The Emergence and Loss of the English Minor Complementizers till and until

Abstract

This article examines the development of the subordinators till and until as minor complementizers in the Late Middle English and Early Modern English periods. An analysis of data obtained from a number of sources shows that till/until underwent a process of secondary grammaticalization, emerging as complementizers introducing clauses governed by the desiderative predicate long. The findings further suggest that the use of till/until-clauses with long was the result of a process of lexical diffusion from the semantically related pattern think (it) long till/until-clause (in the sense of ‘to seem or appear long, to be wearisome (to a person) (until something happens)’). In Late Modern English, till/until-clauses following long were lost and replaced by competing patterns with to-infinitives and for. . .to-infinitives, the latter emerging at the time as a new complement type. The article discusses the motivations and broader implications of the obsolescence of till/until-complements, which failed to spread to other members of the class of verbs of “desire,” such as desire, thirst, or yearn, and thus remained at the margins of the English system of complementation.

1. Introduction

As shown in previous research (López-Couso & Méndez-Naya 2015), subordinators with an originally adverbial use, for example, as if in (1), constitute one of the pathways leading to the emergence of new complementizers, as illustrated in example (2). In (2), as if does not introduce an adverbial clause of comparison (functioning as an adjunct), but rather a complement clause which expresses an internal argument of the predicate seem and serves as its notional subject. In this function, as if and other so-called “minor complementizers” are close equivalents of the major declarative complementizers that and “zero” (López-Couso & Méndez-Naya 2015:189).

(1) He talks as if he has a potato in his mouth

(adapted from Quirk, Greenbaum, Leech & Svartvik 1985:1110)

(2) It seemed as if / that he was trying to hide his true identity

(adapted from Huddleston & Pullum 2002:962)

The use of these complementizers, which are referred to as “minor” because of their low frequency, has been mentioned in the literature on complementation at various times, though mostly in passing (e.g., Lakoff 1968:69, note 7; Huddleston 1971:177-178; Warner 1982:180-185, 221-224; Mitchell 1985:15-17; Noonan 1985:104; Quirk, Greenbaum, Leech & Svartvik 1985:1175, note a; Fanego 1990:19-20; Bender & Flickinger 1999; Huddleston & Pullum 2002:962, 1151-1152). More recently, however, López-Couso and Méndez-Naya (2015) have addressed minor complementizers at length, to examine, on the one hand, the category of minor complementizers itself and, on the other, the history and use of individual clause connectives such as as if, as though, but, if, lest, like, and though (e.g., 1998, 2001, 2012; see also López-Couso 2007; Brook 2014).

The present investigation aims to contribute to this field of research by considering the emergence and loss of two additional clause connectives, namely till and until. Both are attested in complementizer function during Early Modern English (henceforth EModE) up to the Late Modern English period (henceforth LModE). They illustrate a transition from an original use as adverbial temporal subordinators of forward span (for this label, see Quirk, Greenbaum, Leech & Svartvik [1985:533-534]), as in (3), to a marginal, short-lived use as complementizers with the desiderative verb long, shown in (4). In (3), the till-clause is an adjunct, that is, a non-obligatory element in clause structure which identifies a point of time in the future up to which the proposition expressed by the matrix clause (i.e., He slepeth. . . Al nyght) is true (see further Kortmann 1997:183); in (4), by contrast, as will be argued later in this article, the till-clause and the coordinated that-clause which follows it are complements coding the object argument of the predicate long.

(3) He slepeth. . .Al nyght til the sonne gan aryse.

 ‘He sleeps. . . all night till the sun began to rise.’

 (MED, s.v. til, conj. 1a.(a); (c1390) Chaucer CT.ML.(Manly-Rickert) B.791)

(4) the one hath the spirit which saith come, the other shake at the mention of this comming: the one longeth till these shadowes flie away, and that day breake on them; the other can no more desire his comming, then the guiltie fellon can the comming and presence of the iudge.

(EEBO BYU; 1612, T. Taylor, A commentarie vpon the Epistle of S. Paul written to Titus)

On the basis of the data obtained from a number of sources detailed in section 5, the aims of the present study are the following:

•  (i) to observe the diachronic evolution of till/until-clauses occurring with long and argue for their complement status, in the light of a set of criteria for complementhood proposed in the literature (Huddleston & Pullum 2002:219-228; López-Couso & Méndez-Naya 2012:173-178, 2015:189-191);

• (ii) to discuss why till/until-complements emerged as a lexically specific phenomenon occurring only with long, but not with other verbs of desire, for example, desire, hunger, lust, thirst, or yearn (see Levin 1993:194-195);

• (iii) to discuss why till/until failed to survive as complementizers in the history of English.

As a broader implication, the study seeks to address the central question of how and why linguistic elements are lost (section 3), a topic which has not always been brought to the fore in diachronic linguistics, especially as compared to the attention given to the emergence and spread of new elements and new categories (Hundt 2014:167; Kranich & Breban 2021:1). Competition with systemic alternatives will be proposed as one of the factors motivating the loss of till/until-complements, which are seen to decline in the face of the continued growth of to- and for. . .to-infinitives as verb complements in the course of the LModE period (see De Smet 2013).

The remainder of this paper is organized as follows. Section 2 addresses the previous literature on the shift from adverbial subordinator to complementizer function and introduces a number of criteria for complementhood. Section 3 explains the notions of loss, competition, and marginality, which will be invoked as explanatory tools in later sections. Section 4 provides a preliminary overview, based on the existing literature and on lexicographic sources such as the Dictionary of Old English (DOE), the Middle English Dictionary (MED), or the Oxford English Dictionary (OED), of the historical development of the English subordinators of forward span, in particular till/until (section 4.1), and of the verb of desire long (section 4.2). Section 5 is concerned with the corpora and methodology used in the current investigation, while section 6 discusses the findings extracted from the data analysis. Lastly, section 7 summarizes the findings and offers some concluding remarks, with a special focus on the abovementioned research questions and broader implications.

2. The Historical Shift from Adverbial Subordinator to Complementizer

The change from adverbial subordinator to complementizer (henceforth A > C) examined in this article is a phenomenon which has been observed cross-linguistically, in English as well as in other languages, like Latin, Spanish, or Georgian (Hewitt 1987:219; López-Couso & Méndez-Naya 2015:195-196). As discussed by López-Couso and Méndez-Naya (2015:195) in relation to each of the clause links they investigate, this change has been seen to occur in a number of adverbial domains in the history of English, including those of comparison (as if, like, as though), exception or negative condition (but), condition (if), negative purpose (lest), and concession (though) (see also Kortmann 1997:79-89). In addition, López-Couso and Méndez-Naya (2015:193-196) have interpreted the change A > C as a case of secondary grammaticalization; as noted by Hopper and Traugott (2003:175), the traditional process of grammaticalization of lexical items can be extended to cases in which there is an “increased grammaticalization of already grammatical items in specific contexts” (see also Breban 2014:470). This is precisely what happens when adverbial clause links acquire the function of complementizer, since these already possess a grammatical function as adverbial subordinators which becomes even more grammatical when they are used as complementizers.

