TLG Technical Note 001:
Greek Word Definition

Authored: Nick Nicholas, TLG
Maintained by tlg@uci.edu
Created: October 1999
Last Revised: 2002-11-23

The TLG maintains a word index of the Greek words occurring in its texts. Versions of the word index appear on the CD ROMs published by the TLG, and on the online TLG Search Engine. The following documents what constitutes a distinct word in the compilation of the TLG word index.

1. Definitions

Different versions of the TLG word index are referred to in the following:

2. Word delimiters

A new word is assumed to occur whenever a blank, a dash ( ; Beta code _ ), a punctuation sign, or a new line occurs in the source text.

Punctuation is defined as one of the following characters: However, %1 (Roman question mark: ?), which may indicate a doubtful letter, is not considered to be punctuation.

Nor are quotation marks, which may on occasion be used as editorial brackets, or as continuing block quotation markers.

Comma is only considered punctuation when not followed immediately by a letter; otherwise, it is considered a hypodiastole.

On the Web Index, a new word is also assumed to occur whenever an apostrophe occurs in the source text; the current TLG orthographic norms do not admit word-internal apostrophe. Thus, D'O(/S δ'ὅς is considered two words, D' and O(/S. This also applies where the apostrophe has supplanted the coronis: G'OUN γ'ουν (normally GOU)=N γοὖν) is still analysed as the two words G' and OUN.

3. Normalization

3.1. Punctuation

Punctuation is normally stripped out of the word. (See below for exceptions.)

3.2. Hyphenation and Non-Text

Hyphenated words are joined; words in non-text brackets ({}) are indexed separately from words outside non-text brackets, and do not interfere with hyphenation. For example,

A)/N- {STR.}
QRWPOS
ἄν-         στρ.
θρωπος
is indexed as A)/NQRWPOS ἄνθρωπος, and the marginal note STR στρ is indexed separately.

There is one exception: {27 }27 (formerly occurring in work 1595.107), indicating apograph textual emendation, are treated as brackets, and are thus ignored in extracting words. For example, ARIS{27T}27WN αριστ*ων is indexed as ARISTWN, not ARISWN, T. (The escape code has been reassigned as of August 2000, so this exception is no longer relevant.)

All other non-text brackets by definition contain text extraneous to the main text (e.g. marginalia, stage directions, titles, editorial emendations); so their content is never concatenated with main text words, even if no space intervenes.

3.3. Roman Script

Words in other languages are ignored in the word index; this currently concerns only words tagged as being in Roman script, since words from other languages transliterated into Greek are not yet tagged distinctly. Thus, in a text like *)APOMA/SAR &[1addidi]1 $E)/FH ᾿Απομάσαρ (addidi) ἔφη, only the ‘Greek’ words A)POMA/SAR ‘Abu-Masar’ and E)/FH are indexed.

3.4. Beta escapes

Characters other than elision markers, the Greek alphabet, breathings, accents, iota subscripts, and commas—i.e. almost all beta escapes—are also ignored. For example, "6$10A)$%5N?<1[H/]R>1#13 «ν̣|[ή]ρ is indexed as A)NH/R ἀνήρ .

Characters within a word in non-text or Roman font are also ignored (but see Partial Words below.)

3.5. Diacritics

Accents are regularized: grave accents become acute, only the first accent in a word is retained, and words are converted to lower case. Thus,

3.6. Hypodiastolae

Since commas are potentially hypodiastolae, they are retained in the word, and are stripped out only if they do not correspond to a known instance of a hypodiastole; thus, O(/,TI ὅ,τι is distinguished from O(/TI ὅτι. There are nine words with hypodiastolae known in our texts:

O(/,TI, O(/,TIPER, O(/,TTI, TO/,TE, O(/,TE, O(/,TOU, TA/,TE, TO/,T', H(/,TE
(ὅ,τι, ὅ,τιπερ, ὅ,ττι, τό,τε, ὅ,τε, ὅ,του, τά,τε, τό,τ’, ἥ,τε)

Hypodiastolae are new to CD ROM #E; in CD ROM #D, they were stripped out of the word.

3.7. Coronides

All crasis markers in words are now retained; this is a change from CD ROM #D, in which crasis markers were dropped if they occurred after the first syllable of the word. Thus, KALOKA)/GAQOS καλοκἄγαθος is represented in the CD ROM #D index as KALOKA/GAQOS καλοκάγαθος, but KA)GW/ κἀγώ remains KA)GW/; both forms retain their crases on CD ROM #E.

