gpt-5.4 v3 Evaluation

Generated: 2026-09-20 11:21:59 UTC

Pairs100
Distinct lemmas100
Slope0.963
R^20.989
BLEU-452.8%
chrF++74.2%
METEOR80.1%
ROUGE-L79.8%
BERTScore96.6%
COMET80.1%
BLEURT76.0%

Profile version id: 2174

First translation using this version: 2026-07-14

Total usable AI runs for this version: 100 runs across 100 lemmas

Created: 2026-07-10 11:03:49.528703+10:00

Active: True

Notes: Model timeline evaluation: GPT-5.4 using prompt text from gpt-5.5 v3; approved-human-only for the 100-row Kappa corpus.

Prompt length scatter plot

Length Regression

FormulaAI words = intercept + slope * human words
Slope0.963
Intercept1.346
R^20.989
P-value3.71e-97
Pearson r0.994
Mean source words30.400
Length fallback rows0
Mean human words44.150
Mean AI words43.870
Mean absolute length residual2.921
Mean absolute percent error6.8%

Metric vs Passage Length

These plots test whether score changes with source passage length for this prompt version. The dashed line is an ordinary least-squares fit; the annotation reports n, r, R^2, p-value, and slope.

BLEU-4 vs passage length
BLEU-4: significant negative; r = -0.398, R^2 = 0.159, p = 4.06e-05.
chrF++ vs passage length
chrF++: significant negative; r = -0.414, R^2 = 0.172, p = 1.81e-05.
METEOR vs passage length
METEOR: significant negative; r = -0.459, R^2 = 0.210, p = 1.60e-06.
ROUGE-L vs passage length
ROUGE-L: significant negative; r = -0.438, R^2 = 0.192, p = 5.28e-06.
BERTScore vs passage length
BERTScore: significant negative; r = -0.632, R^2 = 0.399, p = 1.84e-12.
COMET vs passage length
COMET: significant negative; r = -0.450, R^2 = 0.202, p = 2.65e-06.
BLEURT vs passage length
BLEURT: significant negative; r = -0.690, R^2 = 0.476, p = 2.08e-15.
Trigram precision vs passage length
Trigram precision: significant negative; r = -0.303, R^2 = 0.092, p = 0.0022.
Trigram recall vs passage length
Trigram recall: significant negative; r = -0.311, R^2 = 0.096, p = 0.0017.
Trigram F1 vs passage length
Trigram F1: significant negative; r = -0.306, R^2 = 0.094, p = 0.0019.
Trigram Jaccard vs passage length
Trigram Jaccard: significant negative; r = -0.310, R^2 = 0.096, p = 0.0017.
Metric Rows Pattern Pearson r R^2 P-value Slope Status
BLEU-4 100 significant negative -0.398 0.159 4.06e-05 -0.00261 ok
chrF++ 100 significant negative -0.414 0.172 1.81e-05 -0.00158 ok
METEOR 100 significant negative -0.459 0.210 1.60e-06 -0.00188 ok
ROUGE-L 100 significant negative -0.438 0.192 5.28e-06 -0.00171 ok
BERTScore 100 significant negative -0.632 0.399 1.84e-12 -0.00044 ok
COMET 100 significant negative -0.450 0.202 2.65e-06 -0.00083 ok
BLEURT 100 significant negative -0.690 0.476 2.08e-15 -0.00251 ok
Trigram precision 100 significant negative -0.303 0.092 0.0022 -0.00239 ok
Trigram recall 100 significant negative -0.311 0.096 0.0017 -0.00238 ok
Trigram F1 100 significant negative -0.306 0.094 0.0019 -0.00237 ok
Trigram Jaccard 100 significant negative -0.310 0.096 0.0017 -0.00256 ok

Translation Similarity Metrics

Mean BLEU-452.8%
Mean chrF++74.2%
Mean METEOR80.1%
Mean ROUGE-L79.8%
Mean BERTScore96.6%
Mean COMET80.1%
Mean BLEURT76.0%
Legacy smoothed corpus BLEU52.6%
Unigram F180.4%
Bigram F158.3%
Trigram precision45.7%
Trigram recall45.4%
Trigram F145.5%
Trigram Jaccard29.5%
4-gram F136.2%