Researchers have put forward a set of criteria to investigate the complement status of clauses introduced by minor complementizers, and how they differ from adverbial clauses. Based mainly on the discussion in Huddleston and Pullum (2002:219-228) and López-Couso and Méndez-Naya (2012:173-178, 2015:189-191), the first and major criterion that is of relevance for complement status is licensing by a predicate (Criterion 1). It is a general property of complements that they need to be licensed, since every predicate projects certain participant roles. Thus, predicates meaning ‘fear, dread’ contain, as part of their argument structure, two participants: an experiencer and a participant that provokes fear. Discussing this kind of predicates in earlier English, López-Couso (2007) finds that these predicates are often followed by clauses with the conjunction lest (Old English, henceforth OE, þy læs (þe); Middle English, henceforth ME, the lesse the, thi les the, lest); this connective originally meant ‘so that not’ and introduced adverbial clauses of negative purpose. Sometimes, a twofold reading is possible, as in (5), containing the OE verb ondrædan ‘fear, be afraid.’ ¹

(5) Alyse me of  Esaues handa, mines broðor, for þam ðe ic hyne 

   deliver me from  Esau’s hands  my  brother because  I him

  swyðe ondræde,  þe læs ðe he  cume & ofslea  ðas modra mid

   much fear    lest    he  come & kill    the mothers with

  heora  cildum

  their  children

 (c1000, Genesis 32.11; cited from López-Couso 2007:22)

Here, on the one hand, the þe læs ðe-clause can be interpreted as a clause of negative purpose: ‘Deliver me from the hands of Esau, my brother, because I fear him very much, lest he come and kill the mothers with their children.’ On the other, the sequence could also be viewed as a type of appositional structure, attested in OE and ME (cf. Warner 1982:91-99), where a noun phrase (NP) (hyne ‘him’ in 5) was followed by a finite clause which specified further the respect in which what was said about the NP held true: ‘Deliver me from the hands of Esau, my brother, because I fear him very much, [I fear] that he may come and kill the mothers with their children.’ Eventually, the inference that the clauses following ondrædan ‘fear’ and similar predicates would explain what people were afraid of resulted in lest-clauses starting to be used exactly as though they were complement clauses, with lest functioning as a complementizer analogous to that, as in example (6) below (see also Los 2015:139-140).

(6) but bycause this texte of sayncte Paule is in latyn, and husbandes commonely can but lyttell laten, I fere leaste they can-not vnderstande it.

   ‘but because this text of Saint Paul is in Latin, and farmers in most cases know little Latin, I fear that they cannot understand it.’

   (Helsinki Corpus; 1534, A. Fitzherbert, The book of husbandry, 99; cited from López-Couso 2007:14, my translation)

The change in status from adverbial clause to complement clause can be put to the test by applying two other morphosyntactic criteria diagnostic of complementhood. One is whether the minor complementizer introducing a complement clause can be replaced by another complementizer “without any perceptible change of meaning” (Criterion 2; Huddleston & Pullum 2002:962; see also Quirk, Greenbaum, Leech & Svartvik 1985:1175, note a, 1184, note b); this is what happens, as pointed out by López-Couso and Méndez-Naya (2012:175, 184-189, 2015:190), in (7), first quoted in section 1 and repeated here for convenience.

(7) It seemed as if / that he was trying to hide his true identity

(adapted from Huddleston & Pullum 2002:962)

A third diagnostic concerns the ability of these clauses to occur in parallel with other complements (Criterion 3). Thus, since “coordination is a relation between two or more elements of syntactically equal status” (Huddleston & Pullum 2002:66), based on the fact that the as if-clause in (8) is coordinated with a that-complement clause, López-Couso and Méndez-Naya (2015:190) conclude that the as if-clause and the that-clause both perform the same syntactic function as complements of the verb feel.

(8) Now, driving the horse and sulky borrowed from Mynheer Schuyler, he felt as if every bone was topped by burning oil and that every muscle was ready to dissolve into jelly and leave his big body helpless and unable to move

(Brown Corpus; 1961, S. Longstreet, Eagles where I walk; cited from López-Couso & Méndez-Naya 2015:190)

I will return to the criteria for complementhood presented in this section in section 6.2.

3. Loss, Competition, and Marginality

In processes that are characterized as grammaticalization, loss is often considered as the last stage of development (Rehn 2021:161). Hopper and Traugott (2003:172) point out that “at the extreme end of the history of a particular form as a grammatical marker we might find loss, either of form alone or occasionally of both form and function.” Discussing the study of loss as a unified phenomenon, Kranich and Breban (2021:1) note that attention in diachronic linguistics has largely focused on the emergence and spread of new elements and new categories. Robust generalizations regarding paths, processes, and motivations have been formulated and tested with regard to the development of new grammatical elements out of lexical material and to the acquisition of new grammatical functions. However, as extensively as emergence has been researched, solid insights into the demise and eventual loss of grammatical elements and categories are still lacking. There are mechanisms, hypotheses, and generalizations that need to be formulated to further our systematic study of loss (Kranich & Breban 2021:2); there are also potential universals specific to loss which need to be discussed against the background of what we know about the emergence of new elements and categories (Kranich & Breban 2021:4, 9-10). An example is the universal principle of “Survival of the Frequent,” which states that “when a grammatical distinction is given up, it is the more frequent category that survives” (Haspelmath 2004:18).

One of the language-internal factors that has been identified as contributing to processes of loss is competition, which can be defined as “a situation in which two or more competing forms exist for the same function and one is eventually selected at the expense of the other(s)” (Rudnicka 2019:84; also Hopper & Traugott 2003:172; Hundt & Leech 2012:176). According to Hundt (2014:171), two prerequisites are to be considered for competition to play a role in an observed language change process: (a) constructions in competition have to be truly equivalent, that is, they have to compete “for the same linguistic environment,” and (b) a decrease in the frequency of the declining element is associated with a frequency rise of competing structures (see also Hundt & Leech 2012:176). Two cases of competition in the history of English investigated by Hundt and Leech (2012) are the decline of which as a relativizer and of for as a conjunction, at three equidistant periods in the recent history of American and British English. Their findings support the view that the demise of which and for is interrelated with the rise in frequency, respectively, of relative that and the conjunction because (Hundt & Leech 2012:183, 186).

Little attention has also been paid to what Joseph (1997) has referred to as marginal data, which involve those peripheral facts in a language that are highly particularized, being often keyed to the behavior of particular lexical items. Marginal data are frequently viewed as “the epitome of ‘anti-generalization’” (Joseph 1997:197). The assumption here is that, while the “core” of language represents the real basis for the emergence of linguistic knowledge and rules, many peripheral facts are nothing more than a list of exceptional behaviors. However, there is much of interest in studying linguistic marginalia because, as Joseph (1997:210) points out, “all linguistic knowledge starts out as marginal — only after enough bits of information have accumulated and the pieces begin to be fit into a system do broad ‘generalizations’ emerge.” Linguistic marginalia thus lead to crucial insights into the nature of language and linguistic competence: speakers have a considerable amount of knowledge that is highly particularized, and the domain of the specific forms the ground for the generalizations they make based on extensions over an ever wider set of particular facts. The margins, therefore, are deserving of a more prominent place in linguistic debate and analysis (Joseph 1997:209-210) and, as I argue in this paper, pose a valuable source of empirical data, not only for the study of particularized facts, but also for furthering our understanding of the inner workings and general principles of grammatical change and loss.

4. Historical Overview

Prior to the discussion of corpus findings in section 6, the present section offers a preliminary overview to provide a larger conceptual and historical backdrop for the study. I base this discussion on the existing literature and on lexicographic sources such as the DOE, MED, and OED. The historical development of the English subordinators of forward span, in particular till/until, is treated in 4.1, and of the verb of desire long in 4.2.

4.1. The Development of English Temporal Subordinators of Forward Span

As discussed by Kortmann (1997:345-348) and Rissanen (2011), among others, the system of adverbial connectives has been subject to considerable change over the history of English. I focus here on the category of subordinators which are the concern of the current analysis, that is, temporal subordinators of forward span referring to a point of time in the future. By far the most common item in OE was oþ (þæt/þe) ‘until’ (Mitchell 1985:II, 388-395; Traugott 1992:259-261; Rissanen 2007:64), illustrated in (9).

