Where did wer go? Lexical variation and change in third-person male adult noun referents in Old and Middle English

Abstract The present study uses variationist quantitative methods to examine the evolution of the semantic field of third-person male adult noun referents from Old English to Middle English, covering a time depth of approximately six hundred years. Results show a shift from the favored variant wer in Old English to man in Middle English, with the diachronic change in frequency following a prototypical s-shaped distribution. Although the replacement seems to take centuries to be complete, lexical frequency and written transmission are proposed as influential explanatory factors, and a homonymic clash is suggested to have accelerated the process of replacement in Middle English. Text type and text origin contribute to variation, with alliteration significantly influencing lexical choices in Old English verse texts. When combined with findings from recent synchronic work, this study highlights a heterogeneously structured semantic domain, which has undergone lexical replacement and change over time, providing some evidence for the applicability of s-shaped patterns for lexical change.

While this semantic field has been explored in Present Day English (Franco & Tagliamonte, 2021;Tagliamonte, 2022) and a list of variants pertaining to this semantic field has been compiled for Old and Middle English (Elsweiler, 2011;Grygiel, 2006;Kleparski, 2003Kleparski, , 2005Stenroos, 2002), how this domain has evolved over time, especially in the earlier stages of the English language, remains unclear. One of the reported hallmarks of linguistic change is the presence of an s-shaped distribution where incoming forms are adopted at a rate of slow to fast to slow (Bailey, 1973:77;Weinreich, Labov, & Herzog, 1968:113-114). A comparison of the current system of variants (e.g., man, guy) with variants used in earlier stages of the language (e.g., wer 'man') clearly shows lexical change within this semantic field. However, whether lexical change follows a traditional s-curve trajectory remains to be investigated. Although lexical replacement is expected to follow an s-curve distribution (Blythe & Croft, 2012:278-279;Chambers, 2002:361), to date s-curve patterns have been modeled predominantly on phonetic (e.g., Labov, 1994), morphological (e.g., Nevalainen, 2015), and discourse-pragmatic features (e.g., Tagliamonte & Smith, 2021). While s-curve patterns for lexical change can be found in the literature (e.g., Chambers, 1995), they are typically based on apparent time as opposed to real time data. However, in work on short-term high density lexical change, Grieve, Nini, and Guo (2017) found s-shaped patterns for several "emerging words" such as baeless and fleek, suggesting that lexical change also follows a prototypical s-curve trajectory. Nevertheless, research on long-term s-curve patterns for lexical change is lacking. Given that the semantic field of third-person male adult noun referents is rife with variation, examining how this system has evolved holds promise for insights into the factors governing lexical variation and change in real time. Recent work on Ontario English shows that this lexical domain is in flux, with factors such as gender and age influencing speakers' choices (Franco & Tagliamonte, 2021;Tagliamonte, 2022). Against the backdrop of this work, the present study examines the system of third-person male adult noun referents in earlier stages of the English language, specifically in Old and Middle English.
Two research questions are addressed. First, what was the distribution of thirdperson male adult noun referents in Old and Middle English? A quantitative comparison of this domain over time can document changes within this semantic field while also providing a platform to test patterns of lexical change. Second, based on the extant metadata, is there any evidence to suggest that the use of third-person male adult noun referents was conditioned, constrained, or influenced by any attested factors of variation? To answer these questions, the Helsinki Corpus of English Texts (Kytö, 1996;Rissanen, Kytö, Kahlas-Tarkka, Kilpiö, Nevanlinna, Taavitsainen, Nevalainen, & Raumolin-Brunberg, 1991) was used as the principal source of linguistic data, as it contains texts from Old English and Middle English, as well as metadata for potentially influential conditioning factors, such as TEXT TYPE and TEXT ORIGIN. Although larger corpora are available, given the size of the system of third-person male adult noun referents, with over ten thousand tokens of man attested in the Helsinki Corpus of English Texts alone (Rauer, 2017:142-143), to ensure confidence in the circumscription of the variable context, larger corpora were avoided.
