The consolidation of þat as an invariable relativizer in the history of English

Two different invariable relative markers were in u se in early English, þe and þat/that. This paper aims to answer the question of how and  why þat replaced þe as an invariable relativizer in Middle English. To this end I analyse the distribution of invariable r lativizers in the relevant periods of the English language (from Old English to late Middle English) as represented in The Helsinki Corpus of English Texts. The following variables are examined: (i) the syn tactic function of the relativizer, which determines the progressio n and recession of relativization strategies, following the Accessibil ity Hierarchy proposed by Keenan and Comrie (1977); (ii) the type of relat ive clause (whether restrictive or non-restrictive), which conditions t he distribution of relativization strategies in particular; (iii) the type of NP antecedent, which also plays a role in the selection of relativ izer; (iv) the text type; and (v) dialect. The analysis reveals that þ /that starts to replace þe very slowly, occupying the environments less favour ed by þe, that is, those of object and resuming inanimate antecedents. Moreover, I will show that this slow, progressive introduction sudde nly evolves into a dramatic change, with þat quickly becoming the only invariable relativizer available.


Introduction
Different relativization strategies have coexisted over the course of the English language.These include the pronominal relativization strategy, the zero relativization strategy and the invariable relativization strategy.
The most frequently used of these has been the invariable relativizer, represented in early English (Old and early Middle English) by þe (example (1)), which from Middle English was wholly replaced by þat2 (example (2)), precursor of Present-day English that, as in (3), which was itself already in existence in Old English, but used with considerable less frequency.
(1) Gemunon we symle þaet we þa god don [rc þe us Godes bec laerað] Remind we ever that we the good things do rel us God's books teach "Let us be always mindful that we do those good things that God's books teach us" [Q O2/3 IR HOM BLICK 6: 139] (2) Morgan wolde haue hade alle þe lande fram bigende Humber, [rc þat Morgan would have had all the land from beyond Humber that Conedag helde] Coneday held 'Morgan wanted to have had all the land from beyond de River which Coneday held.' [Q M3 NN HIST BRUT3 21] (3) I am reading the book [rc that I've always wanted to read].
From Old English onwards, þat existed as an invariable relativizer but was very sporadically used as such.A common view is that relativizer þe was phonologically weaker than the already existing þat, and this led to its disappearance from the system earlier (Fischer et al. 2000: 91).Additionally, the displacement of þe by þat as the invariable relativizer has been linked to the use of þat as the general subordinator of the English language, already used in OE as a subordinating particle in non-relative dependent clauses (especially complement or content clauses and clauses of purpose and result) (Fischer 1992: 294;Rissanen 1997).
My purpose here is to shed light on the consolidation of that as the only invariable relativizer in English and to provide an answer to the question of how the consolidation of þat as an invariable relativizer proceeded.In order to answer this question, I report here on a study set up to investigate the use and evolution of þe and þat/that as invariable relativizers, from Old English to late Middle English (1420), when that had become the only invariable relativizer available in the English language.I will analyze the distribution of these relativizers, taking into account a number of linguistic factors (both internal and external) that have been reported in the literature to influence the choice of the relative word.The ultimate goal of this paper is to reconstruct the conditions under which þe came to be replaced by þat/that in the history of English.