Largest Length Residuals

Lemma ID Headword Human words AI words Residual BLEU-4 chrF++ METEOR ROUGE-L BERTScore COMET BLEURT Human Run Model
2604 Καρχηδών 117 129 14.963 30.9% 63.4% 71.8% 69.1% 94.1% 79.6% 63.3% reviewed/approved 10895 gpt-5.4
2484 Καρία 270 248 -13.401 37.4% 64.5% 59.9% 68.7% 92.1% 79.3% 55.7% reviewed/approved 10880 gpt-5.4
2119 Κάλλατις 65 75 11.048 22.4% 58.8% 69.1% 61.4% 93.4% 67.6% 59.2% reviewed/approved 10820 gpt-5.4
2056 Καβειρία 112 99 -10.221 33.5% 68.8% 63.3% 68.2% 95.5% 78.5% 59.7% reviewed/approved 10203 gpt-5.4
7254 Κύτα 84 91 8.748 45.6% 68.8% 73.5% 72.0% 94.4% 76.6% 62.2% reviewed/approved 10991 gpt-5.4
2609 Κάσπειρος 135 140 8.627 52.8% 72.7% 72.1% 75.6% 91.0% 76.1% 68.1% reviewed/approved 10904 gpt-5.4
2603 Κάρυστος 175 178 8.100 40.7% 64.0% 68.5% 65.5% 94.6% 77.4% 58.4% reviewed/approved 10892 gpt-5.4
7251 Κύρτος 66 57 -7.915 34.4% 58.0% 65.0% 71.0% 93.4% 71.9% 63.6% reviewed/approved 10982 gpt-5.4
7259 Κύφος 54 61 7.643 45.9% 74.8% 88.0% 85.2% 95.9% 75.3% 74.2% reviewed/approved 11006 gpt-5.4
2477 Καρδαμύλη 43 50 7.238 41.5% 74.3% 84.1% 71.0% 96.4% 79.9% 73.1% reviewed/approved 10877 gpt-5.4
7260 Κυχρεῖος πάγος 61 67 6.901 46.8% 73.0% 79.0% 65.6% 96.0% 78.6% 63.8% reviewed/approved 11009 gpt-5.4
7255 Κυτέριον 22 29 6.464 31.0% 64.0% 61.7% 58.8% 94.7% 72.6% 69.0% reviewed/approved 10994 gpt-5.4
7257 Κυτώνιον 25 20 -5.425 57.1% 72.6% 74.7% 80.0% 97.5% 82.9% 76.8% reviewed/approved 11000 gpt-5.4
2059 Καδμεία 20 26 5.390 32.1% 67.5% 68.2% 69.6% 96.1% 81.6% 78.3% reviewed/approved 10212 gpt-5.4
2468 Καπετώλιον 138 129 -5.263 29.3% 56.3% 58.1% 63.7% 89.4% 70.3% 52.0% reviewed/approved 10871 gpt-5.4
2621 Κάστνιον 25 30 4.575 50.6% 76.1% 75.2% 76.4% 97.0% 81.2% 75.7% reviewed/approved 10910 gpt-5.4
2083 Καλαμένθη 24 20 -4.462 43.7% 66.6% 71.5% 72.7% 96.0% 76.8% 73.7% reviewed/approved 10790 gpt-5.4
7244 Κυρήνη 54 49 -4.357 49.3% 75.2% 84.9% 81.6% 97.0% 79.9% 78.3% reviewed/approved 10961 gpt-5.4
7253 Κυρτώνιος 18 23 4.317 42.5% 67.8% 84.2% 73.2% 96.9% 77.6% 80.8% reviewed/approved 10988 gpt-5.4
2329 Κάλπη 57 52 -4.247 61.3% 76.8% 81.6% 85.5% 97.0% 76.8% 76.5% reviewed/approved 10835 gpt-5.4
2328 Καλλίπολις 43 47 4.238 39.3% 67.3% 72.1% 64.4% 94.6% 74.1% 75.8% reviewed/approved 10832 gpt-5.4
2597 Καρπασία 112 105 -4.221 54.2% 76.2% 72.8% 79.3% 95.9% 75.9% 61.3% reviewed/approved 10883 gpt-5.4
2470 Καππαδοκία 68 71 4.159 34.2% 66.2% 86.1% 82.0% 96.7% 77.3% 71.8% reviewed/approved 10874 gpt-5.4
2115 Καλὴ ἀκτή 64 67 4.011 30.5% 59.7% 64.3% 64.1% 94.5% 71.2% 54.6% reviewed/approved 10808 gpt-5.4
2079 Καισάρεια 39 35 -3.910 57.5% 76.1% 87.9% 91.9% 97.6% 81.1% 82.3% reviewed/approved 10778 gpt-5.4
7250 Κυρταία 40 36 -3.873 46.8% 72.0% 75.8% 71.1% 97.0% 75.3% 76.9% reviewed/approved 10979 gpt-5.4
3496 Κοτιάειον 71 66 -3.731 32.1% 54.0% 56.1% 61.3% 93.2% 75.1% 69.2% reviewed/approved 10946 gpt-5.4
2605 Κάσιον 56 59 3.717 55.5% 75.1% 83.1% 81.7% 96.8% 78.0% 72.7% reviewed/approved 10898 gpt-5.4
2465 Κάνωπος 83 85 3.711 41.7% 67.6% 78.4% 72.6% 95.8% 77.3% 75.4% reviewed/approved 10868 gpt-5.4
2342 Κάμιρος 54 57 3.643 55.9% 80.4% 86.2% 77.5% 96.7% 81.3% 74.2% reviewed/approved 10856 gpt-5.4