Based on the premise that lexical choices are influenced by similar conditioning factors that operate on grammatical and phonological variation, Tagliamonte and colleagues probed the sociolinguistic underpinnings of the system of third-person male adult noun referents in Ontario English (Franco & Tagliamonte, 2021;Tagliamonte, 2022). The authors found that guy was making traction over competing variants such as man, a change led predominantly by young men. While guy was used more frequently by men than women, other factors such as socioeconomic status were also found to play a role. Following the "Uniformitarian Principle" (Lyell, 1830(Lyell, -1833cf., Labov, 1972:275), one might expect similar conditioning factors to have affected this system diachronically. Although the lack of similar types of sociolinguistic metadata in the earlier stages of the English language rules out a quantitative analysis of the effect of social factors on this system in Old and Middle English, this scholarship serves as a point of departure for the present analysis.

History of third-person male referents
Variants for describing third-person male adult noun referents have a long history of lexical replacement in the English language. According to A Thesaurus of Old English (2017) and The Historical Thesaurus of English (2023), there are at least twenty-five attested lexical items that denote "man" in Old English, including, but not limited to ceorl, carlmon, esne, freca, folcagende, folcwer, guma, gumrinc, haele[þ], hyse, leod, maecg, man, scealc, wer, woruldman, wiga, waepnedmann, and waepenmann. While some of these variants (e.g., gumrinc) are reportedly more frequent in Old English verse (A Thesaurus of Old English, "13.02.10.01 A man, warrior"), others (e.g., wer, guma, man) are found in both prose and verse (Kleparski, 2003:49;Stenroos, 2002:382-383). Wer, guma, and man are variants that can refer to a male adult in Old English, but the word form man has additional functions: it can be used genderinclusively to refer to both male and female referents (Curzan, 2003:64-65), it can be used to refer to human beings (Rauer, 2017:139-140), and it can be used as an indefinite pronoun, corresponding in translation to "one" (Raumolin-Brunberg & Kahlas-Tarkka, 1997). 1 Examples of wer, guma, and man referring to male referents are provided in (2). 2 (2) a. Ond on ðone ylcan daeg Crist gereorde fif ðusenda wera of fif and on that same day Christ fed five thousand man-GEN.PL of five hlafum ond of twam fixum, eac wifum ond cildum bread-DAT.PL and of two fish-DAT.PL, also woman-DAT.PL and child-DAT.PL þara waes ungerim which.GEN.PL was uncountable 'And on that same day, Christ fed five thousand men, with five loaves of bread and two fish. In addition, he also fed women and children, of which there were many' [Old English Martyrology, 950-1050] b. ðonne onwaecneð eft wineleas guma then awakens again friendless man 'Then the man without any friends woke up' [The Wanderer, 950-1050] c. on þaere fyrde waeron þe ferdon fram Egipte, sixhund þusend in the army were REL traveled from Egypt six-hundred thousand manna butan wifum 7 cildum man-GEN.PL except woman-DAT.PL and child-DAT.PL 'In that army, there were 600,000 men who travelled from Egypt, and that number does not include women and children' [AElfric's Letter to Sigeweard, 1050-1150 By Middle English, only half of the Germanic words for "man" (e.g., beorn, cerl, freca, guma, hearra, leod, man, rinc, scealc, secg, waepenmann, wer, wiga) are reported to have remained in use (Stenroos, 2002:385). Contact with Anglo-Norman led to the emergence of new variants through lexical borrowing, such as sire 'man' (Kleparski, 2005:48) and sergeant 'servant/serving individual' (Kleparski, 2003:51). Modern English gentleman emerged during Middle English by compounding the French loanword gentil 'noble' with Germanic mon 'man' (literally 'nobleman'). Meanwhile, other variants underwent semantic shifts, such as Old English aeþeling 'prince/nobleman' which became Middle English hathel 'man' (Middle English Dictionary [MED] 2021; hathel, n.). However, like in Old