From deceit to pain: Late Latin dolus and the interplay between semantics and analogy

1 Introduction

The Late Latin semantic change where dolus, in Classical Latin ‘deceit’, took on the meanings of dolor ‘pain, suffering, resentment’, is often mentioned, but seldom explored. It is common to see it presented as an example of poor Latin. The usage is called “illiterate”, and the change that produced it is referred to as “lexical contamination”, “malapropism” and “homonymic confusion”.^[1] Over the years, the function by which dolus took on the meanings of dolor has only been discussed in the most general terms. This article will lay out the evidence of dolus taking on the semantics of dolor, discuss the routes by which the change may have happened and place the change in a broader methodological context.

In order to discuss this topic, we must first look at the semantics of both dolor and dolus in Classical Latin.

Dolor was used in reference to a range of negative feelings and sensations.^[2] It could denote physical pain, both localised (e.g. Cic. Brut. 34.130, Plaut. Cist. 1.2.22) and generalised (e.g. Gell. 10.3.17), and mental anguish. The most common such meaning was grief or sorrow (e.g. Cic. fam. 4.6.2), but other negative emotions, such as anger or resentment (e.g. Ov. met. 3.395), were also encompassed by it. Dolor could also refer to the object or cause of grief (e.g. Prop. 1.14.18).

Dolus was most commonly used in legal contexts as part of the standing expression dolus malus ‘ill intent’.^[3] On its own, dolus meant ‘deceit’ or ‘deception’. It was often used in phrases such as non dicam dolo ‘I say without deceit’, i.e. truthfully (Plautus Men. 228; Plaut. Trin. 91), and together with words such as fraus ‘fraud’ (Liv. 1.53), fallacia ‘trick’ (Sall. Cat. 11.2) and insidiae ‘ambush, trap’ (Caes. Gall. 4.13.1). It could also be used of the means by which the deceit is perpetrated (e.g. Ov. hal. 25).

Throughout this article, I will make a distinction between “dolus”, “dolus ‘deceit’” and “dolus ‘suffering’”. When using the word dolus without translation, the formal word is under discussion. Dolus with its classical meaning, as outlined above, will be indicated by “dolus ‘deceit’”. Dolus with any of the meanings of Classical Latin dolor, whether ‘pain’, ‘suffering’, ‘grief’, ‘anger’ or ‘resentment’, will be indicated by “dolus ‘suffering’”.

The attestations of dolus ‘suffering’ will be given with full references and context in Section 2.^[4] All example numbers and references, as well as the attested case-form(s) and any words in agreement with dolus, can also be found in Table 1. Section 3 is dedicated to the issues surrounding the concepts of homonymy, polysemy and semantic change. In Section 4, theories on the process by which dolus came to mean ‘suffering’ will be discussed.

2 Literary and epigraphic attestations of dolus ‘suffering’

Determining whether a semantic change has taken place can be difficult, as the formal word remains unchanged and (depending on the semantics involved) the context may not give us any obvious clues. In the case of dolus, we can be certain that a change did occur. Romance words for ‘grief’ such as Spanish duelo, French deuil and Italian duolo cannot be derived from dolor, but must be from dolus (see Väänänen 2006 [1981]: 87; Löfstedt 1959: 161; TILF Online s.v. deuil). Furthermore, ancient writers commented on the change.

In this section, I will first discuss attestations where dolus ‘suffering’ is discussed, followed by other literary attestations (some of which are more ambiguous) and finally evidence from the epigraphic record.

The longest and most famous discussion of dolus ‘suffering’ appears in Augustine’s tract on the gospel of John, dated between 406 and 420 (Tornau 2020). This passage is in reference to John 1.48, rendered in the Vulgate as Ecce uere Israelita, in quo dolus non est ‘behold truly the Israelite, in whom there is no deceit’.

This passage gives us a terminus ante quem for dolus ‘suffering’. It may be tempting to date the inception of the change to Augustine’s lifetime. This would be ill-advised, as there is a tendency (both then and now) to think of ‘incorrect’ usage as something new-fangled. It is for instance common to hear people bemoan the use of English literally to mark a hyperbolic expression as something that has entered the language recently, but the first attestation in OED of this usage is from 1769 (OED Online s.v. literally, I. 1. c). Therefore, this passage of Augustine only tells us that the change in semantics occurred at some point before its composition.

Augustine does not mince his words when it comes to dolus ‘suffering’, making it clear that he thinks this is bad Latin. However, this is not to say that the fratres imperitiores latinitatis ‘brothers who are less skilled in proper Latin’ are not native speakers of Latin. Latinitas is not Latin as a language, but specifically the correct form of Latin as taught in school.^[6] The implication is rather that they are uneducated or, worse, sloppy with their speech.

Often when dolus ‘suffering’ is discussed, this passage from Augustine is the only evidence referenced.^[7] This centres Augustine’s prescriptivism, and, by extension, plays into the idea of Late Latin as a language on the decline, brought to ruin by careless or incompetent speakers. It also makes it easy to assume that this is the only evidence of this semantic change, which is not the case.

Dolus is ascribed both the meanings of ‘deceit’ and ‘suffering’ in a Latin-Greek glossary found in a ninth-century manuscript (Loewe and Goetz 1888: vii):

As literary works have come down to us through a manuscript tradition, we cannot be completely sure of the word-choice of the author when not as explicit as in (1) and (2). A scribe whose vocabulary includes dolus ‘suffering’ may easily slip up and write dolus instead of dolor, and a scribe who believes dolus should not be used to mean ‘suffering’ may change dolus to dolor. To account for this, I will include information on the oldest manuscript including dolus ‘suffering’.

