1 Introduction
Greek has a very long tradition of historical spelling. It is, in that sense, anchored in the past: in the words of the introduction to this volume, users of Post-classical Koine Greek were “continuously looking over [their] shoulder at the frozen grammar of Classical Attic”, even as the grammar could only be frozen by later generations, who were able to construct an idealized past because it was no longer messily alive. ‘Later generations’ is deliberately vague: while some parts of the history of selecting (and constructing) Classical Attic as a model for later Greek, such as the Atticist movement, were conscious processes driven by elite individuals (Kim 2023, pp. 124ff.), others were more organic. This includes Greek’s orthographic conservatism, one aspect of which, the use of iota adscripts, I will be looking at in this chapter.
With a few minor exceptions,1 spelling norms for writing Koine Greek were the same as those for Classical Greek. In historical orthographies, spelling conventions are slow to change to keep pace with phonological changes (Coulmas 2003, pp. 96–102). This means that the gap between conservative written norms and the sounds of Greek increases with time, and the increasing size of that gap is clearest in writers who do not spell normatively. As a result, spelling variants have frequently been used as an indirect clue to variation and change in speech.
However, the importance of orthography for reconstructing phonology arguably also means that historical linguistics has not always considered orthography much in its own right (Evans 2012). In this chapter, I discuss two orthographic variables in a sample of around 200 Ancient Greek letters to illustrate some of the misconceptions this can encourage. The letters are all papyri (texts written on a reed surface) dated to between 50–250 CE, and sent, received and/or found at ancient Oxyrhynchus, Egypt. At this period, Egypt was under Roman control, and ancient Greek was the default written language.
Some but not all of the texts in this set contain adscripts, tremas, or both. Adscripts denote a specific kind of spelling variation, where the graphemes ⟨
Both variables expose problems with a chain of assumptions sometimes made about spelling in ancient Greek papyri. Crudely speaking, the reasoning runs as follows:
1. Ancient Greek spelling is standardized.
2. In a standardized orthography, well-educated people adhere to the standard.
3. Consequently, spelling variation is a sign of poor literacy.
4. Being less literate, variable spellers’ choices reflect speech sounds: they are phonemically motivated.
5. However, being less literate, variable spellers’ phonemically motivated choices may reflect language contact between spoken Greek and Egyptian.
6. This complicates analyses which aim to distinguish between spellings which reflect this language contact and spellings which reflect diachronic sound changes shared by all varieties of Greek.2
This sequence is a construction anchored in how linguistic normativity operates in the highly codified European national languages used for classical scholarship, and not in the different ways language standards operated in Greco-Roman antiquity, where drivers of standardization such as mass education and the nation-state did not exist (Introduction, this volume). Though there are things to unpack at every stage of the sequence, I will focus on steps 3 and 4. Through a case study of a single papyrus, P.Oxy. 12 1482, I will illustrate how the connection between spelling variation and illiteracy in papyri (step 3) is at least in part a back-formation: a concept from scholars’ own literate cultures transferred, sometimes uncritically, onto ancient texts. Indeed, the assumed link between variation and illiteracy is so strong that it can lead to an overly narrow interpretation of orthographic variation in Greek papyri, ignoring patterns that do not fit that narrative.
Because tremas and adscripts are not written systematically, they have received relatively little attention as text features in their own right, with the notable exception of Clarysse (1976) on adscripts and Ast (2017) on tremas among other diacritics. Perhaps unsurprisingly, this lack of regularity has been interpreted as more evidence of lack of literate sophistication in ancient writers. In an important monograph on the ancient book, Schubart (1921, p. 85) argues that “sie [d.h. die Schreiber] setzen die Punkte auf diese Vokale ohne Wahl, wie es ihnen gerade einfällt”.3 Such authoritative but unsupported claims remain influential, so that editors can feel that adscripts and tremas are used “in what can seem like a capricious manner” (Ast 2017, p. 147).
This is a shame, as where they occur, both features are used in ways which challenge associations between variation, error, and illiteracy. By looking at the distribution of tremas and adscripts, I hope to show narratives that equate orthographic variation with error require more nuance than they are sometimes given. For instance, the variation as error model is partly a result of a tendency in previous linguistic work on papyri to concentrate on spelling variation, i.e. variation in the character choices for representing a particular sound in writing. But if orthographies are treated as sets of conventions for how language is written (Rutkowska & Rössler 2012, p. 213), variation in these conventions extends to differences below character level (e.g. letterforms) and in non-character text features, e.g. punctuation, diacritics, and word spacing. As the example of P.Oxy. 12 1482 will show, tremas can exist happily in texts with atypically high spelling variation; this is significant because tremas are more frequently associated with literary and school texts aimed at pupils, especially Homer (Fournet 1995; Cribiore 1996, pp. 83–84), and in exercises written by them (Cribiore 2005, pp. 190–191). The presence of tremas in a text, however spelled, thus suggests more than elementary formal literacy.