English, it is possible that many variants were affected by text type, with some variants occurring more frequently in verse than in prose due to metrical and alliterative demands. According to Eduard Siever's Altgermanische Metrik (1893), lines in Old English verse, called Langzeile, consist of two short lines (Kurzzeile), and at least one lexical item in each short line alliterates. This alliterative tradition was transmitted into Middle English to a lesser extent, with rhyme emerging as a new feature of verse. Examples in (3) from the Middle English text Sir Gawain and the Green Knight illustrate the rich variation present within the semantic field, seemingly due to the alliterative demands. According to the Oxford English Dictionary (OED, 2021), few of the Middle English variants survived into Early Modern English. For instance, tulk (ON tulk-r 'interpreter') and renk (ON rekk-r 'warrior,' OS rink) were rarely used after the mid-sixteenth century, and gome (OE guma), freke (OE freca), berne (OE beorn 'man of valor,' ON bjo rn 'bear'), schalk (OE scealc 'servant/man,' ON skalk-r 'slave/servant'), and lede (OE leoda 'people,' German Leute 'people') are not attested after the seventeenth century. 4 (3) a. and talk wyth þat ilk tulk þe tale þat me lyste and talk with that same man which tale that me desired 'and talk with the same man about whatever tale is pleasing to me' b. forþy goude Sir Gawayn let þe gome one therefore good Sir Gawain let the man alone 'so, good Sir Gawain, leave the man alone' c. for he is a mon methles and mercy non vses for he is a man measureless and mercy none use 'because he is a violent man and not merciful' d. and ȝe ar a lede vpon lyue þat I wel louy and you are a man upon life that I well love 'and you are a mortal man that I love very much' e. Þe burne þat rod hym by bede his mayster abide the warrior that rode him by commanded his master abide 'The man who rode past him, commanded his master to wait' f. here ar no renkes vs to rydde, rele as vus likez here are no men us to ride, reel as us pleases 'There are no men here to stop us from fighting, as it pleases us' g. Þat oþer schalk wyth a schunt þe schene wythhaldez the other man with a jerk the shiny withholds 'The other man withdrew the blade with a sudden jerk' h. wat, is þis Arþureȝ hous, quoþ þe haþel þenne what is this Arthur's house said the man then 'Is this King Arthur's house, said the man?' The linguistic variable In traditional variationist work, the linguistic variable is defined as "alternate ways of saying 'the same' thing" (Labov, 1972:188). Although this definition was originally applied to phonological variation, over time this concept was extended to the study of grammatical and lexical variation (Terkourafi, 2011). In early work on distributional semantics, Firth (1957:11) highlighted the importance of context when determining referential meaning, and in modern studies of variation, forms do not need to have the exact same denotation, but overlapping uses or a shared history is often a prerequisite to be treated as variants of the same thing. So long as the analyst identifies and includes the contexts in which the referential meaning is equivalent, and removes instances in which they are not, a two-step process known as the Principle of Accountability (Labov, 1969:737-738), and circumscription of the variable context (Labov, 1969:729), the linguistic variable can be used to study variation outside of phonology.
Although there are semantic nuances between Middle English nouns such as segge, hathel, freke, wyȝe, and mon, at the discourse level, they can have the same referential meaning, that is, in some contexts they can be used to refer to the same male adult. The examples in (4) from Sir Gawain and the Green Knight illustrate that these forms could be used interchangeably, as they appear in the same contexts to refer to the same green knight (i.e., quoþ the … 'said the…'). Since these nouns occurred in the same contexts to refer to the same referent, these variants were, at least at the level of discourse and in specific contexts, referentially equivalent.