As dolus is often seen as ‘poor’ usage, editors will often change it to dolor. This is the case in all editions of the fourth-century veterinarian Pelagonius’ work on hippology, the oldest manuscript of which is from 1485.^[8] Both Ihm (1892: 78) and Fischer (1980: 39) repeat the 1826 correction made by Sarchanius of dolori instead of dolo. Adams (1995) supports this correction, as “it is hard to believe that Pelagonius would have written dolus = dolor just once” (Adams 1995: 634). For the purposes of this article, it is not necessary to establish whether Pelagonius himself used dolus in this way, as the manuscript nonetheless shows dolus ‘suffering’ in use. With the form dolo intact, the passage, which describes how to make a poultice, reads:

The History of the Franks by Gregory of Tours includes dolus ‘suffering’ once. The work itself is dated to the second half of the sixth century (Simonetti 2014: 186–187), and the earliest manuscripts of the relevant part is from the seventh century (Arndt and Krusch 1951: xxiii–xxviii). The word appears in a passage about Austrechild’s dying wish of revenge on the physicians she thinks have poisoned her:

Two alternative readings exist. Manuscript B5, dated to around the year 700, instead writes dolor. Manuscript C1, from the ninth century, gives it as dolum instead, but appears to still give unus in the masculine. Notably, the dolor reading is outnumbered, as dolus appears in four manuscripts, two of which predate B5.

The Anonymi Valesiani consist of two separate texts that were contained in the same ninth-century manuscript. They got their name from being published by Henri Valois in 1636. The second part has been dated to the early sixth century, and is thought to have been written in Ravenna (Adams 1976: 1–8). Both extant manuscripts contain one instance of dolus ‘suffering’, though this has been ‘corrected’ in the majority of editions (Mommsen 1892: 328; Cessi 1913: 20; Moreau 1961: 26) to a form of dolor. Without this correction, the sentence reads:

Dolus occurs twice in the Continuations to the Chronicle of Fredegar, which were likely produced by three different writers. (6a) occurs in a section likely written as a continuation of Liber Historiae Francorum, rather than the Chronicle of Fredegar. It would have been finished at some point after 736. (6b) occurs in the third continuation, which was written at the behest of Count Nibelung and was completed at some point after the death of Pippin III in 768.^[9] Both examples can thus be dated to the eighth century. The Continuations are found in manuscripts dating to the ninth or tenth century (Wallace-Hadrill 1960: li–lvi). Both attestations are part of variants of the same set phrase, used as hedging when describing untimely deaths:

Literally, this means ‘which is anguish to say’, but a more idiomatic translation would be ‘and I’m sorry to say’ (cf. Wallace-Hadrill 1960: 82). Although the meaning is different, it is of note that this phrase is similar to non dicam dolo ‘I say without deceit’ discussed above.

One of the Latin translations of Aratus’ Greek astronomical work Phaenomena includes dolus ‘suffering’ as part of a discussion of the mysterious constellation Ingeniculus, On His Knees. The Greek is:

One Latin translation, made by an unknown person, designated Anonymus II, renders the second line as follows:^[11]

The line in question does not survive in Cicero’s translation, but in Germanicus’ version, the corresponding word is labor (German. Arat. 65). Maass (1898: xxi) dates the Anonymous II translation to the sixth century at the latest. The earliest manuscript where it survives is from the ninth century (Maass 1898: 101).

Dolus with the meanings of dolor also appears in a poem recounting stories from the Alexander Romance, found in a manuscript from the ninth century (Strecker 1964: 451). The line in question, which introduces the story where Alexander the Great flies in a basket carried by griffons^[12] to see the whole world, is unfortunately corrupted:

(Rhythmi Aevi Merovingici et Carolini 63.5 = Strecker 1964: 601)

†Fere morte dolus magnus luctursque miserabilis

Zacher (in Strecker 1964: 601, app. crit.) has proposed reading this as:

The epitaph over Grimoald III of Benevento, dead in 806 (see Anderson 2017: 59–60), is quoted in the tenth-century Chronicon Salernitanum, preserved in a manuscript from 1300 (Westerbergh 1956: xix, xxiv). It includes the lines:

The above-quoted examples of dolus ‘suffering’ are fairly unambiguous, but there are also some examples where the meaning intended is not so clear-cut. One such example is from the Christian poet Commodian. The dating of his work is contested, ranging from the third to the fifth century (Baldwin 1989: 337–342). His work survives in a manuscript from the ninth century (Salvatore 1965: 12). Both TLL (5,1: 1837) and Lindner (1992: 113) quote this as a clear example of dolus ‘suffering’:

The poem in question does describe suffering, both physical and spiritual, but the larger theme is the differences between the life on Earth and the existence in the next life. In early Christianity, the description of the next life concentrates on the clarity of the senses rather than the absence of pain (cf. 1 Cor. 13.12.) Considering this, this line may instead mean ‘you can hope to live without deceit’.