If tremas challenge the link between variation and illiteracy, adscripts show that viewing spelling variation as error also depends on how spelling variation is defined. As with tremas, adscripts are a variable feature whose variability is not consistently reflected in editions. They are sometimes included in the transcription, sometimes in a critical apparatus, sometimes not noted, and at other times silently regularized to modern print conventions. The last three options can make print transcriptions appear more regular than the underlying papyri, helping to reinforce modern expectations of regular written norms which are partly a product of the editing process.
2 An Illiterate Writer?
The impetus for considering tremas and adscripts alongside what is traditionally treated as spelling error came from a caustic assessment of an ancient writer’s literacy. In the second century CE, a man called Moros sent a letter that was discarded, preserved, excavated some 1,700 years later, and published in 1916 as P.Oxy. 12 1482. It may or may not have been written by someone else, since using scribes was common in Roman Egypt, whether the sender wrote Greek confidently, hesitantly, or not at all. For isolated documents like this, it is hard to know whether the sender was also the writer, so I will call the latter X, to avoid paraphrases later on. Whoever X was, their writing ability came in for the full weight of scholarly disapproval. According to Grenfell and Hunt, its original editors, the papyrus displays “the rude uncial of an illiterate writer, who makes numerous mistakes of spelling in spite of several corrections” (Grenfell and Hunt 1916, p. 241).
The rude uncial of an illiterate writer is a memorable phrase, all the more so because it makes little sense. The letter, all in one hand, is well preserved. It contains 160 words over 26 lines, enough to say that X’s handwriting does not display the characteristics of Greek papyrus texts produced by inexperienced adult writers. These include:
– letters formed separately;
– unconfident letter shapes, sometimes formed by many strokes (‘multistroke’);
– large spaces between letters;
– lack of uniformity (e.g. uneven lines and margins, letter shapes, and direction of writing).
Cribiore 1996, pp. 103; 116–117
P.Koeln 10 419 (third-fourth century CE), a letter which does contain many of these features, is a useful comparison here. Features that display lack of writing confidence in P.Koeln 10 419 include unequal letter size and spacing, multistroke letter shapes, different amounts of ink used per letter, and writing that does not keep to a consistent line. I have annotated some of these features on the image of lines 15–17 below.
By contrast, P.Oxy. 12 1482 displays none of these features. Instead, it shows even lines, consistent letter shape and spacing, and some letter sequences joined together through ligatures (e.g. the sequence 
,
An image of the first three lines is reproduced in Figure 2.3, but as it is of far lower quality than the one for P.Koeln 10 419, the differences between the two papyri are perhaps clearest when comparing the texts of both papyri traced from the images, rather than the images themselves. Partially legible letters are shown, with dotted lines for the missing portions. Lost letters are not marked. In P.Oxy. 12 1482, l. 1, the image shows what appears to be a small vertical line after
Low literacy features in P.Koeln 10 419, ll. 15–17
Comparison of hands in P.Koeln 10 419 and P.Oxy. 12 1482
The difference in letter sizing, spacing, shape, and direction is immediately visible when comparing P.Koeln 10 419 on the left against P.Oxy. 12 1482 on the right. The contrast with the hand of a truly unconfident writer, as in P.Koeln 10 419, shows that X’s ‘rude uncial’ was, in fact, perfectly fluent. The image of P.Oxy. 12 1482, ll. 1–3 also shows X using some of the genre conventions of Greek papyrus letters from Egypt, namely word spacing and decorative word arrangement in the opening address (Sarri 2017, pp. 114–120), with khairein (l. 2) centered and occupying a line to itself:
Someone who can pen a whole letter fluently, and is familiar with genre layout conventions, can hardly be called illiterate in the sense of struggling to write. The level of literacy represented by fluency should not be underestimated in Roman Egypt. Also, in the second century CE, Petaus, a village scribe charged with investigating concerns about his colleague Ischyrion’s literacy, reported back to the strategos that Ischyrion was not illiterate (agrammatos) because he was able to sign his own name (Kraus 2000, pp. 329–336). Petaus, in fact, struggled to sign his name consistently himself, but that does not detract from the fact that he thought the ability to sign one’s name an acceptable definition of literacy in a document intended for administrative superiors.