(4) a. hit is sothe, quoþ the segge it is sooth said the warrior 'it is true, said the man' b. is þis Arþureȝ hous, quoþ the hathel is this Arthur's house said the nobleman 'is this Arthur's house, said the man' c. now iwyss, quoþ the wyȝe now I know said the wise one 'now on my word, said the man' d. yet firre, quoþ the freke yet further said the warrior 'yet further, said the man' e. madame, quoþ the myry mon madam said the merry man 'my lady, said the merry man' As Tagliamonte and Brooke (2014:11-12) point out, using semantic fields as a foundation for circumscribing the variable context is not new. For instance, Sankoff, Thibault, and Bérubé (1978) analyzed the semantic field of verbs which mean "to dwell," which led to the notion of "weak complementarity": the idea that linguistic variables can be identified through distributional properties and distribution across a speech community (Sankoff & Thibault, 1981:207). Recent studies have followed in this tradition when analyzing lexical variation (e.g., Stratton, 2022b;Tagliamonte & Brooke, 2014;Tagliamonte & Pabst, 2020). The concept of a semantic field, however, predates the variationist tradition and has its roots in structuralist semantics (Trier, 1931), with the important distinction between semasiology and onomasiology (Geeraerts, 2010). Semasiology, a concept that emerged in prestructuralist work, "considers the isolated word and the way its meanings are manifest" (Baldinger, 1980:278) and is therefore concerned with polysemy, that is, the various meanings a given word form can have (Geeraerts, 2010:84). In contrast, onomasiology "looks at the designations of a particular concept" (Baldinger, 1980:278), which can be conceived as studying varying levels of synonymy (Geeraerts, 2010:84). Studying the variants used to denote third-person male adult noun referents can therefore be viewed as onomasiology, which, unlike semasiology, starts with the concept and examines how it can be expressed. In a variationist framework, variants within a semantic field, lexical field, or onomasiological set can be studied as a linguistic variable. Therefore, the present study uses the notion of a semantic field, following "weak complementarity," to study the system of third-person male adult noun referents in Old and Middle English.

Data
To examine the semantic field of third-person male adult noun referents in Old English, the Helsinki Corpus of English Texts (Kytö, 1996;Rissanen et al., 1991) was used, which contains 413,250 words, divided into four subperiods: O1 (2,190 words), O2 (92,050 words), O3 (251,630 words), and O4 (67,380 words). While many texts from the same corpus were also included for the analysis of variants in Middle English, because the corpus of Middle English is substantially larger than the Old English counterpart, to ensure confidence in the circumscription of the variable context only a sample of the texts from the Middle English part of the corpus was used. However, because there are fewer verse texts in the Middle English corpus, to ensure that TEXT TYPE could be included as a factor in the analysis, additional verse texts were added from Sisam (1928 Circumscribing the variable context A list of third-person male adult noun referents for Old and Middle English was compiled through previous literature (Elsweiler, 2011;Grygiel, 2006;Kleparski, 2003Kleparski, , 2005Stenroos, 2002) ). Then, search queries were run to find these variants in the data. Since word forms are not lemmatized in the Helsinki corpus, a list of spelling variants and inflectional forms was compiled with the aid of the Dictionary of Old English (Cameron, Amos, & Healey, 2018) and was subsequently searched for in the corpus data.
Tokens were then downloaded and manually inspected for the removal of any nonequivalent instances.
Since the variable context was circumscribed to third-person male adult noun referents, variants such as OE man 'man' were only included in the analysis when they unambiguously referred to a male adult, as in (5). The presence of names, as in (5a), as well as the sociohistorical context, helped determine the gender of the referent. For instance, in (5b) it is evident that the referent Priam is male because he is a preost 'priest,' a role traditionally confined to men. Similarly, in (5c) the referent is King Arthur, referred to as a god mon 'good man,' who is presumably male. In (5d), the six men are biologically male as we are told they were castrated (their stanes 'testicles' were removed). 7 In contrast, examples in (6) were not included in the envelope of variation. While differentiating the gender-specific, gender-inclusive, and indefinite use of man is no simple task (Curzan, 2003:135;Rauer, 2017), indicators such as a preceding negative particle (e.g., no mon here vnmanerly þe mysboden habbez 'no one here has treated you in an unmannerly fashion') or indefinite adjectives (e.g., forþam nat naenig man 'therefore, nobody knows') helped disambiguate possible readings. Special attention was taken to ensure that the anachronistically homophonous and semantically nonequivalent form mān 'crime,' identified by a macron in editorial textual editions, was excluded. A number of functionally nonequivalent uses, such as the indefinite use of man, as in (6a), vocatives of address, as in (6b), as well as instances where the referential meaning was different, such as 'husband/boyfriend' in (6c), were excluded. 8 Instances in which wer meant 'wergild,' that is, a compensation tariff, were also excluded on grounds of being semantically nonequivalent. 9 Eorl 'earl' was not included when used as a term of address or rank (e.g., Harold eorl 'Earl Harold,' Godwine eorl 'Earl Godwin').