(9) Alecgað hit ðonne forhwæga on anre mile þone mæstan dæl fram þæm dwelling, [. . .] oþ þe hyt eall aled bið on þære anre mila

 ‘Then they lay the largest amount within one mile of the dwelling,

 [. . .] until it is all laid out within the one mile’

 (Or 1 1.20.30; cited from Traugott 1992:175-176)

Apart from oþ (þæt/þe), three other connectives are also found sporadically in OE and ME in uses that can roughly be paraphrased as ‘until,’ namely ere (cf. OED, s.v. ere, C.conj.), to (OED, s.v. to, †C.conj. 1.a.), and OE hwonne/hwænne ‘when’ (DOE, s.v. hwonne, hwænne, Adv. II.A.3.). The short-lived use of to as ‘until’ is illustrated in (10); as regards ere, in the vast majority of its occurrences it served as a conjunction of anteriority (‘before’), but mainly after negative matrix clauses it could sometimes shade into the meaning ‘until’ (cf. DOE, s.v. ær, conj. III.B.1.; MED, s.v. er, conj.(1) 2a.), as in (11).

(10) Þe kyng þere soiourned to he was hoole.

  ‘The king sojourned there until he was healed.’

  (MED, s.v. tō̆, conj. 1.(a); c1400(?a1300) KAlex.(LdMisc 622) 5893)

(11) ne cræwð se cocc ær ðu wiðsæcst me þriwa

  ‘the cock [will] not crow until you deny me three times’

  (DOE, s.v. ær, conj. III.B.1. Jn (WSCp) 13.38; cf. Io: non cantabit gallus donec me ter neges)

Hwonne/hwænne ‘when’ was an interrogative adverb in OE, and had not yet developed into the temporal conjunction it is today; it did so after having replaced the temporal conjunctions þa ‘when’ and þonne ‘when’ in the ME period (Mitchell 1985:II, 402-407; Fischer 1992:352-353). As an interrogative adverb, hwonne/hwænne was sometimes employed, as noted by the DOE (s.v. hwonne, hwænne, Adv. II.A.3.), to introduce a dependent interrogative clause after semantically appropriate heads like “verbs or verbal phrases referring to mental processes or emotion (i.e. of thinking, discerning, considering, heeding, observing, waiting, expecting, hoping, dreading, desiring, longing for, etc.).” In such cases, translators render hwonne/hwænne as ‘[a/the time] when’ or as ‘until,’ though, as Mitchell (1985:II, 403) points out, “hwonne can be explained as an interrogative in all the OE examples — in both native texts and those based on a Latin original.” Instances of this usage can be seen in (12) and (13), with the hwonne-clause following the OE verbs langian ‘to long’ and ondrædan ‘to fear.’

(12) [. . .] hæleð langode [. . .] hwonne hie of nearwe ofer nægledbord [. . .] stæppan mosten [. . .]

   ‘[. . .] the man longed [. . .] [for the time] when from durance they might step over the nail-fastened side [of the ark] [. . .]’

   (DOE, s.v. hwonne, hwænne, Adv. II.A.3.; GenA 1431; see also B&T, s.v. næglian, Verb I.)

(13) hie selfe wæron ælce dæg [. . .] ondrædinge hwonne hie on þa eorþan besuncene wurden

   ‘they themselves were every day [. . .] fearing [the time] when they would be sunk into the earth’

   (DOE, s.v. hwonne, hwænne, Adv. II.A.3.; Or 2 6.50.9)

From late OE, till, borrowed from Old Norse, begins to appear in northern texts, at first only in prepositional use, equivalent to to, and as such governing NPs (OED, s.v. till, A.prep.; also Rissanen 2007:69-70). Its earliest attested use as a temporal conjunction of forward span dates from early ME, illlustrated in (14). The synonymous until is also a borrowing from Old Norse, recorded from the fourteenth century (OED, s.v. until, B.conj.; Rissanen 2007:74).

(14) Þar he nam þe biscop..&..hise neues & dide ælle in prisun til hi iafen up here castles.

   ‘There he arrested the bishop.. &.. his nephews & put all in prison till they gave up their castles.’

   (OED, s.v. till, B.conj. 1.a.; 1154 Anglo-Saxon Chronicle (Laud) anno 1137; see also Mitchell 1985:II, 388)

As shown by Rissanen (2007:74), ME saw the quick and complete replacement of oþ by til as an adverbial subordinator, perhaps because til (CVC) was “phonetically weightier” than oþ (VC), which was, furthermore, formally very similar to the coordinator oþþe ‘or.’ With regard to the synonymous until, its frequency of usage relative to till has varied throughout its history (Rissanen 2005), being dialectally restricted during the ME period. From about 1800 its frequency has risen steadily (see OED, s.v. until, B.conj.), and in Present-day English (henceforth PDE) it has superseded the short form till, even in spoken language (Quirk, Greenbaum, Leech & Svartvik 1985:533-534).

4.2. Syntactic Patterning of the Verb of Desire Long in OE and ME

Long (< OE langian) is a native formation inherited from Germanic (see OED, s.v. long, v.¹ Origin). It forms part of the class of verbs of desire, as defined by Levin (1993:194-195), which includes items such as ache, crave, hope, hunger, lust, thirst, wish, and yearn; these constitute a syntactically coherent group insofar as they involve two arguments: a person that desires something (i.e., a desirer) and a thing that is desired (i.e., a desired; see further Castro-Chao 2021:70).

Prior to the discussion of findings in section 6 below, I provide here an overview of the phrasal and clausal patterns attested with long in OE and ME, based on the exhaustive analysis of four historical dictionaries (Bosworth & Toller’s An Anglo-Saxon Dictionary, B&T, DOE, MED, and OED) and their corresponding databases, and of the Penn-Helsinki Parsed Corpus of Middle English (PPCME2, circa 1.2 million words, 1150-1500; Kroch, Taylor & Santorini 2000). Also briefly discussed is the related collocation think (it) long (till/until) ‘to seem or appear long, to be wearisome (to a person) (until something happens)’ (OED, s.v. †think, v.¹ 3. and v.² 13.†b.); as argued in section 6, I believe this collocation has had a crucial role in the shift of till/until from adverbial subordinators to complementizers which is the concern of this article.

The various sources mentioned above reveal that long is found from OE in both personal and impersonal use, and with five different types of complements representing the desired argument: prepositional complement in example (15), NP complement in (16), “zero” complement in (17), a complement of direction, realized by an adverb or adverb phrase in (18) (in PDE a verb of motion needs to be supplied in order to render the translation grammatical), and a clausal complement in (19).

(15) [. . .] for eche man longeth aftur goode and þe laste goode and best in whych only man schulde reste is blys.

   ‘[. . .] for every man longs for good, and the greatest and best good on which man should only rest is heavenly bliss.’

   (PPCME2; c1400, CMWYCSER)

(16) Hom longeþ  tramtris  þe trewe [. . .]

   Him  longs  Tramtris  the truth

   ‘He, Tramtris, longs for the truth [. . .]’

   (MED, s.v. lō̆ngen, v.(1) 2.(e); c1330 (?a1300) Tristrem (Auch) 1275)

(17) This maide longede sore, And lovede him durneliche [. . .]

   ‘This young woman longed grievously, and loved him secretly [. . .]’

   (OED, s.v. long, v.¹ 6.†b.; c1300 Life & martyrdom Thomas Becket (Harl. 2277) (1845) l. 45)

(18) Me longeð heoneward.

   Me longs  hence

   ‘I long [to go] hence.’

   (OED, s.v. long, v.¹ †5.; c1225 (▶?c1200) St. Katherine (Royal) (1981) 876)

(19) Þe cwen..longede forte seo þis meiden [. . .]

   ‘the queen longed to see this young woman [. . .]’

   (PPCME2; c1225 ?c1200, CMKATHE)

In zero complement examples, the desired argument is syntactically unexpressed, because it can be inferred from the co(n)text (see Fillmore [1986:96] for a discussion of the pragmatic conditions controlling this type of elision in PDE). This can be seen in (17), where the identity of the desired can be retrieved from the pronoun him in the succeeding clause. ² This use of long corresponds to a pattern, now obsolete, which the OED classifies as syntactically intransitive, glossing it as ‘[t]o feel strong desire; to be restless or impatient with yearning’ (OED, s.v. long, v.¹ 6.†b.; also MED, s.v. lō̆ngen, v.(1) 1.(a) and 2.(a)). As may be observed in (17), in this usage long was often supported by intensifying adverbs, such as sore ‘grievously, with great distress’ (OED, s.v. sore, adv. 4.), swith ‘extremely’ (OED, s.v. swith, adv. †1.), or truly ‘loyally, constantly’ (OED, s.v. truly, adv. 1.b.), which served to complete the meaning of the verb and quantify the degree of the longing.