Data and methods
The elements under examination here are invariable relativizers in Old and Middle English.During these periods, two relative words were used in this context, namely þe, as in (1) and þat as in (2), their distribution resulting from personal choice and/or linguistic factors which may have influenced the selection of one over the other.In this study I will consider the two elements as variants of the same relativization strategy, since both fulfil the requisites expected of invariant relativizers.
The 'invariable relativization strategy', also known as 'relative subordinator strategy' (Comrie 1981: 151, Givón 1993, Comrie 1998), represents one of the major relativization strategies typologically speaking, since it is among the most frequent means of marking relativization cross-linguistically. 3 Items used for this relativization strategy are characterized by being indeclinable, and not marked for gender, number or case.Unlike pronominal relativizers, they lack genitive forms (contra Seppänen and Kjellmer (1995) who report examples such as The dog that's leg was run over as examples of that explicitly marked for the genitive) and are not marked for animacy.Additionally, these elements are usually found as complementizers or subordinators in the language, as is the case with that and also þe in earlier English, the former used to introduce complement clauses throughout the history of the language and as a pleonastic marker of subordination up to early Modern English (Rissanen 1997) and the latter frequently used to mark complement clauses and subordinate clauses of purpose and result in Old English (Fischer 1992: 294).
The data used in the present study have been extracted from the Helsinki Corpus of English Texts.As Table 1 shows, the sample is taken from texts from Old English (950-1150) (O3, O4) and Middle English (1150-1420) (M1, M2 and M3).Recent studies on the distribution of relativizers in different British dialects (Herrman 2005: 24-28) confirm that this is still the norm, and is also becoming the trend in New Englishes, particularly in spoken language, as reported by Gut (2009) with respect to Kenyan, Jamaican and Singapore Englishes.
Table 3 presents the distribution of invariable relativizers in the corpus under analysis: it includes raw numbers and their normalized frequencies per one thousand words.relative clauses introduced by þat (the same tendency is also found sporadically in English earlier than this).By early Middle English, the frequency with which þat was used as an invariant relativizer had increased considerably (Kivimaa 1966: 133-136), an increase that was to become dramatic in late Middle English (1150-1250), when þat was by far the more important relativizer (McIntosh 1947(McIntosh -1948: 73;: 73;Kivimaa 1966: 134;Fischer 1992: 196;Fischer et al. 2000: 93).This sudden increase is depicted in Figure 2 (based on Table 3), which demonstrates graphically the progress of þat to become the most important relativizer from the second half of the thirteenth century on; and, similarly, how þe, in recession from the twelfth century onwards, became almost invisible.By the fifteenth century, þat was the only invariable relativizer.Here the use of an innovative invariable relativizer in the English language-in this case þat-grew in frequency and progressively pushed þe away, as demonstrated by the decline of þe in the 12 th century, being compensated by the increase of þat.As happened in the case of wh-relative pronouns, the use of which rose as that of the demonstrative relative pronouns receded (Suárez 2008a), the gap left here by the declining invariable relativizer þe started to be occupied by þat in early Middle English, eventually becoming the only invariant 6 M2 has been excluded from the graphs because the low number of examples-related to the scarcity of available material-hides the clear competition between the two relativizers (see footnote 4).relativizer by late Middle English.This ultimately led to the complete disappearance of þe as a relativizer and hence the prevailing state-ofaffairs in Modern Standard English.Around the thirteenth century, invariant þat was the norm everywhere, potentially occurring with any kind of antecedent and in any context, in most dialectal areas, with both restrictive and non-restrictive relative clauses, and with both animate and inanimate antecedents.The specialization of invariant that in restrictive relative clauses did not take place until around the fourteenth century, when wh-relative pronouns began to occur in the English language with some frequency (Mustanoja 1960: 197;Fischer 1992: 297;Fischer et al. 2000: 93).Before this consolidation, there existed variation between þe and þat; the factors which may have conditioned this variation will be addressed in the following section.