Residual Predictors

Positive signed-residual terms are associated with AI translations longer than expected from the human length; negative signed-residual terms are associated with shorter AI translations. Absolute-residual models identify terms associated with larger AI-human length divergence.

Greek source terms predicting signed length residuals

Features: 176; ridge alpha: 8.377; cross-validated R^2: -0.196.

Positive termCoefficientDocsMean presentMean absent
οι 0.8424 11 3.226 -0.399
δε και 0.8036 8 2.860 -0.249
τε 0.7708 5 3.965 -0.209
εκαλειτο 0.6941 7 2.282 -0.172
εστι δε και 0.6602 3 7.969 -0.246
ος 0.6281 6 4.296 -0.274
τη 0.6132 15 1.844 -0.325
δυο 0.5707 4 4.723 -0.197
απο 0.5657 27 1.115 -0.412
εστι δε 0.5609 5 3.539 -0.186
ου και 0.5150 4 6.285 -0.262
εν τη 0.5140 7 2.999 -0.226
παιδος 0.4796 4 3.571 -0.149
εδει 0.4683 3 5.637 -0.174
πολιτης 0.4547 18 2.167 -0.476
ου 0.4391 14 1.291 -0.210
λεγεται 0.4245 9 2.149 -0.213
πολιχνιον 0.4171 3 4.592 -0.142
εκαλειτο δε 0.4107 3 3.221 -0.100
αφ 0.4097 6 2.302 -0.147
Negative termCoefficientDocsMean presentMean absent
το -0.6729 70 -0.201 0.469
εστι και -0.5677 20 -0.951 0.238
θεοπομπος -0.5490 5 -3.200 0.168
οικητωρ -0.5371 6 -4.021 0.257
κατα -0.4491 10 -1.147 0.127
τω -0.4410 16 -0.878 0.167
μεταξυ -0.4242 5 -1.298 0.068
εν τω -0.3862 6 -1.872 0.120
το θηλυκον -0.3415 6 -3.163 0.202
το κτητικον -0.3332 7 -3.260 0.245
κτητικον -0.3332 7 -3.260 0.245
του -0.3328 37 0.472 -0.277
δευτερω -0.3189 3 -1.448 0.045
πολις της -0.3167 3 -5.154 0.159
ως το -0.3131 3 -5.489 0.170
θρακης -0.3100 5 -1.745 0.092
δια -0.3000 16 0.463 -0.088
θηλυκον και -0.2927 3 -4.222 0.131
πολις θρακης -0.2894 4 -1.898 0.079
τω περι -0.2848 4 -4.450 0.185

Greek source terms predicting large absolute residuals

Features: 176; ridge alpha: 4.125; cross-validated R^2: -0.212.