The clausal complements representing the desired argument seem to have been, in most cases, infinitive clauses, as in (19). However, very occasional examples of long governing a dependent clause introduced by the interrogative adverb hwonne/hwænne ‘when’ can be found in OE (see example 12, section 4.1), as also one ME example of long governing a clause introduced by that (as in 20). In addition, the historical data yield two ME instances, in (21) and (22), where long is followed by clauses introduced by the adverbial subordinators of forward span ere and till.

(20) Somdel hem longede [. . .] þt heuene cloue ofte atwo to sauy eiþer side.

   ‘They very much longed [. . .] that heaven repeatedly came apart in two to save each side.’

   (MED, s.v. lō̆ngen, v.(1) 2.(e); a1350 (?c1280) SLeg.Prol.CV (Ashm 43) 45)

(21) Þa longede swuðe Luces þon kinge ære he mare wuste of þan lauerd Criste.

   ‘Then Luces the king longed greatly until he knew more about the Lord Christ.’

  (MED, s.v. lō̆ngen, v.(1) 2.(a); c1275(?a1200) Lay.Brut (Clg A.9)10124)

(22) [. . .] this worthi kniht of Grece [. . .] Tok ore on honde and sore him longeth / Til he the water passed were. / Whan he cam to that yle there, / He set him on his knes doun strauht [. . .]

   ‘[. . .] this noble knight from Greece [. . .] took [the] oars in [his] hands, and longs him [i.e., he longs] grievously until he had crossed the sea. When he arrived at that island there, he immediately knelt down [. . .]’

   ((a1393) Gower CA (Frf 3) book 5, lines 3684 through 3690; see MED, s.v. lō̆ngen, v.(1) 2.(a))

Importantly for my analysis, examples (21) and (22) seem to be open to a double reading. On the one hand, they could be interpreted as instances of the pattern with a zero complement; the desired arguments are not explicit at the surface, because they can be inferred from the ere- and till-clauses that follow. In other words, the sentences would signal a feeling of yearning on the part of the desirers (King Luces and the noble knight, respectively) until the time when the states of affairs designated by the temporal clauses are achieved. Note also the use of the intensifying adverbs swuðe ‘greatly’ and sore ‘grievously’ quantifying the degree of longing, as in the intransitive example (17).

On the other hand, it could also be argued that, in (21) and (22), the specification of the time longed for has shaded into the specification of the event or situation wished for, such that the expression of a feeling of longing until something happens becomes equated with the longing and desire that something happens; in this second interpretation, the dependent clauses are not time clauses any more, but complement clauses, so that appropriate paraphrases for them would be, respectively: ‘King Luces longed greatly to know more about the Lord Christ’ and ‘he longs grievously to have crossed the sea.’

The historical record does not provide any other example involving the connective ere, seen in (21), but in the case of the connective till similar examples, dating from the sixteenth century onward, can be found in the OED and in my data; it is my hypothesis, discussed in the remainder of this article, that till-clauses were effectively mapped onto one of the participant roles of long (i.e., the desired) and served to complete its argument structure, thus leading to the interpretation of such clauses as complement rather than adverbial clauses. This process of change is discussed in detail in section 6.

Important in this connection is also the semantically related collocation think (it) long (till/until) ‘to seem or appear long, to be wearisome (to a person) (until something happens)’ (later, in personal use: ‘to grow weary with waiting, to be impatient (until something happens)’). It may have played a role, I will argue, in the shift of till/until from adverbial subordinators to complementizers (see OED, s.v. †think, v.¹ 3. and v.² 13.†b.; MED, s.v. lō̆ngen, adj.(1) 2a.(e); also OED, s.v. till, B.conj. 1.d.; MED, s.v. til, conj. 1a.(c)). This verbal phrase, where long is categorially an adjective, is attested as early as OE, as shown in (23).

(23) þincþ him tó lang hwænne hé beó genumen of ðyses lifes earfoþnyssum

    ‘it seems to him too long [to the time] when he shall be taken from the troubles of this life’

    (B&T, s.v. hwanne, Adverb II.; Homl. Th. i. 140, 9)

In the course of ME we find examples where think long is followed by an adverbial till-clause indicating forward span. The construction as a whole expresses ‘long duration of time or delay before the act or state expressed by the subordinate clause begins’ (OED, s.v. till, B.conj. 1.d.). One instance is (24), which antedates the earliest recorded use of the verb long patterning with a till-clause (see example 22). ³

(24) Alle him þenkeþ swiþe longe Til þou comest hem amonge.

    ‘It seems very long to all of them till you come among them.’

    (OED, s.v. till, B.conj. 1.d.; c1330 Assump. Virg. (B.M. MS.))

This historical overview has shown that in ME the verb of desire long could be found co-occurring with clauses introduced by the temporal connectives of forward span ere and till, in structures which appear to be halfway between adverbial subordination and complementation. Data for such usages, however, are very limited: as already noted, examples (21) and (22) are the only ones that I could trace in the extensive MED quotation database, while the PPCME2 (circa 1.2 million words) does not contain a single example of long patterning with either ere or till.

For the sixteenth and seventeenth centuries, however, the OED (s.v. long, v.¹ 6.†b.) provides several clear examples of long plus a till-clause. This suggests that in order to investigate the apparent shift of till from adverbial subordinator to complementizer which is the concern of this article, the relevant period for study is from EModE onward.

5. Material and Methodology

I first carried out a pilot study of till/until in the EModE period. For this purpose, I consulted a subset of about twenty million words extracted from the Early English Books Online Corpus 1.0 (EEBOCorp 1.0, circa 525 million words, 1470-1700; Petré 2013; on the compilation of the subcorpus, see Castro-Chao [2021:79-100]). The data obtained showed that the collocation long + till/until-clause occurs with very low frequency, as only four examples were found, the earliest dating from 1582 (Richard Mulcaster, The first part of the elementarie; see Table 1 for further details). The search in EEBOCorp 1.0 also served to confirm that till/until-clauses are attested only after the predicate long, but not after other verbs of desire such as desire, thirst, or yearn, a finding that was already suggested by the OED entry for till (B.conj. 1.d.).

Table 1.

Syntactic Patterns of Long in EModE (Tokens, Percentages within Parentheses, and Normalized Token Frequencies per Million Words; Adapted from Castro-Chao 2021:188)

Subperiod	Syntactic pattern										Total
	Prep		to-Inf		Zero		till-Clause		until-Clause
	T	Pmw	T	Pmw	T	Pmw	T	Pmw	T	Pmw	T	Pmw
S1 (1500-1550)	74 (54.81)	14.79	54 (40.00)	10.79	7 (5.19)	1.40	—	—	—	—	135 (100)	26.98
S2 (1550-1600)	42 (61.76)	8.40	24 (35.29)	4.80	1 (1.47)	0.20	—	—	1 (1.47)	0.20	68 (100)	13.61
S3 (1600-1650)	49 (67.12)	9.79	22 (30.14)	4.40	2 (2.74)	0.40	—	—	—	—	73 (100)	14.59
S4 (1650-1700)	43 (66.15)	8.72	17 (26.15)	3.45	2 (3.08)	0.41	3 (4.62)	0.61	—	—	65 (100)	13.19
Total	208 (61.00)	10.43	117 (34.31)	5.87	12 (3.52)	0.60	3 (0.88)	0.15	1 (0.29)	0.05	341 (100)	17.11

In light of the low frequency of long + till/until-clause in EEBOCorp 1.0, it became necessary to resort to the much larger Early English Books Online BYU (EEBO BYU, circa 755 million words, 1470-1700; Davies 2017). EEBO BYU contains a comprehensive collection of texts structured into decades and is very diverse in terms of genre, as it contains data from almost every work printed in English over a period of about 250 years. For the purposes of the present study, decades have been organized in circa eighty-year subperiods in order to make up a set of diachronically comparable subcorpora: 1470-1550 (27,498,390 words), 1550-1630 (227,775,908 words), and 1630-1700 (499,804,104 words). The searches were conducted using the lemmatized search strings “LONG TILL” and “LONG UNTIL.” These retrieve all forms and spellings of long followed by any form of till or until. As to the search options, these were set to 5000 hits (grouped by lemmas), a very high figure if we bear in mind that in EEBO BYU the number of hits set by default is one hundred. This procedure allowed me to retrieve a total of 1190 hits for till-clauses, seventy-one of which corresponded to long + till-clause; for until-clauses I retrieved a total of 311 tokens, eighteen of which corresponded to long + until-clause.