(5) a. ða waes Apollonius gehaten sum iung man se waes swiðe welig then was Apollonius called some young man who was very wealthy and snotor and wise 'there was a young man called Apollonius, who was very wealthy and wise' [Apollonius of Tyre, 950-1050] b. þa cleopede þe king Piram, aenne preost maeren then called the King Priam, a priest famous he wes swiðe wis mon and witful on bokken he was very wise man and witful in books 'Then the king summoned Priam, a famous priest, he was a very wise man and well read' [Layamon, 1150[Layamon, -1250 c. Arður wes an weorlde wis king and riche god mon and griðful Arthur was one world wise King and rich good man and peaceable 'In this world, Arthur was a wise and powerful King, a good man and amicable' [Layamon, 1150[Layamon, -1250 d. six men spilde here aegon 7 of here stanes six men deprived their eyes and of their stones 'six men had their testicles and their eyes removed' [Peterborough Chronicle, 1150-1250  For texts translated from Latin, comparisons between the Old English variants and the Latin counterparts aided in identifying the gender of the referents. For instance, in (7) it is evident that the people the idesa 'women' have not slept with are male, not only from context, but also because vir 'man' was found in the Latin text (habeo duas filias, quae necdum cognoverunt virum 'I have two daughters who are yet to have known/slept with men'). 10 Old English wer is cognate with Latin vir, but beorn was likely used because it alliterates with gebedscipe (cf., ge-prefixes are unstressed in Old English verse).
(7) ne can þara idesa owðer gieta not can the-GEN.PL woman-GEN.PL either yet þurh gebedscipe beorna neawest through intercourse men-GEN.PL proximity 'Neither of these women have slept with a man before' [Genesis, 950-1050] After circumscribing the variable context, each token was coded according to the available metadata: TEXT TYPE, TEXT ORIGIN, and TIME. The factor TEXT TYPE had two levels (prose, verse), TEXT ORIGIN had two levels (translated, not translated), and TIME had four levels for Old English (O1, O2, O3, O4) and three levels for Middle English (M1, M2, M3). While some metadata for DIALECT was available, DIALECT was not included as a factor for three reasons. First, not all texts contained such metadata. Second, there is some disagreement regarding the dialect in which specific manuscripts are written. Third, for Old English, there is a bias toward West Saxon texts, which substantially outweigh Northumbrian, Kentish, and Mercian texts. To test whether alliteration had a significant effect on lexical choices in verse, each variant in verse texts was coded binomially for the presence or absence of neighboring words with which the variant could alliterate. For the multivariate analyses, binary mixed effects logistic regressions were developed in Rbrul (Johnson, 2009), with TEXT ID run as a random intercept. In all models, the most frequent variant of the period (i.e., wer in Old English, and man in Middle English) was run against all other variants within that period, coded binomially.

Old English distributional analysis
For Old English, a total of 631 tokens were included in the envelope of variation. Of the twenty-five attested variants (beorn, carlman, cempa, ceorl, cniht, duguð, eorl, freca, guma, haele[þ], hildedeor, hyse, leod, magu, man, rinc, scealc, secg, sundbuend, þegn, waepman, waepned, wer, wiga, wigmen), wer was most frequent, which made up 42.2% of the semantic field. The variants man and guma competed for second place, each occupying approximately 13.5% of the system. The overall distribution of variants is reported in Table 1, with some examples of use in (8).  c. Hie þa aet burhgeate beorn gemitton sylfne sittan sunu they then at city-gate-DAT.SG man met.PL self sit son 'They then met the man, the son of Haran himself, sitting at the city-gates' [Genesis, 950-1050] To explore differences in use across prose and verse, the variants were cross-tabulated by TEXT TYPE (see Table 2). Different variants were favored by different text types. In prose texts, wer was the number one variant, at 63.6%, compared with 19.7% in verse texts. In contrast, guma was the number one variant in verse texts, at 24.8%, compared with 2.2% in prose. The type-token ratio (TTR), a common measure of lexical density, indicates that a wider range of variants was found in verse (n = 23 types, 314 tokens → TTR = .073) than in prose (n = 12 types, 317 tokens → TTR = .037). Of the 314 tokens found in verse, 76% (n = 239) alliterated with words in proximity. A chi-square test found that alliteration had a significant effect ( p < .001) on the lexical choices within the semantic domain of variants found in Old English verse. The need for alliteration may explain the wider range of variants found in verse compared to prose. As for the effect of provenance (Table 3), nontranslated texts contained a wider a range of variants (types = 25, tokens = 409 → TTR = .06) than translated texts (types = 13, tokens = 222 → TTR = .058), but the type-token ratio was almost identical. In translated texts, wer made up 68.3% of the semantic field. In contrast, although wer was also the most widely used variant in nontranslated texts, it made up a smaller share of the system (28.1%).