Spending some time looking at P.Oxy. 12 1482’s supposedly rude hand, the first component in Grenfell and Hunt’s negative assessment of X’s literacy shows that the epithet is unjustified. It also calls for a reassessment of their second reason for calling X an illiterate writer: irregular spelling.
X’s spelling is indeed variable. 26 of the papyrus’ 160 lacuna-free words (or 16%) are marked as spelling variants by the editors, in addition to X’s own corrections. It is also true that, seen as a whole, Greek papyrus spelling is rather consistent: the mean correction rate for the c.60,000 documents available through the Papyrological Navigator (meta)database (PN; https://papyri.info/) is 2%, and the median 2.3% (Depauw & Stolk 2015, pp. 203–204).
X’s letter is thus substantially more variable than average, and also exhibits a range of variation patterns, including:
As shown in the table, there are phonemic factors involved in all these patterns, and ⟨
Phonemically, for example, the variants in P.Oxy. 12 1482 show that for X /i/ corresponded to at least three graphic units: {⟨
In this situation, where a sound, x, corresponds to the written units {a,b,c}, a variable speller could potentially select any of {a,b,c} any time x occurs. Every occurrence of a, b, or c is therefore linguistically interesting, whether the choice, editorially speaking, is a spelling variant or not. Returning to lines 1–3, I have marked every occurrence of ⟨
Seen from this point of view, an iota is an iota. With no apparatus or typographic conventions to set any instance apart from the others, it is easier to see that spelling variation (editorially, whether an iota should be written) sits alongside variation in whether it is written (
3 Dataset
Given the wide range of variation in adscript and trema usage from writer to writer, patterns which hold above an individual level are more likely to emerge from studying a coherent set of documents than by focusing on separate documents with little known context to connect them. Since I am interested in patterns that are shared among writers rather than individual use, I have chosen to restrict other variables (time, date range, and place) and leave authorship open. The sample of texts I used consists of documents which matched particular criteria on a search of the Papyrological Navigator (PN) in March 2018.
This is, rhetorically, its own form of construction: as Sluiter (2017, pp. 31–32) points out, appeals to technology are often used to frame a research method as an advance, even though ‘new’ is not necessarily better, or even new at all. Empiricist approaches to language have a long tradition, and even computer- assisted corpus or corpus-based work is now over 60 years old (Jenset & McGillivray 2017, pp. 67–73). Quantifying spelling variation by using digitized transcriptions is not valuable because it is newfangled; it is valuable because it makes it possible to see patterns in data in a new way, and thus challenge—or confirm—earlier hypotheses. Having anchored using digital tools as an appropriately classical way of innovating to refresh tradition (new is the new old), let me describe my method. The search targeted texts:
in Ancient Greek;
whose provenance was given as Oxyrhynchus (i.e. found and/or sent there);
dated strictly between 50–250 CE;
whose metadata matched one or more search terms in English, German, Italian or French used by editors to describe letters.4
The search retrieved 202 matches. 8 were duplicates or not (electronically) transcribed, leaving a sample of 194 texts.
Letters are an inherently fuzzy category, and several of the documents have been called letters by one group of editors and something else by another. P.Oxy. 77 5111, for instance, is entitled Petition of Temple Personnel in the edition, and Schreiben (amtlich) ((official) correspondence) in the Heidelberger Gesamtverzeichnis data archive. All the same, I have included all the results, regardless of their status as letters, since the main focus was on looking at tremas and adscripts overall rather than grouping them by document type.
To count the number of tremas and adscripts, I relied mainly on checking XML-encoded transcriptions accessed through PN against digital or print images of the document. Access to images matters, since changes introduced by remediation (transferring information from one format to another (Kichuk 2007)) can mean print editions can differ from what can be read from the papyrus or an image of it. Editorial practice on noting tremas and adscripts varies: P.Oxy., for instance, has from Volume 1 in theory always written iotas adscript if present on the papyrus, and regularized them to subscript if not (P.Oxy. 1, p. xvi). Documents edited in condensed form, however, may in fact regularize both <ω> and <ωι> to
A further reason for checking images is that PN’s XML files do not always encode these features, particularly tremas, in the same way as print editions. For example, change logs on the XML files sometimes record that editorial diaereses were missed in a previous version of the PN record. So, a change made to P.Oxy. 63 4353 (not in the sample) on 29 March 2011 specifies that 8 editorial tremas have been added to the file:
In simple terms, therefore, I relied on images to check for the presence of tremas and adscripts where available, checking for tremas on all documents and adscripts when not editorially marked. Where images were not available, I accepted tremas and adscripts noted by editors as printed. If editors of texts without an image did not mark tremas or adscripts, I excluded these texts from my calculations, since there was no way of checking whether the papyrus lacked these features, or whether they had simply not been transcribed.