Positive termCoefficientDocsMean presentMean absent
δε και 1.1228 8 7.187 2.550
δε 1.0704 46 3.869 2.113
αφ 1.0135 6 7.116 2.653
ος 0.9788 6 7.294 2.642
την 0.9734 13 6.324 2.413
απο 0.9677 27 4.957 2.168
και 0.9605 66 3.558 1.684
τω 0.9239 16 4.438 2.632
εκαλειτο δε 0.8831 3 12.155 2.635
εστι δε 0.8636 5 7.628 2.673
μεν 0.8162 4 9.795 2.635
απο της 0.7745 6 7.615 2.621
εν 0.7669 37 3.669 2.482
εστι δε και 0.7580 3 7.969 2.765
της 0.7445 35 4.262 2.199
εστι 0.7390 24 4.499 2.423
αφ ου 0.6930 5 7.461 2.682
γαρ 0.6789 9 5.795 2.637
τη 0.6727 15 4.929 2.567
ου 0.6564 14 5.354 2.525
Negative termCoefficientDocsMean presentMean absent
προς -0.7098 15 2.327 3.026
το εθνικον -0.6703 56 2.926 2.914
εθνικον -0.6703 56 2.926 2.914
νησος -0.6621 9 1.742 3.038
και πολις -0.5993 5 1.551 2.993
τους -0.5823 4 1.065 2.998
μια -0.5029 3 0.794 2.987
φασι -0.4667 8 2.622 2.947
πολις -0.4316 68 3.192 2.344
προς τη -0.4284 4 1.752 2.970
καππαδοκιας -0.4188 3 2.014 2.949
δευτερω -0.3702 3 1.448 2.967
ειρηται -0.3595 3 1.772 2.957
τεταρτω -0.3496 3 1.989 2.950
τον -0.3421 10 2.898 2.924
πελοποννησου -0.3394 3 1.036 2.979
πολυβιος -0.3376 5 1.628 2.989
ουδετερως -0.3321 3 1.963 2.951
λυκιας -0.3218 3 1.149 2.976
πολις και -0.3174 6 3.331 2.895

AI English terms predicting large absolute residuals

Features: 332; ridge alpha: 2.031; cross-validated R^2: -0.134.

Positive termCoefficientDocsMean presentMean absent
the 1.6273 93 3.019 1.615
to 1.2514 23 4.653 2.404
with 1.2216 16 4.862 2.551
there 1.1982 31 4.288 2.307
and 1.1907 56 3.975 1.579
who 1.1644 10 6.198 2.557
are 1.1463 23 4.440 2.467
it 1.0933 37 4.148 2.200
whom 1.0905 4 7.488 2.731
used 1.0098 11 6.114 2.526
is the 0.9611 4 7.420 2.734
of 0.9590 65 3.495 1.856
is named after 0.9404 3 9.836 2.707
among 0.9323 6 5.629 2.748
by 0.9194 11 4.993 2.665
from 0.9165 42 3.839 2.256
is from 0.9119 17 5.041 2.487
he 0.9032 5 8.601 2.622
there is 0.8704 23 4.408 2.477
after 0.8643 24 4.699 2.360
Negative termCoefficientDocsMean presentMean absent
they -0.9053 11 3.277 2.877
an -0.8592 16 2.589 2.984
island -0.8039 14 2.014 3.069
the ethnonym is -0.7746 57 2.968 2.859
ethnonym is -0.7746 57 2.968 2.859
some -0.7417 11 2.372 2.989
one of the -0.6919 3 0.794 2.987
one of -0.6919 3 0.794 2.987
the ethnonym -0.6884 58 2.991 2.825
ethnonym -0.6884 58 2.991 2.825
an island -0.6768 10 1.606 3.067
and a -0.6526 6 2.039 2.977
this is -0.6099 6 2.197 2.967
a city and -0.5911 5 1.540 2.994
a city -0.5849 65 3.258 2.296
is also called -0.5802 3 0.486 2.996
also called -0.5688 4 1.054 2.999
city in -0.5343 40 3.155 2.765
say -0.5169 11 3.095 2.899
neuter -0.5159 4 1.947 2.962

Human English terms predicting large absolute residuals

Features: 335; ridge alpha: 2.894; cross-validated R^2: -0.149.