Old English multivariate analysis
To examine the statistical significance and relative strength of the factors operating on this semantic field, a binary mixed effects logistic regression was computed in Rbrul (Johnson, 2009). TIME, TEXT ORIGIN, and TEXT TYPE were run as fixed effects, with all possible interactions. The output is summarized in Table 4. Factor weights (FW) indicate the probability of the application value (i.e., wer) to occur in the listed context. Factor weights closer to 1 indicate a favoring effect whereas factor weights closer to zero indicate a disfavoring effect. Although TEXT was originally coded with four levels (O1, O2, O3, O4), due to the limited data available for O1 (2,190 words), for which only two tokens were included, O1 and O2 were collapsed into one level.
All three factors were found to significantly affect the probability of wer to occur in the Old English texts, with a significant interaction between TIME and TEXT ORIGIN. Wer occurred more frequently at the beginning of the Old English period (O2) than at the end (O4), in prose than verse, and in translated texts than non-translated texts. The range for the factor groups, calculated by subtracting the lowest factor weight from the highest, indicates that, of the three factors, TEXT TYPE had the strongest effect on the absence or occurrence of wer in Old English. A random forest (Hothorn, Hornik, Strobl, & Zeileis, 2015) was run to confirm the hierarchical ordering of the constraints: TEXT TYPE ranked first, followed by TEXT ORIGIN, and then TIME.
Although the higher frequency of wer in O4 (46.6%) than in O3 (32.3%) suggests the decrease in wer was not linear, when only prose texts are considered it is evident that wer continued to decrease throughout Old English. To confirm this, two follow-up models were run using data from only prose texts: wer occurred at a significantly greater frequency in O2 than O3 and significantly more frequently in O3 than O4, illustrating a significant downward trajectory over time.

Middle English distributional analysis
For Middle English, 246 tokens were included in the envelope of variation, with twenty-seven attested variants (bachelor, baroun, beorn, burne, carlman, cherl, duȝeðe, erl, freke, gome, hathel, kempe, knape, kniȝt, ladde, lede, man, rahȝe, renk, schalk, segge, swein, þein, tulk, wepmann, wer, wyȝe). The number one variant was man, with 57.3% (see Table 5). Examples of use are provided in (9). In contrast with Old English, the only attestation of wer referring to a male individual came Total N = 631, Input = .453, Texts = 72, *p < .05, **p < .01, ***p < .001 from the Early Middle English text Ormulum (9b). The low frequency of wer is consistent with evidence from the OED of its reported demise by Late Middle English (OED, were, n.1). 13 With the exception of (9b), when wer occurred in the Middle English data, it referred to a specific type of man, namely a married man, a use which was later usurped by the lexical item 'husband.' 14 (9) a. Þiss gode mann þiss gode prest þatt we nu maelenn offe this god man this good priest that we now speak of wass alls I seȝȝde nu littlaer Ȝehatenn Zacaryas was as I said now little-early called Zacharias 'This good man, this good priest, that we now talk of, was, as I said earlier, called Zacharias' [Ormulum, 1150[Ormulum, -1250 b. Uss birrþ heroffe witenn wel 7 seon 7 unnderrstanndenn, þatt David us behooves thereof know well and see and understand that David kingess kinness men, off weress oþþr off wifess wiþþ Aaroness kinness men king-GEN kins men from men or from women, with Aaron-GEN kins men Off siþre waerenn sammnedd to streonenn streon to wurrþenn sibb, from lately were gathered to acquire offspring to become relation wiþþ kingess 7 wiþþ preostess with kings and with priests 'It is necessary for us to know, see and understand, that the lineage of King David's kin, from men or from women, were gathered lately to have offspring in order to be related to Kings andto priests' [Ormulum, 1150-1250] c. Þa