4 Adscripts
4.1 Background
Iota adscripts occur after ⟨
This means that where adscripts are written in etymologically correct positions in my texts, the choice of ⟨
the opening portions of document types where a sender addresses a recipient (e.g. letters and petitions);
delivery addresses, indicating where the document was to be sent.
A secondary complication is that modern print conventions for representing etymological ōi, ēi, ai in Ancient Greek differ from papyrus ones. Figure 2.7 shows three options for representing ōi:
Modern print editions use options 1 or 2 (with diacritics). Iota subscripts are a modern print and medieval manuscript convention, generally dated to the twelfth century CE onwards; they are vanishingly rare in papyri and pre-medieval inscriptions (for two exceptions, see Clarysse 1976, p. 151, nn. 6–7). Papyrus letters themselves use options 2 or 3, generally without diacritics.
By convention, any ancient forms using option 3 are regularized to option 1 in print editions: thus e.g. adelphō (papyrus
4.2 Textual Patterns
Where editors use options 1 and 2 in the same text, I have assumed that the difference between subscript and adscript iotas is meaningful and consistent: i.e. that if adscript
Of the 194 documents, 90 contained at least one adscript marked by the editor. This left 104 documents where adscripts are not editorially marked. I have again excluded documents which only print subscripts and do not have an image available. There were 38 of these, leaving 73 texts for which I checked the images manually. I found only one adscript not marked by the editor:
This makes a total of 91 total documents in which an adscript was identified. These 91 documents contained a total of 290 adscripts. As expected, these are not evenly distributed. In absolute terms, there is one clear outlier: whereas the other documents range between 1–11 adscripts, SB 22.15708 contains 22, as illustrated in Figure 2.9:
Distribution of adscripts in 193 papyri (excluding TM 20506)
However, absolute numbers do not say anything about the proportion of words with adscripts in each text, particularly as SB 22 15708 is also the second-longest document in the sample in terms of both total and lacuna- free words. Proportionally, SB 22 15708 is much less remarkable: just under 5% of its words are adscripted, whereas 12 texts have proportions of 10% or above.7
All of these 12 texts are short. None is longer than 75 editorial words, which includes partially or fully restored wordforms, below both the mean (109 words) and median (82.5) for the dataset. Seven are short because of extensive damage to the papyrus and can be discounted. Five, however, are both materially and textually fairly complete.8 Concentrating on these, there is a tendency for adscripts to be written in the first few and final lines of the text, matching Youtie’s and Clarysse’s observations about where adscripts occur.
This pattern holds out across the dataset: 39% (113 / 290) of all adscripts occur in the first 2 lines, and 52% (152 / 290) in the first 4. The high proportion of adscripts in these 5 texts is, therefore, partly a function of their length: the shorter a text is, the more of it is made up by the opening greeting and delivery address. This makes SB 22 15708 stand out again, since its proportionally high score is attributable to adscripts occurring regularly in the body of the letter, i.e. in a portion of the text where it is typically less frequent. ¾ of the texts in the sample have fewer than 130.5 words; of the 52 texts which are longer, only 5 have an adscript proportion of 4% or above.9
These texts are also unusual in other ways. Two are associated with cultured contexts: P.Oxy. 18 2192 is a letter about books, and the sender of P.Oxy. 73 4959, Ammonios, was an ex-gymnasiarch, a position of considerable social prestige. The set also account for 3 of the 5 instances of adscripted -āi,10 which is underrepresented. This provides a tentative suggestion to explore in future work, namely that adscript -āi may potentially be more of a clue to a writer’s formal educational level than -ōi or -ēi. Character by character, -
This frame provides three dative slots, all of which provide the opportunity for writing adscript forms. (In the example, all three opportunities are realized, but that is not always the case.) Most recipients in the sample are men. -
Typical opening greetings as exemplified by P.Oxy. 47 3357, ll. 1–2
The gender skew towards male recipients in papyrus letters thus helps account for the greater frequency of -
Comparison of word-final adscripts and subscripts
These numbers are very crude and would need further refinement for analysis. For example, the table doesn’t separate nominal from verbal forms, both of which can end in adscript slots. It also includes hypercorrect spellings in forms where there was no etymological long diphthong.11 Despite their noisiness, however, the data suggest some hypotheses to explore in future work.