The dataset was then expanded to include also LModE and PDE. For this purpose, I used two additional corpora: (a) the Corpus of Late Modern English Texts, version 3.0 (CLMET3.0; De Smet, Diller & Tyrkkö 2013), a corpus comprising about thirty-four million words structured into three seventy-year subperiods (1710-1780, 1780-1850, 1850-1920), and covering five major genres (narrative fiction, narrative non-fiction, drama, letters, and treatise); and (b) the Corpus of Historical American English (COHA, circa 475 million words, 1820-2019; Davies 2010), covering also five major genres (TV/movies, fiction, magazine, newspaper, and non-fiction). The searches in CLMET3.0 were conducted with WordSmith Tools 8.0, 64 bits (Scott 2022) by means of the search strings “long* till” and “long* until,” which yielded only two instances of long + till-clause, both occurring in the first subperiod (1710-1780), while no instances were retrieved for until. ⁴ In COHA, the searching method I used is the same as the one adopted for EEBO BYU (as both databases share the same interface); my searches yielded no instances of long + till/until-clause, which confirmed the information already provided by CLMET3.0 that this collocation is not used beyond the eighteenth century.

Lastly, to explore the hypothesis that till/until-complements died out because of competition from to-infinitives and for. . .to-infinitives, data for these patterns were collected as well. For EModE, I consulted the twenty-million-word extract from EEBOCorp 1.0 as well as EEBO BYU. The pattern with to-infinitive complements was found to occur with notable frequency in EModE (see section 6.1), so I used only the twenty-million-word subcorpus from EEBOCorp 1.0. After manual sifting, this corpus yielded 117 instances of long + to-infinitive. By contrast, in the case of for. . .to-infinitives, which are infrequent in this period (see De Smet 2013:78-79, 90; also section 6.4.2), I employed the larger EEBO BYU; I used the lemmatized string “LONG FOR * * TO VERB,” which retrieves instances of the sequence long for followed by any constituent(s) intervening between for and a following to-infinitive. After filtering out the false positives, I was left with only one instance of the pattern long + for. . .to-infinitive in EModE.

For LModE, I made use of CLMET3.0 to retrieve data on to-infinitives—using the search string “long* to”—and on for. . .to-infinitives—using the string “long* for.” After manual sifting, the corpus yielded a total of 800 tokens of to-infinitives and nine tokens of for. . .to-infinitives.

6. Findings and Discussion

6.1. Syntactic Patterning of the Verb of Desire Long in EModE

Based on data drawn from the twenty-million-word subcorpus extracted from EEBOCorp 1.0 (see section 5), Table 1 provides an overview of the syntactic patterns attested with the verb long in EModE. The purpose of this information is to gain further insights into the range of complement types which may have entered into variation with the till/until-clauses which are the concern of this article.

As can be observed, two of the complement types attested in earlier periods of English (see section 4.2), namely noun phrase complements and complements of direction realized by adverbs, have disappeared in EModE. At this stage, prepositional phrases introduced by for (155 tokens; example 25) or after (fifty-three tokens; example 26) constitute the most frequent type of complement. This is followed by to-infinitive clauses (example 27) and zero complements (example 28). There are only four instances of till/until-clauses (example 29) that are ambiguous between an adverbial and a complement reading (see 6.2). Example (28) illustrates the intransitive use of long, now obsolete (OED, s.v. long, v.¹ 6.†b.; see 4.2); the antecedent (a pore woman with childe) of the relative pronoun which feels a desire for something that is syntactically unexpressed, but can be inferred to be meat (flesh), which is mentioned in the following coordinate clause.

(25) There is no doubt, but the sickeman for his health sake longeth for a medicine [. . .]

  (EEBOCorp 1.0; 1607, L. Ducci, Ars aulica or the courtiers arte)

(26) Let vs thinke vpon this with diligent mynde, let vs long after this with our whole hart [. . .]

   (EEBOCorp 1.0; 1574, St. Augustine, Certaine select prayers gathered out of S. Augustine’s meditations)

(27) And therfore wept he tenderly & longyd to lyue lenger [. . .]

   (EEBOCorp 1.0; 1529, T. More, The supplycacyon of soulys made by syr Thomas More)

(28) I knew a pore woman with childe which longed and beinge overcome of her passion eate flesh on a fredaye which thinge she durst not confesse in the space of xviij. yeres [. . .]

   ‘I knew a poor woman with child who felt a strong yearning [for meat], and being overcome by her desire, ate meat on a Friday, a thing which she did not dare to confess for eighteen years [. . .]’

   (EEBOCorp 1.0; 1528, W. Tyndale, The obedie[n]ce of a Christen man)

(29) And if he see a storm coming, he will make some haste to secure him before it falls; as the Husbandman hurries in his corn, when he sees danger of the rain. Thou long’st till thy corn be in the barn; And Christ longs till thou be in heaven. He is not compleat without thee [. . .]

   (EEBOCorp 1.0; 1668, R. Steele, The husbandmans calling shewing the excellencies, temptations, graces, duties)

From a diachronic perspective, prepositional patterns increase from 54.81 percent of uses in subperiod 1 to 66.15 percent in subperiod 4. To-infinitives, by contrast, decrease, dropping steadily from 40 percent in subperiod 1 to 26.15 percent in subperiod 4. Zero complements, for their part, remain below 6 percent of uses throughout the period.

6.2. Frequency and Status of till/until-Clauses with Long from 1470 Onward

The results for till- and until-clauses in EEBO BYU (1470-1700) and CLMET3.0 (1710-1920) are displayed in Tables 2 and 3. Two categories of clauses have been distinguished: complement clauses (five tokens), which can be identified as such on the basis of the morphosyntactic criteria for complementhood expounded in section 2, and indeterminate clauses (eighty-six tokens), which constitute the great majority of instances. The category “indeterminate” includes tokens where both an adverbial and a complement reading are possible; an example is (30).

(30) you have not met with moderate men, i am one that on occasion can spend, and then again spare, but pray you enter, i long till my wife and my son see you [. . .]

  (EEBO BYU; 1659, Aristophanes, The worlds idol)

Here, as was the case with examples (21) and (22) discussed in section 4.2 earlier, a double interpretation seems possible. Long could be employed in its intransitive sense of ‘[t]o feel strong desire; to be restless or impatient with yearning’ (OED, s.v. long, v.¹ 6.†b.); the till-clause would then be an optional adjunct, that is, an adverbial clause referring to a point of time in the future, so that an appropriate paraphrase for the sentence would be: ‘I am restless until my wife and son see you.’

Table 2.