Syntactic function of the relativizer
The syntactic function of the relativizer has generally been considered one of the main factors governing the choice of relativizer.Syntactic function, in fact, is the variable most frequently mentioned in the literature on the expansion of wh-relativizers to occupy the space left by pronominal relativizers in earlier English (Suárez 2008a: 345-346) and may also be of importance in the case of invariant relativizers.For my analysis I have adopted (and adapted) Keenan and Comrie's 'Accessibility Hierarchy' (1977) so that three different categories are distinguished.The 'Accessibility Hierarchy' is a hierarchical ordering of noun phrase positions illustrated as follows (Keenan and Comrie 1977: 66): SU > DO > IO > OBL > GEN > OCOMP (where SU stands for ''subject'', DO for ''direct object'', IO for ''indirect object'', OBL for ''oblique'', GEN for ''genitive'' and OCOMP for ''object of comparison'') The syntactic positions included in the hierarchy are ordered from the most to the least accessible, that is, the easiest and hence most frequently relativized on the left, and the most difficult and less frequently relativized syntactic functions on the right.
For the present paper, the Accessibility Hierarchy was simplified and the following syntactic functions were distinguished: subject (S), as in example (4), object (examples (1), repeated here as ( 5)), and oblique (Obl), which refers to relativizers that function as prepositional phrases (all of which are stranded, as must be the case with invariable relativizers in the periods under analysis), as in example ( 6).The categories GEN and OCOMP were not included since no examples were found of invariable relativizers being used in either of these types of positions.
(4) se [ RC þe ne can þa beorhtnesse þaes ecan leohtes], se bið blind he rel neg knows the brightness of the eternal light he is blind 'he who does not knows the brightness of the eternal light is blind' [Q O2/3 IR HOM BLICK 2: 62] (5) The question of which positions in a language may be subjected to relativization is closely linked to the degree of explicitness of the relativization strategy.Relative clauses which are introduced by invariable relativizers allow a narrower range of positions to be relativized in comparison with those introduced by pronominal relativizers.With this in mind, we would expect invariable relativizers to be able to relativize only the most accessible positions in the hierarchy.Table 4 shows the distribution of invariable relativizers in the periods under analysis: The results show that, with respect to late Old English, the indeclinable relativizer þe is most frequently used when it appears as the subject or the object of the relative clause, as Traugott (1992: 226) and Kivimaa (1966: 28) have previously shown.In other words, the most frequently relativized positions are those which correspond to the roles highest on the 'Accessibility Hierarchy', as the behaviour of þe demonstrates.The same situation is observed in the distribution of þat (subjects and objects being the most commonly relativized positions), although the number of examples in this case is extremely low. 7The most notable aspect of þat's behaviour, however, is that the expected hierarchy is not respected: objects are as frequently relativized as subjects.In relation to early Middle English, Kivimaa (1966: 135) and Jack (1975 8 : 104; 19889 : 49) have observed that the invariable relativizer þe is used almost exclusively in the subject slot; such exclusivity was not observed in the case of þat.The correlation between the syntactic function of the relativizer and the choice of relativizer has also produced some interesting and significant results (χ 2 , p <.0001) in relation to my corpus of early Middle English (see Table 4).All syntactic roles are more frequently performed by the invariable relativizer þe, in particular, though not exclusively, that of the subject (just as Kivimaa and Jack had previously found).The most notable development in this period (1150-1250) involves þat whose importance as an invariable relativizer rises and which, although more commonly used as subject, also achieves an extremely high frequency as an object (in comparison with the other periods).
In its expansion þat is stronger-and ousts þe earlier-as object, a position in which þe is proportionally less frequent, as illustrated in O4 and M1, illustrating a similar process to that observed in the substitution of se pronominal relativizers by wh-words.The unavailability of relevant data from M2 noted above does not allow us to confirm this tentative hypothesis.

Restrictiveness of the relative clause
Another important variable affecting the distribution of relativizers is that of the restrictiveness of the relative clause, according to which relative clauses are divided into restrictive and non-restrictive types.The hypothesis being examined here is whether the distribution of invariable relativizers correlates significantly with the distinction between restrictive and non-restrictive relative clauses.The distribution of relativizers in Present-day English changes depending on this variable: restrictive relative clauses can be introduced by any of the available items, namely zero, that and wh-pronouns, and nonrestrictive relative clauses favour (indeed almost require) a wh-word; only very rarely are they introduced by the relativizer that (Jacobsson 1994;Huddleston andPullum 2002: 1059),10 and never by the relativizer zero.Throughout the history of the English language, invariable relativizers (as well as other relativization strategies) have been used to introduce both restrictive and non-restrictive relative clauses, though more commonly in relation to restrictive relative clauses.Scholars such as McIntosh (1947McIntosh ( -1948: 79 fn. 19: 79 fn. 19), Mitchell (1985 and Traugott (1992: 223) are not as categorical on this point but, broadly speaking, are of much the same opinion.Jack (1975Jack ( : 106-107, 1988: 52-53): 52-53), on the other hand, denies that the choice between the two major relativizers-þe and þat-is a function of the type of relative clause; nonetheless, his analysis still indicates a higher proportion of invariable þe in non-restrictive relative clauses, and the prevalence of þat in restrictive relative clauses (Jack 1988: 53).Table 5 sets out the distribution of relativizers according to the type of relative clause.The results show that the choice of þe does not relate to any type of relative clause in particular, but is simply a reflection of the general distribution of higher frequency in restrictive than in non-restrictive relative clauses across all the periods, and especially in late Old and early Middle English.During these periods of coexistence, þat exhibits a higher tendency to introduce restrictive relative clauses, and this is statisticially significant in Old English (χ 2 , p <.0001).However, from M2 onwards, when þat becomes the only invariable relativizer, it inherits the behaviour of þe in O4 and M1, so that the number of nonrestrictive relative clauses introduced by þat also increases.This new distribution leads us to the tentative conclusion that when þat substituted þe, it adopted þe's distribution in relation to the type of relative clause, thereby reinforcing the substitution process already observed in terms of syntactic function.