namen hi þa men þe hi wenden ðat ani god hefden, bathe then took they the men who they turned that any goods had both be nihtes 7 be daeies, carlmen 7 wimen, 7 diden heom in prison by night and by day men and women and did them in prison 'Then they seized those people who had any goods, both during the night and during the day, both men and women, and threw them in prison' [Peterborough Chronicle, 1150-1250 As for the distribution by TEXT TYPE (see Table 6), man was the overwhelming choice in Middle English prose (92%) but occupied 41.1% of the field in verse. Of the 168 tokens of third-person male adult noun referents in Middle English verse, forty-eight (18%) alliterated, suggesting that while alliteration still influenced lexical choices in Middle English verse, it played less of a role than in Old English verse. The diminished role of alliteration in Middle English, however, is indicative of a larger change in verse style, as 34% of the Middle English variants that did not alliterate in verse, instead rhymed (e.g., kniȝtriȝt, manþan). In texts such as Sir Orfeo and the Dancers of Colbek, rhyme is the emphasis, not alliteration.

Middle English multivariate analysis
A binary mixed effects logistic regression was run on the Middle English data, using man (the most frequent variant) as the application value. The output is reported in Table 7. The model found TEXT TYPE to significantly influence the probability of man to occur over any other variant, appearing more frequently in prose than in verse. TEXT ORIGIN and TIME also significantly affected the occurrence of man. While, like in Old English, the model suggests that the increase in frequency of man was not consistent across time, when only prose texts are considered it is evident that man continued to increase in frequency throughout Middle English. TEXT TYPE significantly interacted with TIME due to high frequency of genre-specific variants in verse (e.g., knight). These variants occur frequently, not necessarily because they were used frequently in everyday discourse, but due to the nature of verse content, wherein references to knights and chivalry are common. The range for the factor weights, along with a random forest indicates that TEXT TYPE had the strongest effect on the variation, followed by TIME and TEXT ORIGIN.

Changes from Old English to Middle English
To examine changes from Old English to Middle English, the frequency of wer, guma, and man was plotted across time (Figures 1 and 2). Frequency was measured by comparing the number of times a variant occurred versus the total number of referentially equivalent tokens by subperiod. Figure 1 shows the demise of wer from Old English to Middle English and its gradual replacement by man. The spike in frequency of wer in O4 (Old English: 1050-1150 CE) and the dip in frequency of man in M3 (Middle English: 1350-1420 CE) is a function of TEXT TYPE interference. When only prose texts are considered, the trend is clearer (see Figure 2). The change from wer to man follows a clearly identifiable s-curve pattern.

Discussion
To examine changes within the set of third-person male adult noun referents in the early history of the English language, the present study examined the frequency of variants pertaining to this semantic field from Old English to Middle English. Distributional and multivariate analyses show a clear shift from wer in Old English to man in Middle English. Although the gender-inclusive use of man continued into Middle English, the number of instances in which man was used with reference to male-only individuals increased. The shift from wer to man follows an s-shaped  distribution, a pattern typically observed in other areas of linguistic change (Nevalainen, 2015;Tagliamonte & Smith, 2021). While language change does not have to follow an s-shaped pattern (Kauhanen, 2017;Newberry, Ahern, Clark, & Plotkin, 2017), an s-curve temporal trajectory is often viewed as a clear indication of lexical replacement (Blythe & Croft, 2012:278-279;Chambers, 2002:361).