One is that the proportionally higher occurrence of
Morphological ambiguity therefore appears to decrease rather than increase the proportional rate of adscript use. Final -
Further qualitative support for the importance of predictability comes from a trend in the sample. Of the 30 documents which preserve only one adscript, 24 place it on the masculine dative singular definite article,
[sender’s name].NOM
τῶι [recipient’s name].DAT, [verb phrase]
assuming the recipient is a man.
It would be interesting to see whether the greater predictability of -
Adscripts in 194 papyri by inflection type
As ever, raw frequency counts are problematic: virtually every letter will contain several definite articles which the writer may or may not choose to write a particular way, whereas not every letter contains a subjunctive. Without a fully part-of-speech tagged dataset, identifying the proportion of subjunctives that are adscripted is error-prone, as (ironically) many wordforms are morphologically ambiguous. I am planning to tag my dataset in the next stage of my work. Even without tagging, however, there is something other than raw frequency to suggest that writers are less comfortable adscripting verb forms than substantives. 9 of the 18 adscripted verbs are in fact etymologically incorrect, with adscripts used on indicative forms.
Overall, the distribution of adscripts thus shows a preference for substantive endings that are morphologically unambiguous. For verb forms, frequencies are low, but the rate of irregularity suggests something similar: morphological ambiguity causes difficulty in writing adscripts in historically appropriate circumstances. The preference for unambiguous morphemes (i.e. dative -
5 Tremas
On this analysis, adscripts can be seen as a case of deliberate spelling variation which conveys respect to the reader. Tremas can also be interpreted as a reader-oriented feature, though perhaps in a more textual and less social sense. Most non-literary papyri are written largely without orthographic word division (a practice termed scriptio continua). In letters, (some) word spacing is fairly common, though by no means the default. Like adscripts, word spacing is concentrated in opening greetings, as illustrated in Figure 2.3.
A few letters in the sample use tremas to signal they are not following optional adscript and spacing conventions. In texts without word spacing, pen lifts are unlikely to coincide with word breaks, and may clash with them. P.Oxy. 41 2956 l. 22, for example, contains the sequence
Space / word ambiguity in P.Oxy. 41 2956, l. 22
5.1 Background
This use of a trema as a word separator ties in well with an observation made in previous discussions of tremas in Greek papyrus documents: that they are more frequent at the beginning of words (so-called inorganic diaereses) than word-internally (organic diaereses) (Cribiore 1996, pp. 83–84; Ast 2017, p. 151, n. 29).
Tremas are less common in the sample than adscripts, occurring 124 times over 57 texts. These counts were derived from checking all papyri with images, since editions differ in noting tremas. The majority of the tremas, 106, are word- initial. However, saying tremas are more common word-initially (a description of occurrence patterns) is not quite the same as saying they function as word separators. When it comes to describing the function of tremas, scholars have tended to comment, if at all, on their “errati[c]” use (Cribiore 1994, p. 2). Part of the problem may be an expectation that tremas, when they occur, all signal the same thing. My discussion of tremas here will be tentative, as the overall figures are lower than for adscripts, but it appears they fulfil a variety of functions, some of which stand out more clearly from raw numbers than others.
5.2 Textual Patterns
While the trema in P.Oxy. 41 2956, l. 22 is apparently acting as a word separator, there is good reason to think the role of tremas is not that straightforward. For a start, why do tremas overwhelmingly occur over only two graphemes, iota ⟨
Such usage, however, appears to be genuinely rare. All the tremas in the sample are over iota (92 tokens) or upsilon (32 tokens). Separately, I also looked at 4,420 papyri, from all time periods, which contain the wordform khairein (GREET.PRES.IND.ACT.INF.). This is frequently, though not exclusively, associated with letters. As many of the letters in the smaller sample also contain khairein, the larger set of documents overlaps with the smaller one. 871 of the documents in the larger set contain at least one trema encoded in PN XML files using the diaeresis attribute:
<hi rend=“diaeresis”>.
There were 3,753 editorial tremas across these 871 documents, of which iota and upsilon together account for over 99% of the total:
The larger dataset also shows tremas occurring more frequently over iota than upsilon. Without knowing character counts (i.e. the frequency of iota and upsilon as characters), this may or may not be a meaningful difference. In the next stage of my work, I am gathering character counts to see whether tremas occur over a greater proportion of iotas. Nevertheless, the difference in raw frequencies makes it worth thinking about whether there are any qualitative differences in the contexts in which iota and upsilon appear.