The NP antecedent
One of the most complex variables hypothesized to affect the distribution of relativizers is the NP antecedent.According to Jack, the preference of a relativizer, in particular þe and þat in early Middle English, is to be interpreted in light of the nature of the antecedent (Jack 1988: 53).The influence of the antecedent on the distribution of relativization strategies, however, is denied by Romaine, who states that "the effect of different types of antecedent (grouped according to certain characteristics or features of the modification which precedes the head noun) is in most cases negligible" (Romaine 1982: 143), at least in her investigation of relative clauses in Middle Scots (c.1530-1550).
In the current study, NP antecedents have been analyzed for the following two characteristics: • Category/form of antecedent • Type of antecedent: animate vs inanimate Regarding the category of the NP antecedent, the aim is to observe whether there exists any correlation between the category of the NP antecedent and the selection of þe and þat in the periods under analysis.
For the analysis of this variable I reduced the form of the NP antecedent to the following six categories: count nouns in the singular (example 7), count nouns in the plural (example 6), non-count nouns (example 8), pronominal items (example 4), proper names (example 9) and a miscellaneous category which would comprise every other antecedent (basically nominalized adjectives (example 5) and coordinate NPs (example 10).
( The bearing of the antecedent on the choice of relativizer in Old English has been studied by Kivimaa (1966), Mitchell (1985) and Traugott (1992), and among the few descriptive notes we find that the invariable relativizer þe is favoured when the antecedent is modified by a pronominal element (Kivimaa 1966: 44;Mitchell 1985Traugott 1992: 226), especially in the singular (Traugott 1992: 226) (with the exception of the indefinite pronoun eall, which when modified by a relative clause, is invariably introduced by þat (see Mitchell 1985).The distribution of relativizers þe and þat according to the antecedent they resume is shown in Table 6: The results in Table 6 show those differences in the two periods of Old English and early Middle English (M1).In O3 (950-1050), although invariable relativizer þe is used in all environments, it is clearly favoured when it resumes pronominal items and count nouns in the plural.Relativizer þat is preferred in combination with count nouns in the singular (either masculine or feminine), 11 but we must bear in mind that the low number of examples here does not allow us for definite conclusions.Regarding O4 (1050-1150), relevant differences are also observed (χ 2 , p <.0001).Þe is favoured with count nouns, both in the singular and in the plural, and is much less frequently used with pronominal items, which becomes the favourite NP antecedent of the emergent þat.This is reinforced by the results shown for þat in M2 (1250-1350).Regarding M1 (1150-1250), a very similar distribution is observed in þe and þat, both favoured with count nouns in the singular and with pronouns, thus mirroring the general distribution of the category of antecedents in the whole corpus.The distribution in this case is also significant (χ 2 , p <.0001).
In terms of the animacy of the antecedent, in the codification of the results from my corpus, 'animate' (example 9) and 'inanimate' (example 7) antecedents were distinguished.
There is no consensus in the literature as to the influence of the animacy of the NP antecedent on the choice of relativizer.As noted above, from the results of her analysis of relativizers in Middle Scots (c.1530-1550), Romaine concludes that the influence of the antecedent "is in most cases negligible " (1982: 143), and, more categorically, that "the animacy of the antecedent has virtually no effect in determining which form of the relative will occur" (1982: 142).The results from Table 7 demonstrate that at least some relativizers are sensitive to the animacy of the antecedent.The results show that þat is preferred with inanimate antecedents both in Old English and early Middle English; less frequently, it is also found resuming animate ones in O4 and M2 (the latter somehow reflecting the general distribution of animate vs inanimate antecedents).By contrast, no such clear preference regarding the animacy of the antecedent has been observed with invariable þe.It is used as frequently with animate as with inanimate antecedents, contra McIntosh (1947-48: 74), Kivimaa (1966: 135) and Jack (1975: 101;1988: 58), who all observed a preference for þe to be used with animate antecedents.The distribution in M1, the most representative period, confirms that, when relativizer þat becomes the favoured relativizer, it tends to be used with animate antecedents, occupying the space dominated by þe (Fischer 1992: 295), and also behaving as the only invariable relativizer used with both animate and inanimate antecedents.
Figures 5 and 6 illustrate the findings discussed.Invariable relativizer þe shows a clear preference for combining with animate antecedents, while þat favours inanimate ones.This could be related to the fact that the pronominal relativizer þat in Old English was used with nouns marked for the neuter gender, and these are very frequently inanimate. 12igures from Table 7 above confirm that the choice of þe and þat is sensitive to the animacy of the antecedent, both in Old English and in early Middle English (χ 2 , p <.0001): þe is more frequently used with animate antecedents than with inanimate ones and þat is more frequently used with inanimate antecedents than with animate nouns.However, when þat becomes the only invariable relativizer it is used very frequently both with animate and inanimate antecedents, adopting the roles covered earlier by þe.
In sum, interesting tendencies emerge from the analysis of the NP antecedent.In Old English, the influence of the antecedent on the relativizer's choice has to do with the form.It is concluded that the use of the invariable relativizer þe is favoured when the antecedent is pronominal or singular.In early Middle English, animacy seems to be the most relevant factor.On these grounds we can conclude that þe is mostly found with animate and inanimate plural antecedents and þat with inanimate antecedents and some very specific animate nouns.