On CD ROM #E the coronis is not normalized: the rough coronis is treated as distinct from the smooth coronis, although these are mere notational variants. Thus, XW(S χὡς is listed as a distinct word from XW)S χὠς. The two coronides are conflated on the online version of the index. However, a rough breathing is considered a coronis only when following an aspirated stop; otherwise, it is an internal breathing, and left as is (e.g. *)ABRAA(/M ᾿Αβραἃμ)

3.8. Internal breathing marks

The internal breathing marks on double rho are removed only if they are predictable; thus, R)R( is converted to RR, but R(R( is left as is. For example, A)/R)R(WSTOS ἄῤῥωστος is indexed as A)/RRWSTOS ἄρρωστος.

This is another change from CD ROM #D, which also converted R(R(, R(R) and R)R) to RR.

3.9. Character Stacking

On the web index, instances of characters superposed above other characters, as tagged with the Beta escapes <10 ... >10<11 ... >11 are deemed to be textual variants, and two versions of the word are recorded in the word index: that with the above reading, and that with the below reading. Thus, the string A)<10QH=>10<11MU/>11NAI μύθῆναι is parsed as two words in the same spot, A)QH=NAI and A)MU/NAI.

3.10. Dittography

On the web index, if a bracket is known to indicate dittography or editorial or scribal deletion for a work, the indication is respected, and the marked portion of the word is excluded from the indexed word. If however the deletion bracket incorporates a word in its entirety, the word is indexed, as a variant reading. E.g. given the information that [3 ]3 { } in a given work are editorial deletions, E)/GW[3GW]3GE ἔγω{γω}γε is indexed as E)/GWGE ἔγωγε -- i.e. E)/GWGWGE ἔγωγεγε. However, in a text like E)GW\ [3DE\]3 EI)=PON ἐγὼ {δὲ} εἷπον, all three words are indexed.

4. Partial words

When a partial word is discovered in a text, the partial word is included in the current word index only if it contains two or more Greek letters (excluding diacritics) in sequence. This is at variance with CD ROM #D, which required three Greek letterals (including diacritics) in sequence. Thus, CD ROM #E includes word fragments like BA! βα. and excludes A)/! ἄ.; CD ROM #D would exclude BA! and include A)/!.

On CD ROM #D, only the missing letter code (!) was regarded as a partial word boundary. The repertory has been significantly expanded for CD ROM #E. A partial word boundary is discovered from the occurrence of:

or:

Thus,

However,

A-
@@B]
α-
    β]
does not have the hyphen as a partial word boundary: the @ act merely to indent, and not to indicate a lacuna. Likewise, the bracket pair in AB[1%1]1 αβ(?) does not indicate a lacuna: the presence of any code between brackets is taken to indicate the bracket pair is no lacuna --- unless the code itself denotes a lacuna (e.g. AB[1%40]1G αβ(˘)γ ).

Furthermore,

A-
$13]B
α-
β
contains a partial word boundary; the $13 is a font shift which can be ignored as a non-spacing Beta escape, not representing any actual text intervening between the hyphen and the right bracket. However,
A-
#13]B
α-
does not contain a partial word boundary, since the asteriskos is a spacing Beta escape, and is deemed to start the current text line.

By this reckoning, the following are incomplete words:

The following are complete words, by contrast: And the alternative readings A)MAMA/CUD[1%19OS, %19ES]1 (0009.001: ἀμαμάξυδ(-ος, -ες) = A)MAMA/CUDOS vel A)MAMA/CUDES) are indexed as A)MAMA/CUDOS, !ES.

If a beta escape denoting a lacuna occurs at the beginning of a word, it is deemed part of the word unless a blank, dash, or new line follows it: the words %40A ˘α and [%40]A [˘]α contain the breve, and thus are words with incomplete beginnings; but the word [%40] A [˘] α is not incomplete, since the breve is considered a separate word.

Exceptionally, as of November 2002, if a lacuna is followed immediately by a vowel with a breathing mark, the word is deemed complete. This is because the breathing mark is deemed to mark the beginning of a word rather than a coronis, and the source texts are rarely consistent in delimiting words from lacunae. For instance, !!!A)NH\R ...ἀνήρ is treated as the complete word A)NH/R ἀνήρ .