Similarly, there is a passage from one of St Ambrose’s letters that is given as an example of dolus ‘suffering’ by TLL (5,1: 1837). Ambrose lived during the latter half of the fourth century, and the earliest manuscript of the passage in question is from the ninth century (Faller 1968: 1). In this letter, Ambrose is using Deut. 21.12, which describes shaving the head of a captive bride when first bringing her into one’s home, as a metaphor:

Does dolus in this context have the meaning ‘suffering’ or ‘deceit’? Reading dolus as ‘deceit’ makes little sense. We can exclude the possibility that the intended meaning is that a razor that is too sharp might slip and ‘betray’ the user, as it would be highly unidiomatic. This would seem to indicate that ‘pain’ is the more likely meaning. However, it is not clear whether a razor that is too sharp would cause pain. Modern straight razors, if sharpened too much, would not, but fourth-century razors were considerably different, being made from bronze, iron or a steel-iron combination (Boon 1991: 27–28). Things become even murkier when we consider Ps. 51.4, which is rendered in the Vulgate as:

This very translation occurs in Ambr. Off. 3.10.75, so Ambrose was aware of this phrasing. The occurrence of the words noucacula, acuta and facio raises the question whether Ambrose is alluding to the passage in Psalms, where it means ‘deceit’. It seems more likely that dolus here means ‘pain’, but this possible allusion makes the issue far from straightforward.

The epigraphic evidence of dolus ‘suffering’ consists of eight funerary inscriptions. As this material is seldom discussed, the inscriptions will be quoted in full.^[13]

An inscription from Aquileia uses the form dolium in a similar way. Due to the fragmentary nature of this inscription, only the relevant lines are quoted below.

Many of these inscriptions do not consistently use Classical epigraphic formulas, and a number of them appear to be Christian. CIL V.1638 ([14b]) and XI.3054 ([14e]) are adorned with christograms, and CIL X.4510 ([14d]) is adorned with crosses. Phrases containing pax ([14d], [14e], [14g]) also indicate a Christian belief-system.

A number of the inscriptions contain dating formulas. The consulate of Eusebius and Hypatius ([14e]) was in 359 CE. The sole consul Basilius, mentioned in (14d), (14g), may be one of three sole consuls of that name, in 480, 527 or 541 CE. This means that (14d) is from 492, 539 or 553 CE,^[14] and (14g) is from 513, 560 or 574 CE. Although (14h) is included in CIL XIV, a volume dedicated to old inscriptions from Latium, the inscription must be far more recent. Not only is the inscription written on marble, which was uncommon until the early Roman Empire (see Bruun and Edmondson 2015: 16), but the consul mentioned appears to be Flavius Stilicho, Western consul during the year 400 CE. Although he had a co-consul, Aurelianus, this was not made public in the West, which would explain why he is listed as sole consul.^[15]

A final piece of evidence for dolus ‘suffering’ does not in fact feature dolus or meanings relating to pain or grief. It is found a Greek-Latin bilingual psalter kept in Verona, dated to the sixth century, rendering the following Septuagint passage:

This is translated into Latin as follows:

Blanchinus indicates in a footnote that dolore may be dolorem, implying that there may be a tilde or line above the <e>, indicating a nasal (see Bischoff 2017 [1990]: 157). However, he makes no mention of dolo or dolum, and the uncial script makes a mistake on Blanchinus’ part unlikely.^[16] This is the only time dolor is used to mean ‘deceit’, and is thus not evidence of dolor undergoing a change. Instead, it is a case of hypercorrection. The translator or scribe likely had dolus ‘suffering’ in their vocabulary, but was aware that this usage was not considered ‘proper’. As a result, they would correct themselves when saying dolus, in this case even when they meant dolus ‘deceit’.

The use of dolus ‘suffering’ is often called a medieval feature (e.g. Dinkova-Bruun 2011: 290–291; Harrington et al. 1997 [1925]: 7; McGillivray et al. 2022: 56), but it may be more correct to call it a feature of Late Latin, extending into early mediaeval Latin. Figure 1 shows the dating range of those attestations of dolus ‘suffering’ where dates can be ascertained (all but [14a]–[14c], [14f], [14g], [14i]). (15b) has also been included, as it is also evidence of dolus ‘suffering’. The century of composition and the century of the first surviving manuscript are marked in black, with any intervening centuries being marked in grey.^[17] In cases where dolus ‘suffering’ is expressly discussed ([1], [2]), where the date of composition is unknown ([2], [9], [15b]) or where the evidence is epigraphic ([14d], [14e], [14g], [14h]), only one century has been marked.

Figure 1:

The dating range of all datable examples of dolus ‘suffering’, as described in Section 2. In cases where the attestation has survived through a manuscript tradition, the first black square indicates the date of composition and the second black square indicates the first surviving manuscript. The intervening centuries are indicated in grey.

The largest concentration is found between the fourth and ninth centuries, with most compositions occurring between the fourth and sixth centuries. The datable epigraphic evidence also dates between the fourth and sixth centuries. Only four attestations ([3], [6a], [6b], [10]) extend beyond the ninth century. Pelagonius ([3]) has an unusually late first surviving manuscript from 1485. The epitaph Grimoald III of Benevento is from the early ninth century, recounted in a tenth-century chronicle preserved in a manuscript from around the year 1300. The manuscript of the Continuations ([6a], [6b]) may be from either the ninth or tenth century.