Returning to the sample of 194 letters, it turns out that tremas tend to cluster on a relatively small number of lexical items. 21 of the 117 total tremas in these documents (18%) appear on the subordinator
The three leftmost columns mark up particular features of the character the tremas occur on. At first, there appears to be no neat correspondence between e.g. word position, aspiration, and character. However, if iota and upsilon are treated distinctly, it is possible to develop some hypotheses about usage patterns.
Starting with hina, it is interesting that the item most frequently equipped with a trema should be such a short word. 21 tokens with tremas is a remarkably high proportion of the 25 total instances of editorial hina, both read and restored. One hypothesis is that tremas become associated with
ιν for acc.sg. of -i-stem substantives (e.g. polin, CITY.ACC.SG.FEM.);ι (ν ) in various 3rd person plural verb forms (e.g. phēsi(n), SAY.3.SG.PRES. IND.ACT.);ειν , an infinitival suffix (e.g. graph|ein, WRITE.PRES.INF.ACT., or graps|ein, WRITE.FUT.INF.ACT.).
Given the range of suffixes that the sequence -
Hina is frequent enough as an individual wordform to show a clear pattern for tremas on iotas: marking word division in morphologically ambiguous written sequences. This pattern is not restricted to hina, as shown by tō|idiō, but neither is it the only pattern. The other major use of tremas in the dataset, in proper names, generally involves clarifying syllable division within words rather than boundaries between them. 33 of the 124 total tremas occur in proper names, all on iota. 17 of these occur word-medially, e.g. in the male name Zōilos. As seen above with adscripts, the sequences -ōi- and -ō could both correspond to /o/ at this period. The spelling Zōïlos helps to disambiguate the sequence, indicating that the iota is not part of a digraph, which would produce /zolos/, but instead an independent grapheme (producing /zoilos/).
While this ‘organic’ use of diaereses is well-known (Turner 1987, pp. 10–11), it is interesting to note that 11 of the 17 medial tremas in the sample occur in etymologically non-Greek names, both Egyptian (e.g. Phmoïs, Thaïs, Talobaïs, Hēraïs) and Latin (alternative spellings of Flauius, e.g.
Seen in this way, the use of tremas, if not systematic, is internally consistent. In texts written largely without word division, the distinction between organic (medial) and inorganic (initial) diaereses based on lexical position is arguably less relevant than their functional similarity. Whether medial or initial, tremas most commonly act as sequence separators, clarifying boundaries between syllables and words. Used as sequence separators, tremas are also a clear sign of a writer’s textual awareness. Since their primary purpose is to assist a reader by removing the ambiguity from a sequence where there are multiple ways of decoding spelling to sound, tremas are in itself an acknowledgement of how language as written differs from speech.
This does not mean that all tremas function neatly as separators between sequences that could be vowel digraphs. Initial upsilon with trema appears both in sequences that form graphically ambiguous combinations across word boundaries, e.g. -au- (/aw/ ~ /ay/) in proskunēmaumōn (
6 Conclusion
Looking beyond spelling irregularity to orthographic variation more broadly can thus prove interesting in a variety of ways. On the one hand, variation can reveal nuances of how writers perceived and dealt with phonemic ambiguity in ways that may not be obvious through spelling alone. In addition, the use of tremas to signal a mismatch between word spacing and word boundary, as in tō|ï diō, shows writers using graphic variation in an explicitly textual way. The same applies to adscripts. Even in writers whose confidence in selecting between ⟨
As the data also provides empirical support for previous observations that many writers only felt comfortable writing adscripts in openings and delivery addresses, documents which preserve multiple adscripts outside these document sections may imply higher levels of familiarity with literate conventions. These texts thus emerge as a fertile group in which to explore the relationship between non-phonemic and phonemic spelling variation. P.Oxy. 73 4959, for instance, contains both tremas and adscripts, and is atypical of the set for including adscript -ai. At the same time, these marks of text awareness sit alongside iotacistic interchanges (humein for regular humin, /ymin/).17
Looking at the distribution of orthographic variants can thus help build a more nuanced picture of the practice of literacy in Roman Egypt. Several of the patterns discussed here are only preliminary sketches, and there is plenty to do in filling them in. Sometimes, this work might result in revising estimates of writers’ literacy upwards, as spelling variation seen as error by editors clearly sits alongside non-error variation (adscripts) (e.g. adscripts) and trema use in texts which use them proficiently. At other times, it might lead to questions about what high or low literacy means. If the underuse of -ai compared to -ōi adscripts holds across a larger dataset, for example, this would confirm a potential area of uncertainty even for competent writers. Combining a quantitative approach to papyrus spelling with careful contextualization can thus help suggest gradations of literacy, clarifying what it initially seemed to complicate.