Frequency of till-Clauses with Long from 1470 Onward (Tokens and Normalized Token Frequencies per Million Words within Parentheses)

Corpus	Subperiod	Complement	Indeterminate	Total
EEBO BYU	1470-1550	—	1 (0.04)	1 (0.04)
	1550-1630	3 (0.01)	26 (0.11)	29 (0.13)
	1630-1700	1 (0.00)	40 (0.08)	41 (0.08)
CLMET3.0	1710-1780	—	2 (0.19)	2 (0.19)
	1780-1850	—	—	—
	1850-1920	—	—	—
COHA	1820-1890	—	—	—
	1890-1960	—	—	—
	1960-2019	—	—	—
Total		4 (0.00)	69 (0.05)	73 (0.06)

Table 3.

Frequency of until-Clauses with Long from 1470 Onward (Tokens and Normalized Token Frequencies per Million Words within Parentheses)

Corpus	Subperiod	Complement	Indeterminate	Total
EEBO BYU	1470-1550	—	—	—
	1550-1630	—	8 (0.04)	8 (0.04)
	1630-1700	1 (0.00)	9 (0.02)	10 (0.02)
CLMET3.0	1710-1780	—	—	—
	1780-1850	—	—	—
	1850-1920	—	—	—
COHA	1820-1890	—	—	—
	1890-1960	—	—	—
	1960-2019	—	—	—
Total		1 (0.00)	17 (0.01)	18 (0.01)

A different interpretation is that the till-clause is mapped onto the second argument of long, and serves as a complement which specifies the event or situation wished for by the desirer; an appropriate paraphrase in this case would be: ‘I long for my wife and son to see you.’

In the case of a few of the examples I have counted as indeterminate, the interpretation of the till/until-clause as an adverbial clause of forward span appears to be somewhat preferable, because of the presence in the overall context of some kind of temporal specification. Such is the case, for example, of the temporal frame implicit in the noun day and the future-oriented verb group shal come in (31), which support a reading as ‘I am restless until that day shall come.’

(31) [. . .] i know while i am in this body i shal never be with the lord as i should be, but I long til that day shal come

   (EEBO BYU; 1661, S. Marshall, The vvorks of Mr Stephen Marshall)

Yet, even in cases like this, a complement reading (‘I long for that day to come’) cannot be ruled out, and the application of the two morphosyntactic criteria for complementhood outlined in section 2 fails to solve the ambiguity.

Yet despite this persistent ambiguity, the corpus data do offer evidence that the shift A > C, that is, the reinterpretation of an originally adverbial till/until-clause as a complement, effectively took place. Clear indications of such shift can be found in the ability of the till/until-clauses following long to occur in coordination with that-complement clauses in object function, as seen in (32) and (33) (see Criterion 3, section 2).

(32) the one hath the spirit which saith come, the other shake at the mention of this comming: the one longeth till these shadowes flie away, and that day breake on them; the other can no more desire his comming, then the guiltie fellon can the comming and presence of the iudge

  (EEBO BYU; 1612, T. Taylor, A commentarie vpon the Epistle of S. Paul written to Titus)

(33) they long for death, even as a hungry man longeth for any meate, or as a woman with child longeth for some speciall meate, as a fish longs for the baite, or as an angler longeth till the fish bites, or as a beleever (which as it is the most spirituall, so the most ardent desire of all) desires, to have any promise fulfilled [. . .]

   (EEBO BYU; 1643, J. Caryl, An exposition with practical observations)

In (33), the sequence as an angler longeth till the fish bites is coordinated with four long-clauses containing prepositional objects introduced by for, which supports a complement interpretation also for the till-clause itself. Note further the to-infinitive complement, dependent on the verb desire, which closes the sequence.

A roughly analogous example is (34), where the until-clause occurs in parallel to the prepositional object for his inheritance.

(34) let us therefore (as an holy divine would have us) spend many thoughts upon it [i.e., this most glorious place]; let us enter into deepe meditations of the inestimable glory of it: let us long untill we come to the fingering and possession of it: even as the heire longeth for his inheritance

    (EEBO BYU; 1632, R. Bolton, Mr. Boltons last and learned worke)

A second, albeit more indirect, piece of evidence also suggesting that till/until-clauses with long came to be parsed by language users as complements, rather than adverbial clauses of time, can be found in the fact that in the course of the eighteenth century they were lost, being replaced by infinitival complements (see 6.4).

The data from EEBO BYU and CLMET3.0 also reveal that till/until-clauses with the predicate long are not restricted to any specific register; examples can be found in a variety of genres, including religious prose (thirty tokens), drama (thirty tokens), poetic works of various kinds (eighteen tokens), or narrative non-fiction (five tokens), among others. Another finding is that the use of till/until-clauses with long, though first attested in late ME (4.2), occurs almost exclusively from the middle of the sixteenth century, and specifically between 1550 and 1700. Eighty-eight out of the total of ninety-one tokens of till/until-clauses following long recorded in the corpora (see Tables 2 and 3) are found in subperiods 2 (1550-1630) and 3 (1630-1700) of EEBO BYU; beyond those dates only two more tokens occur in subperiod 1 (1710-1780) of CLMET3.0, and then the usage disappears altogether. It can be said, therefore, that the collocation long + till/until-clause was very short-lived. Two questions arise out of these results: one relates to the factor or factors which may have prompted the sudden popularity of till/until-clauses in collocation with long; the other relates to the reasons for the quick obsolescence of that structural pattern. These issues are explored in sections 6.3 and 6.4 respectively.

6.3. The Extension of till/until-Clauses from the Adjective Long to the Predicate Long

Section 4.2 referred briefly to the lexicalized collocation think (it) long (till/until) ‘to seem or appear long, to be wearisome (to a person) (until something happens)’ (OED, s.v. †think, v.¹ 3. and v.² 13.†b.), and hinted at its potential role in the shift of till/until from adverbial subordinators to complementizers. This collocation consists of the verb think, the adjective long, and an adverbial clause of forward span introduced by till or until; a dummy, semantically empty pronoun it (see Huddleston & Pullum 2002:1481-1482) is also present in some instances. EModE examples of the construction are (35) and (36).

(35) so deepely did the kings words pierce their brests, that with a generall voyce, Reuenge they cry, now euery man the inuasiue Greeke detests, and thinkes it long, til they can greece defie, Soone after this, the king his nobles feasts [. . .]

   ‘the king’s words pierced their hearts so deeply, that with a collective voice they cry ‘Revenge!’; now every man detests the invading Greeks, and thinks it [a] long [time], until they can defy Greece. Soon after this, the king feasts his nobles [. . .]’

   (EEBO BYU; 1609, T. Heywood, Troia Britanica)

(36) this made me the more impatient; and so it was, that being invited thither to dinner again the next day, i thought long till the time was come, and had not (the next day) been long there, but told the master of the house, before much company, that were then present, what i had heard of the man [. . .]

   (EEBO BYU; 1672, M. Casaubon, A treatise proving spirits, witches, and supernatural operations)

Though the sequence think (it ) long till/until-clause is attested from ME (see 4.2), it only becomes frequent in the sixteenth century, as shown by Table 4; during the seventeenth century it continues in use, totaling 162 tokens in EEBO BYU, and then dies out, to judge from the fact that CLMET3.0 does not contain any instances. Chronologically, therefore, it exactly parallels the collocation long + till/until-clause, though the latter, with ninety-one tokens altogether, is comparatively less common (see Tables 2 and 3).

Table 4.