Establishing a geographical area is more difficult, as we cannot always be sure whether the original authors or later copyists were responsible for the occurrence of dolus ‘suffering’. However, some conclusions can be drawn. Augustine’s discussion of dolus ([1]) proves that it was in use in North Africa. Most of the epigraphic evidence ([14b]–[14e], [14h], [14i]) is from Italy. We may tentatively add the epitaph of Grimoald III ([10]), with the caveat that it has only survived as part of a manuscript tradition in Chronicon Salernitanum. There are also two inscriptions ([14f], [14g]) from what is now Austria, and one ([14a]) from Dalmatia.

3 Homonymy, polysemy and semantic change

Depending on one’s outlook, one could describe the process by which dolus ‘suffering’ was developed as one or several of the following:

1. The development of a homonym to dolus ‘deceit’

2. Increased polysemy of dolus

3. A semantic change relating to dolus

Most linguists, when met with one formal word-form having multiple meanings, will distinguish between polysemy, when one lexeme has multiple meanings, and homonymy, when two lexemes are formally the same but have different meanings. Take for example English seal:

Example (16a), referring to the pinniped, is clearly a homonym to examples (16b) and (16c). Examples (16b) and (16c), referring to the physical wax seal and the design it bears, are polysemous.

In this case, the distinction is obvious, but frequently, it is far less clear-cut. Many of the problems stem from the issue of how homonymy and polysemy are defined. Some linguists favour a diachronic definition, where words with the same etymology are polysemous while words with no shared history are homonymous.^[18] Others argue for a synchronic definition, where the impressions of living speakers define what is polysemy and what is homonymy. Neither is entirely reliable. The diachronic approach is difficult to put into practice, as we cannot always establish the etymology of words, and a historic connection does not necessarily mean the words are seen as polysemous. As Panman (1982: 118) points out, English port ‘harbour’ and port ‘fortified wine’ are most certainly homonyms, but both go back to Latin portus, though through different routes (cf. Saeed 2011: 65).

The synchronic approach assumes access to living speakers, something that makes historical discussions of polysemy and homonymy impossible. Although cognitive linguistic (see Short 2016: 3) and computational linguistic (McGillivray et al. 2019; McGillivray et al. 2022) approaches to polysemy have been applied to both Greek and Latin, there is no way to use eye-tracking technology or neurolinguistic approaches such as EEG or fMRI (Gries 2019: 34–37) on native speakers of these languages. Even when speakers are available to be interviewed and engaged in studies, consensus on specific words is not a given. Not only will different speakers make different judgements on whether something is polysemy or homonymy, the same speaker will make different judgements when presented with the same material at different times (Lehrer 1974: 10).

Not all scholars believe that the differentiation between homonymy and polysemy is useful. Traugott and Dasher (2004) argue that homonymy should only be invoked where there is “no clear semantic relation between the meanings of a phonological string” (Traugott and Dasher 2004: 14). Both cluster of meanings associated with dolus are negative concepts, often abstract and subjectively felt. As shown by the examples (11) and (12), it is not always clear which meaning of dolus is intended, which would speak for this being a case of polysemy rather than homonymy. Nonetheless, this is not clear-cut. One common taxonomy of polysemy identifies two types of polysemy – regular (where two polysemous senses have the same relationship to one another as another set of meanings, e.g. “liquid for portion of liquid”, coffee, beer) and irregular, which does not (Vicente and Falkum 2017). On the face of it, dolus would be a case of irregular polysemy. However, irregular polysemy is often associated with metaphor (Apresjan 1974). One may make an argument that pain or suffering is an example of our bodies or minds deceiving us, but this feels far-fetched. This raises the question whether ‘deceit’ and ‘suffering’ are in fact close enough to be even irregular polysemy. Perhaps instead we are dealing with two homonymous words, dolus ‘deceit’ and dolus ‘suffering’.

As with the distinction between polysemy and homonymy, there is debate over the definition of semantic change. There are those who treat semantic change as only occurring when the old meaning falls out of use in favour of the new meaning. This is a view likely influence by the concept of sound-change, which is defined as being regular and exceptionless according to the Neogrammarian regularity hypothesis, and is not infrequently discussed as if it were immediate. It has been shown through observations of ongoing sound changes, most famously the changes in the diphthongs /ai/ and /au/ in Martha’s Vineyard (see Labov 1963), that the changes spreads by diffusion through both speakers and their vocabularies. Similarly, semantic change is not immediate. Instead, when a new meaning is acquired or developed, it is added to the range of the word. As Traugott and Dasher (2004: 11) put it, a semantic change is not A > B, but rather A > A ∼ B, sometimes followed by > B. Dolus did not lose the meaning ‘deceit’ when taking on the meanings of dolor (e.g. Ambr. Off. 3.10.66). Instead, both meanings coexisted.

The traditional typology of semantic change, based on work done by Michel Bréal in the late nineteenth century and still frequently cited, define six mechanisms, usually presented in pairs: metaphor and metonymy, broadening and narrowing, amelioration and pejoration (Trask 1996: 42–44). In recent years, this typology has come under criticism (e.g. Fortson 2003: 651–652; Traugott and Dasher 2004: 4). These three pairs of mechanisms do not describe the process by which the change happens, but instead, focus on the relationship between the meanings (Fortson 2003: 652). In addition, they do not account for every type of change – the change in dolus does not fit comfortably into any of Bréal’s categories. The only option, metaphor, feels unlikely, as outlined above.

There appears to be some predictability in semantic change when it comes to verbs, adjectives and adverbs on a macro-level (Traugott and Dasher 2004), but there is little such predictability when it comes to nouns. Nouns, more than any other word-class, are the subject of extralinguistic pressures, ranging from social change to technological advancements. As the context of the semantics of nouns is cultural, the routes of change can seldom be predicted.