Position of the relative clause
The last intralinguistic variable analyzed to explain the distribution of invariable relativizers is the position of the relative clause with respect to the antecedent.Unlike Present-day English, this variable is of relevance in earlier English because there was a strong tendency in the language to avoid non-clause-final embedded structures (Carkeet 1976).Therefore, relative clauses were usually separated from the antecedent they depend on.
Relative clauses have been coded as 'intraposed' (if the antecedent and the relativizer were adjoined), as illustrated by example (11), or 'non-intraposed' (if the antecedent and the relativizer were separate), represented by so-called 'extraposed' (example 5) and 'left-dislocated' (example 4) relative clauses.An important aspect that needs to be clarified in relation to this variable is that it is only relevant for a reduced number of examples, that is, only those instances in which the separation of the antecedent is possible.Instances such as (2) and ( 7), in which the entire NP antecedent is clause-final, were excluded from the analysis.
(11) Forþon þaere burge nama [rc þe is nemned Gerusalem] is gereht sibbe Because the city name rel is called Gerusalem means of peace gesyhþ, forþon þe halige sawla þaer restaeþ.sight because holy soul there rest 'For the name of the city which is called Jerusalem signifies 'sight of peace', because the holy souls rest there.' [Q O2/3 IR HOM BLICK 6: 25] The results are included in Table 8: The data from Table 8 show that, with the exception of O4 (1050-1150), both þe and þat show a strong preference for adjacency, which might lend support to the hypothesis that þat enters the language to occupy the slots left empty by þe. 13 The data of M3 (1350-1420) are irrelevant not only because it is the period when þat consolidates as the only invariable relativizer, but also because by this time the language had already fixed the word-order and intraposition was the norm in the case of adnominal relative clauses.

Type of text
In addition to the intralinguistic variables, two extralinguistic variables were analysed, namely the type of text and the dialect.Although four types of texts were initially included (medicine handbooks, history, religious treatises and homilies), in the end only religious treatises, homiletic texts 14 and history were analysed, since these text types were available for all the different periods covered by the corpus.The relevance of this variable lies in the fact that the choice of relativizer can be stylistically determined, as demonstrated by Romaine (1982) in her study of relative markers in Middle Scots.Similarly, in the present study the label 'style' will be used to compare the distribution of the invariable relativizer used to introduce the relative clause.Taking into account that þe is earlier than þat, it is then expected that, when coexisting, a higher presence of þe would render a more conservative text type and, on the contrary, a higher presence of þat a more innovative text type.The question under investigation here is whether the text types show a different distribution of invariant relativizers.Table 9 includes the distribution of þe and þat in religious treatises, history and homiletic texts in the different periods covered in the present study.The comparison of different text types in Table 9 shows a significant correlation between the choice of invariable relativizer and text type, especially in early Middle English (χ 2 , p <.005).The general trend shows a decrease of relative clauses introduced by þe from Old English to Middle English in all text types.In Old English, þat is clearly disfavoured.Diachronically, however, history shows a different distribution from religious treatises and homiletic texts, in that it evinces an important earlier increase of þat.In homiletic texts, the consolidation of þat takes place later, indicating that these texts are more conservative texts in nature than in the case of history texts.