Appendix
Bibliography
Ast, R. (2017). Signs of Learning in Greek Documents: The Case of Spiritus Asper. In G. Nocchi Macedo & M. C. Scappaticcio (Eds.), Signes dans les textes, textes sur les signes. Érudition, lecture et écriture dans le monde gréco-romain (pp. 143–157). Presses Universitaires de Liège.
Clarysse, W. (1976). Notes on the Use of the Iota Adscript in the Third Century BC. Chronique d’Égypte, 51(101), 150–166.
Colvin, S. (2009). Greek Koine and the Logic of a Standard Language. In A. Georgakopoulou & M. S. Silk (Eds.), Standard Languages and Language Standards: Greek, Past and Present (pp. 33–45). Ashgate.
Coulmas, F. (2003). Writing Systems: An Introduction to Their Linguistic Analysis. Cambridge University Press.
Cribiore, R. (1994). A Homeric Writing Exercise and Reading Homer in School. TYCHE, 9, 1–8.
Cribiore, R. (1996). Writing, Teachers, and Students in Graeco-Roman Egypt. Scholars Press.
Cribiore, R. (2005). Gymnastics of the Mind: Greek Education in Hellenistic and Roman Egypt. Princeton University Press.
Culpeper, J., & Kytö, M. (2010). Early Modern English Dialogues: Spoken Interaction as Writing. Cambridge University Press.
Dahlgren, S. (2016). Towards a Definition of an Egyptian Greek Variety. Papers in Historical Phonology, 1, 190–108.
Dahlgren, S. (2017). Outcome of Long-Term Language Contact: Transfer of Egyptian Phonological Features onto Greek in Graeco-Roman Egypt [Doctoral dissertation]. University of Helsinki.
Depauw, M., & Stolk, J. (2015). Linguistic Variation in Greek Papyri. Greek, Roman and Byzantine Studies, 55, 196–220.
Evans, M. (2012). A Sociolinguistics of Early Modern Spelling? An Account of Queen Elizabeth I’s Correspondence. In J. Tyrkkö, M. Kilpiö, T. Nevalainen, & M. Rissanen (Eds.), Outposts of Historical Corpus Linguistics: From the Helsinki Corpus to a Proliferation of Resources [online]. https://varieng.helsinki.fi/series/volumes/10/evans/.
Fournet, J.-L. (1995). L’‘homérisme’ à l’époque protobyzantine: L’exemple de Dioscore d’Aphrodité. Ktèma, 20, 301–315.
Gignac, F. T. (1976). A Grammar of the Greek Papyri of the Roman and Byzantine Periods: Vol. 1. Phonology. La Goliardica.
Grenfell, B. P., & Hunt, A. S. (1916). The Oxyrhynchus papyri., Vol. 12. London: Egypt Exploration Fund.
Horrocks, G. (2010). Greek: A History of the Language and Its Speakers (2nd ed.). Wiley- Blackwell.
Jenset, G. B., & McGillivray, B. (2017). Quantitative Historical Linguistics: A Corpus Framework. Oxford University Press.
Kichuk, D. (2007). Metamorphosis: Remediation in Early English Books Online (EEBO). Literary and Linguistic Computing, 22(3), 291–303.
Kraus, T. J. (2000). (Il)literacy in Non-Literary Papyri from Graeco-Roman Egypt: Further Aspects of the Educational Ideal in Ancient Literary Sources and Modern Times. Mnemosyne, 53(3), 322–342.
Loprieno, A. (1995). Ancient Egyptian: A Linguistic Introduction. Cambridge University Press.
Milroy, J., & Milroy, L. (1999). Authority in Language. Routledge.
Parker, D. C. (2000). A New Oxyrhynchus Papyrus of Revelation: P115 (P. Oxy. 4499). New Testament Studies, 46(2), 159–174.
Rutkowska, H., & Rössler, P. (2012). Orthographic Variables. In J. M. Hernández-Campoy & J. C. Conde-Silvestre (Eds.), The Handbook of Historical Sociolinguistics (pp. 213– 236). Wiley-Blackwell.