In order to explain dolus ‘suffering’, it may be necessary to look beyond semantics and explore other means of change, something which will be the focus of the next section.

4 Routes of change: analogy and lexical morphology

This section will include quantitative analyses drawing from two sources: the Latin linguistic corpus LatinISE (www.sketchengine.eu/latinise-corpus/) and the dictionary search tool of the Perseus Digital Library (www.perseus.tufts.edu/hopper/search), accessing Lewis and Short (1951 [1879]).^[19]

LatinISE includes annotated texts ranging from archaic Latin to neo-Latin, and totals around 13 million words (McGillivray and Kilgarriff 2013). It is hosted on Sketch Engine, which hosts a large number of corpora and dictionaries, and has robust search options. The material used in Table 3 has been collected using the ‘concordance’ search function, which allows the user to access all attestations of a certain word, regardless of case-form (see McGillivray and Kilgarriff 2013: 7).

Perseus Digital Library makes a huge number of Greek and Latin texts available, often with multiple English translations, as well as material outside the Classical world, both temporally and geographically.^[20] The “Dictionary Entry Lookup” function is linked to dictionaries in English, Greek, Latin, Arabic and Old Norse, and allows the user to search both a fronte and a tergo. In Tables 2 and 4, material has been collected using this search function using the a tergo option. When used for Latin, this search function leads to both Lewis and Short (1951 [1879]) and its less exhaustive sibling, Lewis (1891). Only lemmata from Lewis and Short (1951) have been included for consistency’s sake.

Having reviewed the evidence, considered terminology and discussed the methods used in this section, we can now look closer at some of the theories regarding the development of dolus ‘suffering’.

The most common explanation can be best represented by quoting Adams (1976): “in vulgar texts by a process of lexical contamination certain words frequently acquire the sense of other, unconnected, words of similar form. Rare words are particularly susceptible to semantic change of this type. […] So dolus comes to mean the same as dolor” (Adams 1976: 25).^[21] The claim about rare words appears to have originated with the German philologist Hermann Paul, who in 1886 suggested, with no examples cited, that uncommon words undergo semantic change more often than common ones, as speakers are less likely to have heard it used ‘correctly’ (Paul 1886: 75–77). To date, no evidence has been presented to support this. Both common and uncommon words undergo semantic change. If we by ‘common words’ mean core vocabulary, it would be true that uncommon words change meaning far more often, but that is not on account of them being more susceptible to change, but that there are simply a larger number of uncommon words than common core-vocabulary words (see Fortson 2003: 666 n.36).

Furthermore, outside of core vocabulary, it is difficult to ascertain what makes a word ‘common’ or ‘uncommon’. When there are too many attestations of a word for TLL to list them all, a representative sample is cited instead, and the entry is marked with an asterisk. This may lead us to think that a word with an asterisk is common, while one without is uncommon. The entry for mulctra and mulctrum ‘milk pail’ (which are listed together; TLL 8: 1565) is not marked with an asterisk – all attestations take up 36 lines. By contrast, only a representative sample of attestations is given for clipeus ‘shield’ (TLL 3: 1351–1355). Nonetheless, the entry is a total of 354 lines. For a soldier (or, indeed, an epic poet), the word clipeus would be more common than mulctra, but to a farmer, it would not. What is an uncommon word will differ from speaker to speaker, and is dependent on social standing, gender roles and a number of other factors. While the extant corpus of Latin is extensive, we do not have access to a perfect representation of the language. As most Latin texts of antiquity were written by men of the Roman elite, there is a bias in the material, in favour of the everyday details of wealthy free men’s lives and the topics that interested them. Thus ascertaining whether a word is common or not is difficult, if not impossible. Explaining the advent of dolus ‘suffering’ by pointing to incompetent speakers is lapsing into the same prescriptivism as we see in Augustine, and serves to evade a far more interesting discussion.

Dinkova-Bruun (2011: 291) touches upon dolus ‘suffering’ briefly, calling it “homonymic confusion”. Although an argument could be made that dolus ‘deceit’ and dolus ‘suffering’ are homonyms, such an argument cannot be made for dolus ‘deceit’ and dolor ‘suffering’. The expanded semantic scope of dolus was not down to dolus and dolor becoming phonetically indistinguishable. This would require a merger of short u and short o, which did not occur in any regional variant of Late Latin. The most common merger of vowels was instead between short u and long o (see Adams 2007: 262). Another approach is thus necessary.

Many scholars have sought to explain dolus ‘suffering’ as a separate postverbal from doleo (Baldonado and Allen 1981: 64; Baldonado 1981: 64; Dinkova-Bruun 2011: 291; Lindner 1992: 122; Stotz 2000: 59; Väänänen 2006: 91–92). However, there is a strong association between second declension statives and nouns in -or, e.g. frigeo ‘be cold’ and frigor ‘cold’ (see Weiss 2020: 429), making a derivation in -us less likely. Table 2 shows the conjugations of verbs which are the basis for deverbal statives in -or. Forty-seven, almost two-thirds of the total, are derived from second conjugation verbs. For another five (amaror ‘bitterness’, dulcor ‘sweetness’, rancor ‘stink’, tardor ‘slowness’ and uuor ‘moistness’), associated third conjugation verbs in -ēscere (dulcēsco ‘become sweet’, rancēsco ‘become rancid’, tardēsco ‘become slow’, uuēsco ‘become moist’) are attested. As can be seen by the long e, these are derived from second conjugation verbs (cf. frigeo ‘be cold’ and frigesco ‘become cold’), which in these cases are not attested.