Dialect
The final variable included in the analysis is the dialect.Since Old English texts are mostly in the West-Saxon dialect, texts are very homogeneous in terms of dialect, making an analysis of variation pointless.In Middle English, however, geographic diversity is the norm and thus it is from this period that dialectal variation was analyzed (1150-1250), and indeed the significance of this variable will be seen clearly in Middle English.Although five main dialects are distinguished in Middle English, the most revealing regional indicators group them into two macro-dialects: (i) Northern, which comprises the inheritors of Old English Northumbrian and Mercian, namely, Middle English Northern, East-Midland, and West-Midland; and (ii) Southern, which comprises the descendants of Old English West-Saxon and Kentish, namely Middle English Southwestern and Southeastern (or Kentish).Table 10 shows the distribution of invariant relativizers in early Middle English: The distribution of invariable relativizers in early Middle English is highly significant (χ 2 , p <.0001).Invariable þe, the favourite relativizer in late Old English and in early Middle English, turns out to be preferred in the South and Kent, the areas which were less affected by the Scandinavian invasions, and therefore more conservative linguistically speaking (Milroy 1992: 181).More surprisingly, it is also favoured in the East-Midlands, showing an even higher frequency than in the other two areas.This contradicts Kivimaa's claim that this relativizer levelled out first in this dialectal area (and the North).
Taking into account that the southern dialect is the direct inheritor of Old English West-Saxon, the overuse of þe looks like a conscious strategy by the scribes in order to preserve this early form and resist the loss of a tradition which had already been lost in other areas.Much more difficult is to account for the high frequency of this relativizer in the East-Midlands, a linguistically advanced area which, accordingly, should integrate the innovations and abandon the old traditions earlier.
The West-Midlands warrants special notice, for it is here that invariable þat has become the preferred relativizer, being even more frequent than invariable þe, and thus becoming an innovation.As mentioned before, this invariable relativizer was already present in late Old English, but it was very rarely used.It is important to remember that invariable þat scores higher in all dialects from early Middle English than in Old English, but the West-Midlands dialect warrants special notice, for it is here that invariable þat has become the preferred relativizer, more frequent even than invariable þe, and thus becoming an innovative feature.

Summary and conclusions
Invariable relativizer þe was the predominant relativizer in late Old and early Middle English, but was replaced by þat in Middle English.Early Middle English marks the turning point in the status of þat, as its use expanded into a growing number of contexts, irrespective of its agreement with the antecedent.Emerging in late Old English, its frequency then rose slightly in early Middle English, before increasing still further until, by Middle English, it had become the only remaining invariable relativizer (13 th century).In the 13 th century, there is a (weak) tendency for þat to be used more frequently as subject and object, introducing intraposed restrictive relative clauses and resuming inanimate antecedents, mostly realized by singular count nouns in the West-Midlands and in history writing, that is, in most of the environments where þe is less prevalent.However, the notion of a complementary distribution is unsustainable; instead, þat's emergence is one of almost immediate substitution (from c. 1150, where þe was favoured, to c. 1200, where þat was almost the only choice).By the end, it had become possible for þat to be used with any kind of antecedent and in any kind of context.In response to the question posed at the beginning of this paper, then, it can be concluded that that spreads by occupying the gaps occupied by þe, and as a consequence finally ousts it from the system of the invariable relativizers.Why?The weakening and disappearance of inflectional markers which took place in the English language from late Old English onwards and the progressive levelling of declensions which was completed by Middle English may well account for the sudden nature of the substitution.The overuse of þe probably favoured the introduction of, or rather the preference for, an already existing, if yet infrequent, element in the paradigm.Though initially occupying the environments less favoured by þe, þat rapidly became the only choice in all environments.The replacement of þe, likewise, coincides to some extent with the period during which þat was gaining ground as the general subordinator.The critical period of analysis in this regard is probably M2, but with so little material available from early Middle English, such hypotheses are no more than tentative.

Figure 2 .
Figure 2. Competition between þe and þat in the history of English (normalized frequencies per 1000 words) 6

Table 1 .
Description of the corpus The last period included in the analysis is late Middle English (M3), since by then (1350-1420) the relativizer þat/that was the only remaining invariable relativizer in use, thus removing the need to examine texts dating from any later than this time.In order to keep the corpus as homogeneous as possible, only original prose texts have been selected.Translations (as classified in the Helsinki Corpus) were discarded, so as to avoid any potential influences from language contact. 5same selection criterion, however, had the additional effect of upsetting the balance between sub-samples, a problem which I have corrected by normalizing the frequencies per one-thousand words.Invariable relativizers have always been present in the English language and have in general been used more frequently than any other relativization strategy, as Figure1below (based on Table2) illustrates.

Table 5 .
Invariable relativizers and type of relative clause (normalized frequencies per 1000 words)

Table 6 .
Invariable relativizers and category of NP antecedent (normalized frequencies per 1000 words)

Table 7 .
Invariable relativizers and animacy of the antecedent (normalized frequencies per 1000 words)