Schubart, W. (1921). Das Buch bei den Griechen und Römern (2nd rev. ed.). Vereinigung Wissenschaftlicher Verleger.
Sluiter, I. (2017). Anchoring Innovation: A Classical Research Agenda. European Review, 25(1), 20–38.
Stolk, J. V. (2017). Dative and Genitive Case Interchange in Greek Papyri from Roman- Byzantine Egypt. Glotta, 93, 182–212.
Turner, E. G. (1987). Greek Manuscripts of the Ancient World (2nd rev. ed.). Institute of Classical Studies.
Van Guilder, L. (2007). Cross-Language Perception in Foreign Name Transcription [Doctoral dissertation]. Georgetown University.
Youtie, H. C. (1947). Parerga Papyrologica. Transactions and Proceedings of the American Philological Association, 78, 105–122.
The most famous is probably the spelling of Classical GIGNOMAI ‘become’, which is spelled ⟨ginomai⟩ in Koine Greek. Following practice in Early Modern English spelling research (cf. e.g. Culpeper & Kytö 2010, p. 199, n. 2), I will be using small caps for the citation forms of words, which, like a lemma, are superordinate: e.g. YOU includes you, u, yew. Angled brackets highlight specific forms under discussion, which may be variants. For example, I might well spell YOU ⟨u⟩ when messaging friends online.
For an overview of 1–3, see (Colvin 2009, pp. 34ff.); 4–6: (Horrocks 2010, pp. 111–113).
‘They [i.e. writers] place the points over these vowels indiscriminately, as the fancy takes them.’ Own translation.
The specific terms were: letter OR correspondence OR Brief OR Schreiben OR Korrespondenz OR lettre OR lettera.
Depauw and Stolk (2015) [accessed 28 June 2019] give 9 examples of second-century BCE texts with homologōi
Eidēis, though semantically present, is morphologically perfect: KNOW.2.SG.PERF.SUBJ.ACT.
These are: P.Oxy. 2 297; P.Oxy. 3 523; P.Oxy. 3 589; P.Oxy. 18 2184; P.Oxy. 41 2976; P.Oxy. 44 3176; P.Oxy. 47 3341; P.Oxy. 47 3357; P.Oxy. 75 5054; P.Oxy.Hels. 47c; SB 10 10275; SB 16 13058.
Truncated: P.Oxy. 3 589; P.Oxy. 18 2184; P.Oxy. 41 2976, P.Oxy.Hels. 47c; SB 10 1027; damaged text: P.Oxy. 44 3176; P.Oxy. 47 3341; (near-)complete: P.Oxy. 2 297; P.Oxy. 3 523; P.Oxy. 47 3357; P.Oxy. 75 5054; SB 16 13058.
P.Oxy. 1 113 (4.6%); P.Oxy. 47 3356 (5.9%); P.Oxy. 70 4773 (5.1%); SB 22 15708 (4.7%); P.Oxy. 73 4959 (4.6%).
P.Oxy. 73 4959, ll. 1, 5; SB 22 15708, l. 10.
E.g.
P.Oxy. 77 5111; PSI 4 281; P.Oxy. 12 1482; P.Oxy. 41 2996; P.Iand. 7 139; P.Oxy. 6 932; P.Oxy. 10 1294; P.Oxy. 14 1758; P.Oxy. 14 1760; P.Oxy. 47 3341; P.Oxy. 73 4959; PSI 121248; P.Amh. 2 136; P.Oxy. 14 1673.
⟨u⟩ is transliterated variously as ⟨
Outside the frequent lexeme huios ‘son’, where variants like ⟨uos⟩ (e.g. BGU 7 1559, ll. 2, 4) and ⟨ios⟩ (e.g. iois son.dat.PL., P.Sarap. 67, l. 10) suggest that ⟨
This particularly affects forms of huios (SON.NOM.SG.). In the larger sample, documents with this pattern are: PSI 15.1535; P.Oxy. 56.3855; PSI 14.1423; BGU 2.412; BGU 4.1080; P.Oxy. 14.1768; P.Oxy. 51.3612; SB 10.10725; BGU 20.2874.
For example, tremas may sometimes reflect quantity differences in formerly distinct phonemes whose quality had already merged to /i/. Parker (2000, p. 160) suggests this for a New Testament papyrus, and the larger sample has several cases of the conjunction ei ‘if’ (/i/, historically /e:/) written as
Ll. 4, 11, 11a, 14.