When looking at the total number of second conjugation verbs, this connection becomes even stronger. Lewis and Short (1951) lists 151 second conjugation verbs with distinct roots.^[22] This means that almost a third of distinct second conjugation verbs have a derived stative in -or. Baldonado and Allen (1981: 65), who themselves subscribe to dolus being derived from doleo, point out that the creation of o-stems from verbs is commonly seen in first conjugation verbs, rather than second conjugation verbs such as doleo.

A more fruitful endeavour is Lindner’s suggestion of linked heteroclisis (Gelenkheteroklisie). He observes that both dolus and dolor share the genitive plural dolōrum (Lindner 1992: 121). As a result, it is possible to extract either word from this form. Considering that most meanings of dolor are abstract, it is worth asking to what extent it is in fact countable. Table 3 shows the distribution of singular and plural forms of dolor and, for comparison, five abstract nouns with meanings relating to sorrow or pain and five concrete nouns referring to wounds and injury in the Latin linguistic corpus LatinISE. As might be expected, concrete nouns occur more often in the plural than abstract nouns, 38 % as opposed to 19 %. Dolor is relatively close to the average percentage for abstract nouns, at 15 %. While plurals are not particularly common, they do occur, making the shared genitive plural dolorum therefore a possible reason for the semantic change of dolus.^[23]

The semantic change of dolus did not happen in isolation, and was not caused solely by its similarities to dolor. Instead, we must consider it in its wider context. The English example disinterested may serve as a parallel:

The original meaning of disinterested was ‘impartial’, as in (17a), but the word has also taken on the same meaning as uninterested ‘with no interest in a certain thing’ ([17b], [17c]). The fact that these words look similar is an important starting point, but the existence of similar words gave the change in semantics more fuel. There is a large number of words in English using the prefix dis- to form an antonym. Many of these words are verbs (allow – disallow; arm – disarm; empower – disempower; favour – disfavour) that have associated participles: disallowed, disarmed, disempowered, disfavoured. These verbs communicate not only the opposite, but an active opposition. Something that is disallowed is not simply not allowed, but is declared thus. Someone who is disempowered is not just someone who does not have power, but someone whose agency has been actively withheld. There are also adjectives such as dishonest, which strengthens the sense of dis- as a privative suffix.^[24] In addition, the word interest developed new meanings, adding the meaning of ‘emotional engagement’ to the existing meaning of ‘stake’ (cf. OED Online, s.v. interest I, 1 and I, 7). In light of these other words, the new meaning disinterested ‘not at all interested’ develops by analogy.

In a similar way, the broader derivational patterns associated with both dolus and dolor may shed light on the situation. Derivational grammar is not a dogmatic system, but an organic set of developments which have later been codified by linguists and philologists seeking to describe what can be observed. A more flexible approach is therefore often more fruitful. The following analysis will focus on adjectives derived from s-stem nouns using the derivational suffix -oso-.

Lewis and Short (1951) list 26 adjectives ending in -orosus.^[25] Two of these, summorosus and indecorosus, are derived from other adjectives, and will therefore be excluded. Of the remaining 24, three – elleborosus ‘in need of hellebore’ (from elleborus ‘hellebore’), mŏrosus ‘lingering’ (from mora; this adjective should not be confused with mōrosus ‘capricious’ from mos ‘habit’) and torosus ‘muscular’ (from torus, tori ‘muscle’) – are derived from o- or a-stems. This leaves 21 adjectives that are derived from s-stem nouns, one of which is dolorosus ‘painful, suffering’. The adjectives and the nouns from which they were derived, as well as variant forms of those nouns, are presented in Table 4.

Most often, s-stems ending in -or and -os are masculine, and s-stems ending in -us or -ur are neuter (cf. Weiss 2020: 327). The exceptions to this will be discussed below. In oblique case-forms, masculine and neuter s-stems can be told apart from the length of the penultimate vowel as well as accent. The adjectives retain the difference in vowel length, making it possible to tell which grammatical gender the noun from which the adjective is derived is, but have the same accent as one another. For comparison, the vowel length and accent of the genitive singulars of the masculine humor ‘moisture’ and the neuter pectus ‘chest’ and the adjectives derived from them compare as follows:

Here, the geographic distribution of the attestations of dolus ‘suffering’ is relevant. Augustine, based in Africa, is responsible for the earliest datable mention of dolus ‘suffering’. There are also multiple attestations from Italy, as well as epigraphic evidence from Dalmatia. Romance developments in Sardinia, the Balkans and some parts of Southern Italy show a merger between Latin long and short o. There is also evidence of a similar merger in African Latin, as when Augustine complains that the imperiti (the same adjective used of those who use dolus ‘suffering’) cannot tell apart ŏs ‘bone’ and ōs ‘mouth’ (Aug. doctr. christ. 4.10.24).^[26] Considering this overlap, we may assume that the environment under discussion here is one where the distinction between long and short o has been or is in the process of being lost. Old accent patterns would nonetheless persist. All adjectives formed from s-stems and -oso- still had the stress on the penultimate, while oblique case of masculine and neuter s-stems would still have different stress, péctoris as opposed to humóris.