5. Special hyphen rules

If a beta escape denoting a lacuna occurs on a new line after a hyphen, it is deemed part of the hyphenated word automatically; if a blank, dash or new line ensues, the word is then terminated. Thus,

A-
%40B
α-
˘β
is indexed as A!B = A!, !B, but
A-
%40 B
α-
˘ β
is indexed as A!, B.

A hyphenated word is also terminated immediately if a space follows a left bracket in a hyphenated word continuation; what follows that space is not deemed part of the word. For instance,

I(STORI/-
[     A)NAGKAI/AN]
ἱστορί-
[     ἀναγκαίαν]
is indexed as I(STORI/!, A)NAGKAI/AN—not I(STORI/A)NAGKAI/AN ἱστορίἀναγκαίαν .

If a hyphen follows a lower case letter within a portion of a word containing an unclosed left bracket, and the continuation of the word in the next line consists of a capital letter not followed by at least one more capital letter, then it is deemed impossible for the two fragments to be part of the same word, and the hyphen is considered a final partial word boundary. Thus, in an instance like

F[ILE-
*)AS]KLHPIOU=
φ[ιλε-
᾿Ασ]κληπιοῦ
(0020.004), the program refuses to join the two fragments, and indexes them as FILE!, *)ASKLHPIOU=. This results because the bracket in F[ILE- explicitly flags what follows in the word as fragmentary. The casing of *)ASKLHPIOU=, on the other hand, with its initial capital, indicates that it must constitute a new word. Without that casing, the algorithm would still join the two word fragments:
F[ILE-
S]KLHPIOU=
φ[ιλε-
σ]κληπιοῦ
is interpreted as FILESKLHPIOU= φιλεσκληπιοῦ, and
F[ILE-
*S]*K*L*H*P*I*O*U
φ[ιλε-
Σ]ΚΛΗΠΙΟΥ
as FILE*S*K*L*H*P*I*O*U φιλεΣΚΛΗΠΙΟΥ.

If a hyphen preceded by an unclosed left bracket is followed by a continuation word with a breathing mark, as distinct from a coronis (for which the heuristic is that the vowel with the breathing mark occur before any consonants in the line), it is likewise deemed impossible for the two words to be joined. Thus,

*KLEANAK[TID-
H(
Κλεανακ[τιδ-
(0009.003) is indexed as KLEANAKTID!, H(: by having a breathing mark, H( cannot be anything but the beginning of a new word.

The necessity of the bracket in such checks is shown by the instance of

PERI-
E(/LKESQAI
περι-
ἕλκεσθαι
in the non-fragmentary text 2734.013.

If a hyphen is followed in the same line by a letter or punctuation, the word is deemed to be incomplete, and is not joined with the next line. Thus, if a line ends in FILE-., that text is not joined with the next line, but is indexed as FILE! .

6. Ellipses & Abbreviations

An instance of more than one contiguous dot, either within or on the boundary of a word, is treated as a lacuna: thus, ..A, A.., and A..B are analyzed as !A, A!, and A!B = A!, !B. If a space intervenes between the multiple dots and the word, the dots are not treated as part of the word, which thus remains complete; this is how ellipses as punctuation are distinguished from lacunae. Thus, A)NH/R ... ἀνήρ ... is analyzed as A)NH/R, not A)NH/R!.

Multiple dots in the continuation of a hyphenated word are treated like the other lacunae: A- B... is indexed as AB!, and A- ...B as A!B. A single dot, on the other hand, terminates a word: A.B is indexed as A followed by B, and not as AB or A!B = A!, !B. If the dot was to represent a missing letter, it would have been encoded as ! in the Beta code instead.

Abbreviations are thus treated as separate words rather than a single word: e.g. K.O.K. κ.ο.κ. (= KAI\ OU)/TW| KAQECH=S καὶ οὔτῳ καθεξῆς ‘and so forth’) is indexed as K, O, K.

The one exception to this (new to CD ROM #E) is K.T.L. κ.τ.λ. (‘etc.’) which is indexed as the single word KTL κτλ (with no periods.)

The foregoing is an algorithmic approach to determining what constitutes a complete or partial word. It is by no means infallible, and several words will inevitably be indexed wrongly. Substantial further improvement of the process, however, can only be done manually.