Frequency of Think (it) Long till/until-Clause 1470-1920 (Tokens and Normalized Token Frequencies per Million Words within Parentheses)

Corpus	Subperiod	N
EEBO BYU	1470-1550	2 (0.07)
	1550-1630	81 (0.36)
	1630-1700	79 (0.16)
CLMET3.0	1710-1780	—
	1780-1850	—
	1850-1920	—
Total		162 (0.21)

These findings suggest that the use of till/until-clauses with the predicate long may have been the result of a process of lexical diffusion from think (it ) long till/until-clause. In models of diachronic change such as Harris and Campbell (1995:97-115; see also De Smet 2013:46), the type of lexically governed change usually known as lexical diffusion is one of various linguistic changes falling under the broader rubric of extension, the analogical transference of a pattern to further related cases: “[o]bserved extensions generalize to a natural class based on categories already relevant to the sphere in which the rule applied before it was extended” (Harris & Campbell 1995:101). In other words, the comparatively frequent co-occurrence of think (it) long with adverbial till/until-clauses provided a favorable scenario for the extension and diffusion of such clauses to the verb long, which was formally and semantically related. In contrast, due to the absence of any formal resemblance to the adjective long, other verbs in the class of desire, such as desire, hunger, lust, thirst, and yearn, were not under the influence of the said pattern. This provides an answer, therefore, to one of the research questions formulated in section 1, relating to why till/until-complement clauses are attested with long, but not with other verbs essentially synonymous.

The evidence presented in section 6.2 (Tables 2 and 3) also shows that instances which receive a clear, unambiguous adverbial reading are never encountered in the data; this suggests that, once till/until-clauses had spread to long, a mechanism of conceptual integration may have come into play: since such clauses could potentially express what the desirer longed for, they could become integrated into the matrix clause by being used as though they were complements, capable of alternating with other kinds of complements of long, such as prepositional objects and that-clauses in object function (see, e.g., 32 and 33). As already explained, long projects two participant roles, so that in its intransitive usage the desired argument, though syntactically unexpressed, needs to be inferred from the co(n)text, as already discussed in 6.1 in connection with an example such as (37).

(37) I knew a pore woman with childe which longed and beinge overcome of her passion eate flesh on a fredaye which thinge she durst not confesse in the space of xviij. yeres [. . .]

    ‘I knew a poor woman with child who felt a strong yearning [for meat], and being overcome by her desire, ate meat on a Friday, a thing which she did not dare to confess for eighteen years [. . .]’

    (EEBOCorp 1.0; 1528, W. Tyndale, The obedie[n]ce of a Christen man)

When long started to be followed by till/until-clauses in such intransitive patterns, an inferential mechanism seems to have been triggered to the effect that, rather than the adverbial reading—‘to be restless until something happens’—there was a second meaning—the complement reading ‘to long for something to happen’—which came into play because it offered a fuller interpretation of the utterance concerned (see Heine 2002:84-86).

6.4. The Loss of till/until-Complement Clauses

Section 6.2 showed that till/until-clauses following long cease to be used in LModE (Tables 2 and 3). The hypothesis put forward in section 1 and explored here is that they may have been in competition with other complement types, which took up the functional space till/until-clauses had begun to occupy. Two potentially competing patterns, both attested with long, are to-infinitives and for. . .to-infinitives; the development of each of these patterns, in relation to till/until-clauses, is examined in detail in sections 6.4.1 and 6.4.2 respectively.

6.4.1. To-Infinitives versus till/until-Complements

As discussed in section 6.1 (Table 1), to-infinitive clauses figure prominently among the complementation patterns of long already in EModE, although with a diachronic tendency to decrease. Table 5 displays the LModE development of to-infinitives. As may be observed, the tendency to decrease observed in EModE is reversed in LModE, since to-infinitives witness a rise from 199 tokens in the first subperiod of LModE (18.99 pmw), to 276 in the second (24.46 pmw), and 325 in the third (25.75 pmw). Also, the overall normalized frequency increases quite considerably from EModE to LModE (5.87 versus 23.27, respectively; see Tables 1 and 5).

Table 5.

Frequency of to-Infinitives with Long 1710-1920 (Tokens and Normalized Token Frequencies per Million Words within Parentheses)

Corpus	Subperiod	N
CLMET3.0	1710-1780	199 (18.99)
	1780-1850	276 (24.46)
	1850-1920	325 (25.75)
Total		800 (23.27)

The EModE periods in which to-infinitives show the lowest frequencies—that is, EEBOCorp 1.0 subperiods 2 (1550-1600, 4.80 pmw), 3 (1600-1650, 4.40 pmw), and 4 (1650-1700, 3.45 pmw; see Table 1)—coincide chronologically with the periods when the collocation long + till/until-clause is most commonly found—that is, EEBO BYU subperiods 1550-1630 and 1630-1700 (see Tables 2 and 3). By the same token, to-infinitives are seen to rise during the time period when till-clauses disappear—that is, CLMET3.0 subperiod 1780-1850 (Table 2)—with normalized token frequencies shifting from 18.99 in the earliest subperiod of LModE (1710-1780) to 24.46 in the second (1780-1850). This implies that, from a diachronic perspective, the corpus data support the hypothesis that till/until-complements were in competition with to-infinitives; as Hundt’s (2014:171) view of competition proposes, a decrease in the frequency of a declining element—in this case, till/until-clauses—is associated with a frequency rise of competing structures—that is, to-infinitives (see section 3).

Such replacement might be interpreted as largely resulting from the operation of the process of “equi-deletion” (Noonan 1985:65-68). As is well known, “[e]qui-deletion (equi) deletes subjects of complements when they are coreferential with (i.e., refer to the same individual or thing as) some argument in the matrix” (Noonan 1985:66); its application yields reduced complements, such as to-infinitives. Thus, a sequence like *Zeke_i expects that Zeke_i will plant the corn is anomalous in PDE and has to be rephrased as Zeke expects to plant the corn.

Equi-deletion is a syntactic rule that applies solely to complementation structures, as in the example just mentioned. Adverbial clauses, which are adjuncts, fall outside its scope, hence the grammaticality of sentences like I_i have lived here since I_i moved from London or I_i did it because I_i wanted to, where the matrix and subordinate clause subjects are coreferential. The till/until-complements following long, since they originate in former adverbial clauses of time, often feature subjects which are coreferential with the matrix subjects, as in (38) and (39).

(38) Come lead the way, I long till I am there

  (EEBO BYU; 1594, C. Marlowe, The troublesome raigne and lamentable death of Edward the second)

(39) [. . .] that villanous [. . .], sir richard clare longs till he bee at the Nunry (EEBO BYU; 1608, Author unknown, The merry deuill of Edmonton)

Coreferentiality occurs in a total of thirty-six out of ninety-one tokens, or 39.56 percent, of the till/until-clauses in my data. This suggests that, as such clauses came to be parsed as complements by language users, the process of equi-deletion may have resulted in to-infinitives replacing till/until-complements in same-subject situations. In addition, this development can no doubt also be connected to the general tendency for finite complement clauses to be replaced by non-finite ones that has been at work in English in the course of history. As Denison (1998:257-258) points out, “[a] long-term trend in English has been the growth of nonfinite complement clauses at the expense of finite clauses,” a trend which continues, and becomes very noticeable, in LModE (see also Aarts, López-Couso & Méndez-Naya 2012:878).

6.4.2. For. . .to-Infinitives versus till/until-Complements

The rise of for. . .to-infinitives has been examined in depth by De Smet (2013). Formally, this pattern involves a structure in which a noun phrase introduced by for acts as the subject of a following to-infinitive with which it forms a constituent (De Smet 2013:73); two corpus examples are (40) and (41).

(40) Sometimes when I am rather tired I lean rather languishingly on a rock, and long for some famous Beauty to get down from her Palfrey in passing, approach me, [. . .], and give me - a dozen or two capital roastbeef Sandwiches.