In most cases, the s-stems associated with -oso- adjective occur in one form, e.g. nidor ‘vapour’, nemus ‘woods’. However, there are four cases where variant forms exist. Vapor ‘steam’, a masculine, had the alternative form uapos (a form also recognisable as a masculine) attested in Lucr. 6.952 and Non. 782 L. Marmor ‘marble’, a neuter, is the exception to the rule that s-stems in -or are masculine. The more recognisably neuter marmur is commented on by Quintilian (inst. 1.6.23), and is attested throughout in the fourth-century veterinary treatise Mulomedicina Chironis (110, 576 592, 605 = Oder 1901: 35, 185, 190, 194) in reference to hard tumours in horses’ joints.^[27]

Robur ‘oak’, a neuter, is recognisably neuter, but has an alternative form in -or (Lucr. 2.1131; Char. GL 1.30.5, 43.31, 86.5, 119.2, 457.16, 543.29, 548.16). Quintilian discusses this form at length:

Other writers saw robor and robur as belonging to different spheres, as shown in a mnemonic verse repeated by a number of grammarians:

Robur also has a third form, robus, which occurs in Cato (agr. 17.1) and Columella (2.6.1).

Finally, there is the case of decŏrōsus, and two nouns from the same root, decus and decor. Both nouns are s-stems, but differ in gender and, in the oblique case, both vowel length and accent, neuter decus, décŏris versus masculine decor, decó̄ris. Dictionaries list these two as separate words, with different, though quite similar, meanings. Decus is to be used to refer to honour, dignity and virtue. Decor is used of beauty, grace and elegance. However, both decus and decor can be used of ornaments. In TLL, the definitions of the first meaning in both entries are almost identical, id quod decet atque exornat ‘that which is comely and adorns’ for decor (TLL 5,1: 206) and id quod hominem uel rem decet atque exornat ‘that which is comely and adorns a person or thing’ for decus (TLL 5,1: 236). Numerous Roman grammarians lay out rules regarding which form should be used when. Some also repeat a mnemonic similar to (20):

In prose, the oblique forms of these two words would be indistinguishable, e.g. <DECORIS> (cf. TLL 5,1: 206). In speech, the oblique forms would be distinguishable by the stress, and in poetry, the vowel length would often be evident due to the metre, but the grammarians’ advice indicates that they think speakers are not telling these words apart properly (e.g. Diff. Suet. 331.9 = Roth 1858: 311; Pseudo-Palaemo GL 5.536.41–537.1). This indicates that to at least some Latin speakers, decus and decor were variant forms, much like robur and robor are presented in our dictionaries. It may also indicate that the stress and vowel length was seen as variable in this type of word.

Formally, the adjective decŏrōsus is derived from decus, décŏris, as indicated by the short vowel. However, an adjective derived from decor, decó̄ris with the suffix -oso- would differ only in vowel length. When long and short o merged in some regional forms of Latin, decorosus may have been seen as associated with decus, decor or both. For this reason, I have included decor in parentheses as a variant of decus.^[28]

We have established that adjectives that look like dolorosus can be derived from either masculine or neuter s-stems, and that the nominative singular of several of those s-stems have variant forms. We have also established that the geographical area where dolus ‘suffering’ occurred saw a merger of long and short o, making it impossible to tell whether adjectives ending in -orosus were derived from a neuter or masculine s-stem. This presents a number of avenues for proportional or four-part analogy (Blevins and Blevins 2009).

The fact that these adjectives could go back to nouns ending in -us may lead to the following analogy:

litorosus: dolorosus == litus: dolus

The existence of variant forms, such as decor/decus, where both suffixes are used, may spawn similar variant forms:

decor: dolor == decus: dolus

In both cases, the resulting dolus would not be declined as dolus, doli ‘deceit’, an o-stem, but would instead be a neuter s-stem, declined like corpus:

For Latin speakers, the stress placement would differentiate this from the paradigm of dolor, but to the modern scholar who only has access to the written record, it would be difficult to identify this as a paradigm distinct from dolus and dolor. The nominative singular would look just like dolus, doli ‘suffering’, and most oblique forms would be identified as forms of dolor. While some attestations of the nominative dolus are paired with adjectives that make the gender clear (examples [4], [9], [10]), others have nothing to indicate their gender (examples [6a], [6b], [8]). The only forms that would differ in writing would be the accusative singular and the nominative/accusative plural. As mentioned above, the instances of abstract nouns in the singular far outweigh the instances in the plural, so *dolora would be comparatively uncommon. In the material underlying Table 3, the form dolores, which covers the same cases as *dolora would, was the most common plural case-form, but still only made up 9.5 % of the total 4,401 attestations in LatinISE.

How then did the proposed *dolus, dólŏris become dolus, doli? This would require two (often related) features to change – gender and declension. While dictionaries and grammars give the impression of a clear-cut system, and deviations from it are examples of ‘substandard’ Latin,^[29] there were variations in grammatical gender already in Classical sources. Vulgus ‘the multiple, the people’, which occurs in both the masculine (e.g. Verg. Aen. 2.99) and the neuter (e.g. Cic. Att. 2.22.3). Vocalic stems show alternation in both gender and declension. The above-mentioned words for ‘milk pail’, the feminine mulctra and the neuter mulctrum, is such an example. O-stems may have both masculine and neuter forms, such as acinus and acinum ‘small berry’. Indeed, this alternation is seen in dolus ‘deceit’, where the form dolum is attested, like in an early fragment of a Bible translation:

TLL does not list any instances of dolum ‘suffering’, but only one of the accusatives dolum can be securely identified as masculine ([14f]). The other examples of dolum ([12], [14a], [14c], [14d], [14g]) could conceivably be neuter.