Assuming the data in the present study are representative, the low frequency of wer and its eventual demise in Middle English illustrates lexical change within this onomasiological set. While the actuation problem occludes the causation of this change (Weinreich et al., 1968:102), one might speculate that the Anglo-Norman loanword werre 'war' (MED, werre, n.), which shows up in twelfth century texts, had an influence on the demise of wer, as wer and werre could have been homophonous. 15 Although homophony of forms may be too simplistic of an explanation to account for the loss of wer "man," an explanation of this kind would be in line with the notion of a "homonymic clash" (Samuels, 1972:67-75), which has been proposed as a mechanism of change for several lexical items in Middle English. Since data show that wer was already decreasing throughout Old English, a homonymic clash could not have been the sole cause of this change, but this clash may have accelerated a change that was already underway.
Wer temporarily retreats to use as part of a related, but different, semantic field, namely "husband," which later too was replaced by the competing lexical item husband. 16 To the question Where did wer go?, the present study shows that wer was gradually replaced by the competing variant man from Old English to Middle English, with a clear relationship between the increase in frequency of man and the decrease in frequency of wer. The present-day compound werewolf, literally 'man-wolf,' is one of the few remaining breadcrumbs of this once frequently used noun and remains in the language today only as a vestige. 17 One implication of this study is that lexical replacement is gradual, as the shift from wer to man appears to have taken place over approximately 400-500 years. However, there are two alternative explanations for the tardiness of this change. First, because frequently used lexical items are typically replaced less frequently (Pagel, Atkinson, & Meade, 2007), the high frequency of wer in Old English may account for why this replacement took centuries to be complete. The second factor to consider is the written transmission. Given that lexical choices are known to shift from generation to generation (Tagliamonte & Brooke, 2014;Tagliamonte & Pabst, 2020), and the locus of linguistic change is generally acknowledged to be in spoken as opposed to written language (Milroy, 1992:32), the Old and Middle English extant manuscripts may leave the impression that this replacement was gradual even though the change may have been accelerated in spoken language but remained in the language in formal written contexts, as is attested in the extant manuscripts. 18 After all, there are well documented register effects that condition and constrain language variation and change (Biber, 2012), which may have contributed to the longevity of this lexical replacement. The seemingly gradual nature of this change may therefore be a byproduct of the limited data that remain, that is, the notorious "bad data problem" (Labov, 1994:11), as it is inevitable that an analysis of this kind would be construed through a written lens.
As for the factors contributing to variation, distributional and multivariate analyses indicate that TEXT TYPE and TEXT ORIGIN significantly affected lexis. Variants such as wer and man were more frequent in prose than in verse, with Old English verse texts making use of a wider range of variants, likely due to alliterative requirements. Variants such as shalk (OE scealc) and renk (OE rinc) rarely occurred in Old English prose, suggesting that these variants were bound by stylistic tradition. Whether a text was translated from a Latin source also significantly influenced the lexical decisions in Old English, with wer occurring more frequently in translated texts. In contrast, in nontranslated texts a wider range of variants was employed (e.g., freca, rink), but this effect may be due to the skewed proportion of verse in nontranslated texts compared to translated ones. The higher frequency of wer in translated texts may also be attributed to the fact that Old English wer and Latin vir 'man' are cognates (Proto Germanic *uiraz/ uiraR), although counterexamples in translation choice were found, as in (7) above.

Conclusion
The semantic field of third-person male adult noun referents is a dynamic and heterogeneous one, with analyses of present-day varieties of English pointing to recent changes within this domain. The present study showed that variation within this onomasiological set is not new and has existed since the beginning of the history of the English language. Wer was once the most frequently used variant to refer to a male adult, but it was gradually replaced over time by man. This diachronic shift in lexical preference followed a prototypical s-shaped distribution, suggesting that, like other areas of linguistic change, lexis may follow similar patterns of change. While research using apparent time data (Chambers, 1995) or short periods of time (Grieve et al., 2017) point to the applicability of s-shaped trajectories for lexical change, the analysis of change within the semantic field of third-person male adult noun referents over approximately six hundred years adds an important diachronic dimension to this discussion.