   (CLMET3.0; 1816-20, J. Keats, Letters of John Keats to his family and friends)

(41) She welcomed back her old enchanted dream of life, and longed for the old dark door to close upon her [. . .]  (CLMET3.0; 1844, C. Dickens, Dombey and Son)

For. . .to-infinitives emerge in object position only at the beginning of the eighteenth century (De Smet 2013:78-79, 90; see also Visser 1973:2244-2248), and they spread to predicates such as ache, care, hope, long, look, pray, seek, thirst, or yearn, among others (De Smet 2013:87). This, according to De Smet, is connected to the strong overall tendency of for. . .to-infinitives to combine first with: (a) verbs which govern prepositional objects headed by for; (b) verbs which govern to-infinitives; and (c) verbs belonging to semantically coherent classes. For. . .to-infinitives spread to verbs governing for-phrases, as is the case with many verbs of desire. This is due, conceivably, to the formal link existing between the two complement types: the new infinitival pattern “would have been more easily accepted in environments that were already associated with a superficially similar sequence” (De Smet 2013:82). On a more general level, this pattern of diffusion is connected to the fact that morphosyntactic change is “sneaky” and advances most easily in environments where it is least noticeable, presumably supported by superficial resemblances to existing patterns (see De Smet 2013:51, 82).

Semantically, for. . .to-infinitives are associated with goal-oriented meaning (De Smet 2013:77, 88), as illustrated by (42) with the verb yearn. Here, “the predicate-complement construction expresses an activity or a state of mind that is directed to or conducive to the realization of the event denoted by the for. . .to-infinitive” (De Smet 2013:88). Along similar lines, example (43) shows that the clausal complement of long may also be interpreted as a goal (or metaphorical endpoint) of the desirer’s attention: that is, those who are diseased with a rheumatism direct their attention or state of mind toward the future realization of the event denoted by the for. . .to-infinitive clause (see Castro-Chao 2021:156-158).

(42) He has to talk to the country in terms which will make the voters yearn for him to replace John Major before another year is out.

  (Collins Cobuild Corpus; cited from De Smet 2013:88)

(43) as those who are diseased with a rheumatism, being worse in the night than the day, impatiently long for the rising sun to dispel the oppressing Humors, and cheer their spirits [. . .]  (EEBO BYU; 1684, W. Bates, The great duty of resignation to the divine will)

Table 6 shows the evolution of for. . .to-infinitives with long in EEBO BYU and CLMET3.0, as this allows us to determine whether the fate of till/until-clauses can indeed be connected with the rise of for. . .to-infinitives. If we compare Table 6 with Tables 2 and 3 for till/until-clauses (section 6.2), the following picture emerges: for EModE, there are eighty-eight instances of till/until-clauses versus only one for. . .to-infinitive. Conversely, for LModE there are only two instances of till-clauses, whereas the total of instances of for. . .to-infinitives amounts to nine. This suggests that from the second subperiod of LModE (1780-1850) onward for. . .to-infinitives begin to rise and replace till/until-clauses as complements to long. Notice that, in terms of normalized token frequencies, the rate of for. . .to-infinitives in the third subperiod of CLMET3.0 (1850-1920, 0.55) is higher than that of till/until-clauses in any of the subperiods of EEBO BYU or CLMET3.0—the highest frequency of till-clauses is in CLMET3.0, at 0.19 in subperiod 1710-1780 (Table 2). This is to be expected if we take into account that for. . .to-infinitives, in different functions, generally tended to grow in frequency in LModE (see De Smet 2013:79).

Table 6.

Frequency of for. . .to-Infinitives with Long 1470-1920 (Tokens and Normalized Token Frequencies per Million Words within Parentheses)

Corpus	Subperiod	N
EEBO BYU	1470-1550	—
	1550-1630	—
	1630-1700	1 (0.00)
CLMET3.0	1710-1780	—
	1780-1850	2 (0.18)
	1850-1920	7 (0.55)
Total		10 (0.01)

In 6.4.1, I invoked coreferentiality as a likely factor behind the replacement of till/until-clauses by to-infinitives. However, 60.44 percent (fifty-five tokens) of till/until-clauses following long do not exhibit an equi-relation to their matrix clauses, as in (44). It is in such different-subject contexts that till/until-complements could have found a functional niche to survive, but they did not, presumably because they existed in competition with for. . .to-infinitives and were eventually ousted by them, as is suggested by the data summarized above. Hundt’s (2014:171) prerequisite for competition, expounded in section 3, is that competing constructions have to be truly equivalent, that is, they have to compete “for the same linguistic environment.” For. . .to-infinitives and till/until-clauses fully comply with this requirement, in that they both allow the language user to overtly express the complement subject, and both constructions convey goal-oriented (or endpoint) meaning.

(44) I long till edward fall by Warres mischance, for mocking marriage with a dame of France.   (EEBO BYU; 1623, W. Shakespeare, Mr. William Shakespeares comedies, histories & tragedies)

7. Summary and Conclusion

This article has addressed the development of the adverbial subordinators till and until as minor complementizers introducing clauses governed by the desiderative predicate long. An analysis of data that were obtained from a number of sources has shown that the collocation long + till/until-clause is first attested in late ME (section 4.2), but does not become common until the sixteenth and seventeenth centuries (section 6.2). In LModE, however, till/until-clauses were lost, being replaced by competing patterns with to-infinitives and for. . .to-infinitives (section 6.4).

One of the factors which seems to have prompted the sudden, though short-lived, popularity of till/until-clauses with long is the relatively frequent occurrence, in the period under investigation, of the pattern think (it) long till/until-clause (section 6.3), which provided a favorable scenario for the diffusion of such clauses to the verb long itself, as it was formally and semantically related. In contrast, till/until-complements did not go on to spread to other verbs in the class of desire, thus failing to comply with the trend for grammaticalized elements to be generalized through new environments (e.g., Hopper & Traugott 2003:63-68). The diffusion of linguistic elements entails, contrary to processes of loss, “the gradual unidirectional expansion of a linguistic item over a new range of lexicogrammatical environments” (De Smet 2013:45). The case of till/until constitutes an exception to this general tendency in that the complement use does not extend to other environments, but rather remains confined to the margins of the English system of complementation. In terms of Himmelmann’s (2004:32) formulation of grammaticalization as a process of context expansion at different levels, till/until-clauses may be said to present a case of incipient grammaticalization: they acquire a new grammatical function in a specific environment, but the grammaticalized element does not establish itself because the class of items it “is in construction with, i.e. the host class” does not expand to include a larger number of verbs which are synonymous.

The case of till/until-complements thus provides a good example of what Joseph (1997) has referred to as marginal data (section 3). The common assumption is that the core aspects of language lend themselves to a greater degree of generalization, whereas marginal phenomena are only exceptional and tend to resist systematic investigation, being “often hard to integrate into broader, more far-reaching analyses” (Joseph 1997:197). The present study, however, has shown that the spread of till/until-clauses to long can be seen as a manifestation of the mechanism of change commonly known as lexical diffusion (section 6.3), and that it can be integrated in a broad trend leading to the emergence of new complementizers across languages (section 2). Till/until-complements are an unusual sort of phenomenon that is lexically particular to the extreme, but which is still captured by some of the mechanisms and pathways that govern processes of change in core aspects of grammar.

The decline of till/until, on the other hand, can also be approached and understood in terms of developments involving competition between variants and the eventual replacement of grammatical elements (e.g., Hundt & Leech 2012; Hundt 2014). It thus furthers our insights into the role of competition in processes of loss by exploring obsolescence in a lexically specific domain, allowing us to reconsider the generalizations and universals involved in it from the perspective of linguistic marginalia. Taking frequency of occurrence as a measure of the entrenchment of linguistic units and constructions (Langacker 1987:59), the newly-emerged, low-frequency till/until-complements had few chances to survive when facing the competition as verb complements of the more frequent to- and for. . .to-infinitives and their continued growth in the course of LModE; since frequency of exposure leads to greater memory strength, a frequent linguistic unit is remembered better (Bybee 1985:119). The present case thus provides evidence in favor of the principle Survival of the Frequent (section 3) as a potential universal which claims that “[w]hen a distinction is given up, only the most entrenched category survives” (Haspelmath 2004:19; see further Kranich & Breban 2021:9-10; cf. Kuo 2021). As Traugott notes (2014:17), “each [construction] has its own history within the constraints of larger patterns,” and constructional loss, like constructional emergence, involves interacting changes affecting constructions within a network (Kuo 2021:144-153).