There is also evidence of Late Latin speakers reassigning words to other declensions. In the Mulomedicina Chironis, the noun armus ‘shoulder of a quadruped’, in Classical Latin a masculine o-stem (e.g. Ov. met. 10.112), is treated as a neuter s-stem armora (Mulomedicina Chironis 19, 241, 580 = Oder 1901: 9, 72, 187; cf. Weiss 2020: 556 n.9). This would have been created by a process such as this:

corpus: armus == corpora: armora

There is also evidence of change in the other direction. In one manuscript (A) of the fourth-century pilgrim Egeria’s Itinerarium, the ablative of corpus is given as corpo (CSEL 39.71.4 = Geyer 1898: 71, where it is corrected to corpore; cf. Weiss 2020: 556 n.8), prompted by a four-part analogy such as:

lupus: corpus == lupo: corpo

In the same way, we may posit:

lupus: dolus == lupi: doli

The process would then be:

However, there is also an important pull-factor – the existing word dolus, doli. While *dolus, dólŏris may have been in use for some speakers, for others, it may have been simply a stepping-stone which was not used.

The semantic change that saw dolus take on the meanings of dolor was fuelled by analogy, completely or in part. While we can identify analogy broadly as the driving force, we cannot pick out one source of analogy as the only one causing the change. It is likely that many different speakers hit upon dolus ‘suffering’ independently. Children are frequently identified as the source of many semantic changes, but these processes continue throughout life. Hearing someone use dolus to mean ‘suffering’ for the first time may trigger one or several of the four-part analogies described here. Children learning the word dolus ‘suffering’ would have further established it.

The study of semantics poses unique challenges to linguists. We have direct access to phonetics, morphology and syntax (although for the scholar of corpus languages, phonetics and phonology are not as accessible as the other features). Though these features are part of a speaker’s internal grammar, there are linguistic approaches to describe them, such as minimal pairs. However, word meanings cannot be quantified in the same ways as e.g. phonemes. The study of semantics requires other approaches, not least because semantics are closely associated with culture.

If we are studying a language with living speakers, we could distribute questionnaires and do in-depth interviews, but even if we do that, we are unlikely to be able to speak to every speaker. Even in a hypothetical scenario where we could interview every speaker, we would not get an objective look at the semantics of all words, and not only because of the time it would take. The vocabulary used depends not only on the speaker, but the interlocutor, and the relationship between them. Taboos may prevent the discussions of certain aspects of semantics, whether that be meanings that are considered too intimate or meanings with religious or cultural significance that are not seen as appropriate to discuss. In the case of corpus languages, our opportunities are even more limited. As discussed above, extant Latin material skews heavily towards free, elite men, who were most often from Rome and its environs and who up until at least 300 CE were most often polytheists who worshipped the Graeco-Roman gods. There are likely to be many examples of polysemy and homonymy that have been lost due to this, as these writers were not aware of them or did not think them appropriate to write down. There are cases where such meanings have been recorded, such as when Varro as part of an antiquarian argument mentions that wet-nurses used the word porcus to mean ‘vulva’ (rust. 2.4.12).^[30] The survival of this case of polysemy is essentially a fluke, and there must be many such examples that are lost to us.

Speech communities are not formless masses, but rather huge numbers of speakers functioning within a number of complex systems – families, generations, neighbourhoods, professional networks, ethnic groups, religions and so on. The lexica of individuals will not be identical, not only by virtue of them being individuals but because of the different groups they are part of. As a result, the same vocabulary item may have different meanings to different people for a variety of reasons. Different semantic meanings can be a regional feature. The word grina means ‘laugh’ in the southern parts of Sweden but ‘cry’ in central and northern Sweden. The word is polysemous only in the sense that both senses are listed in the dictionary, but few speakers will use both senses.^[31] Words can also have distinct meanings among groups we tend not to describe as speech communities. Work jargon is not confined only to specialised vocabulary, but also informal words and alternative meanings. To someone not familiar with the vocabulary of the epigraphist, for instance, a sentence such as ‘I will do an autopsy and make a squeeze’ is impenetrable.

Not only are cultural aspects important to understanding the existence of polysemy – it shapes any semantic hierarchies that we as linguists try to establish. What is considered a core meaning and what is seen as slang, ‘niche’ or ‘incorrect’ is dependent on linguistic hegemonies. Does read mean ‘to insult a peer for the purpose of humour and bonding’? It is not listed as a meaning in the OED Online or Merriam-Webster, but this meaning is in use in some LGBT communities, having its origins in the Ballroom Culture, a predominantly Black and Latino queer subculture.^[32] Therefore it would be incorrect to say that read does not have that meaning.

The value judgements of semantic variation are closely tied to power structures, both past and present. When studying the ancient world, these complexities often elude us, as written sources seldom represent them well, and what was written down may have been lost. Elite literary sources often outnumber and overshadow writing that has survived. Leaving out the many examples of dolus ‘suffering’ in favour of Augustine’s complaints about incompetent speakers reinforces the idea of linguistic variation as a form of decay. When we move beyond this narrative, the interaction between dolus and dolor becomes an example of something else – the plasticity of derivational morphology and the lexical creativity of speakers.