabulated for statistical independence. Finally, a step-down logistic regression analysis was conducted using Rbrul (Johnson 2009) with the aim to find out which multivariate model accounts for the data in the most effective way. The alpha value in this study is 0.05 (this is the most usual significance level). This means that p-values < 0.05 are statistically significant, whereas those >0.05 are not. The data and the results of the tests are presented and discussed next.

4 Quantitative analysis of the variables that affect the position of verb and object in Early Middle English

5 Summary and conclusion

This investigation has analyzed quantitatively the impact of five morphosyntactic variables on the position of the object with respect to the verb in Early Middle English prose (ca. 1150–1250), when variation between VO – OV was still possible. The corpus comprises all prose texts marked as M1 and MX1 in The Penn-Helsinki Parsed Corpus of Middle English (210,769 words). Four of those variables have been the subject of previous research on word order: object type, object length, clause type and type of verbal cluster (±auxiliary). However, their effect in relation to each other has not been quantified before. One of the main results of this article has been to establish a hierarchy of the factors influencing the placement of the object in Early Middle English. The analysis concludes that the first two variables, object type and object length, have the strongest conditioning force on the choice between VO and OV in our sample, which only confirms the results of preceding studies. With respect to clause type, there is no significant difference in the position of the object in main, subordinate and infinitive clauses in our data sample of Early Middle English prose. Only conjunctive clauses diverge significantly by favoring OV out of proportion, although this may well be a function of the frequent omission of the subject in this type of clause and merits further research. The type of verbal cluster unequivocally emerge as a significant predictor of the position of the object in Early Middle English prose, with verb phrases with an auxiliary exhibiting OV more frequently than those without one. This supports the differential treatment of the two types of verb phrase in recent research, if not necessarily their authors’ assumption that clauses without an auxiliary are not diagnostic of word order.

The central achievement of this paper has been to introduce labilility – the fifth variable under consideration – as a possible contributor to the rigidification of word order. Lability is an instance of morphological syncretism which, contrary to the loss of case marking, has not been associated with word order changes in scholarly discussions in the past. Labile verbs are ambiguous with respect to their valency. This study has argued for the hypothesis that they might favor a fixed word order pattern because of its disambiguating function. From that hypothesis follows that labile verbs, when used in a transitive sense, favor the expected order VO, or what is the same, that the receding variant OV is significantly more exceptional in Early Middle English clauses with labile transitive verbs than in clauses with non-labile transitive verbs. The hypothesis was tentatively tested in one single Early Middle English text, the First and Final Continuations of the Peterborough Chronicle, with promising results. The data of the present study provide robust proof that lability is a reliable predictor of VO in Early Middle English. A multiple logistic regression analysis with Rbrul shows that a model including lability as a predictor is more consistent with the data than one without. This implies that labile verbs anticipate the trend towards a fixed (S)VO order, which will be the norm for all verb types by 1400. The study shows further that lability interacts with object type, with NPs (largely unmarked for case) showing OV much more frequently than PROs (with case generally unambiguous) in transitive clauses with a labile main verb than in those with a non-labile verb. This interaction provides additional support to the functional arguments in which the main hypothesis of the study is grounded, namely that lability plays a role in the move to a more rigid constituent order in English. The results of this study may have relevance beyond the scope of English and inspire further language-specific and typological studies on word order.