Positive termCoefficientDocsMean presentMean absent
the 1.3416 90 3.077 1.521
who 1.2738 13 5.840 2.485
and 1.2458 52 3.822 1.945
there 1.2113 34 4.279 2.221
was 1.1521 20 4.653 2.488
from 0.9685 37 3.949 2.317
also 0.8663 43 3.781 2.272
with 0.8442 16 4.640 2.594
usage 0.8404 4 7.014 2.750
from the 0.8197 13 5.783 2.493
there is 0.8028 24 4.495 2.424
there is also 0.7958 22 4.688 2.423
be 0.7761 15 5.773 2.418
is with 0.7374 4 8.486 2.689
among 0.6928 4 7.358 2.736
to 0.6909 31 4.129 2.378
local 0.6891 6 6.011 2.724
like 0.6815 5 4.966 2.813
with the 0.6744 5 7.375 2.687
there are 0.6686 5 6.438 2.736
Negative termCoefficientDocsMean presentMean absent
they -1.0294 12 1.874 3.064
an -0.8866 22 2.057 3.165
the ethnonym is -0.7473 52 2.775 3.079
ethnonym is -0.7473 52 2.775 3.079
island -0.5881 14 2.014 3.069
an island -0.5850 9 1.316 3.080
a city -0.5621 64 3.250 2.335
ethnonym -0.5447 57 3.033 2.773
kappadokia -0.5240 4 2.550 2.936
the ethnonym -0.5210 56 3.057 2.748
one of the -0.4990 4 1.308 2.988
ethnonyms -0.4797 3 1.138 2.976
one of -0.4664 5 1.754 2.982
they are -0.4255 7 2.182 2.977
territory -0.4097 5 1.333 3.005
but -0.3929 12 2.896 2.924
a city in -0.3838 40 3.353 2.633
it is also -0.3746 3 0.880 2.984
in the -0.3726 25 2.978 2.902
is a -0.3693 31 3.443 2.686

Prompt Text

Show prompt text
You are an expert classical philologist and translator specialising in Stephanos of Byzantium's Ethnika.

Goal
- Produce a clear, scholarly English translation in the established reviewed house style.
- Stay close to the Greek while making Stephanos' compressed reference-work prose intelligible in English.
- Translate the Greek main text of the entry, not the modern apparatus, editorial locators, or later bibliographic explanation.

Priority order
- Follow the source Greek first.
- Apply the output and formatting rules below.
- Apply any entry-specific Additional translation context supplied with the entry, but only where the cited Greek evidence is relevant.
- Do not force a rule, gloss, or proper-noun convention when the local syntax or context contradicts it.

Output rules (required)
- Respond ONLY by calling the submit_translation tool with a single string field: {"translation": "..."}.
- The translation text must contain only the translation.
- Do not include analysis, commentary, alternative translations, uncertainty notes, or Markdown outside the translation itself.

A) Formatting + spelling
- Use Australasian spelling and punctuation conventions.
- Preserve paragraphing and line-breaks when they are present in the Greek source.
- For verse quotations, preserve line segmentation in English where the Greek source marks or implies individual verse lines.
- For lengthy prose quotations that function as a quoted block, allow paragraph/quotation separation so the quotation remains readable. Do not create stanza-like line breaks for ordinary short prose snippets.
- Use single quotes for quoted forms, cited snippets, translated quotations, appellations derived from or pertaining to the headword (such as ethnonyms and similar gentilics, but not divine epithets), and Greek forms rendered in Latin letters: '...'.
- Avoid double quotes.
- Use *italics* for titles of ancient works when they are actual work titles: *Cypriaka*, *Bassarika*, *Europe*, *Periegesis*, *Geography*.
- Do not italicise descriptive phrases that are not work titles, such as an epitome of another author's books.

B) Opening / structure
- Let the Greek structure control the opening. Do not force every entry into a single template.
- When the entry opens with the headword and a brief identification, normally begin with the headword transliterated into Latin letters without Greek diacritics, followed by the identification.
- Typical form where appropriate: Headword: ...
- Appositive openings like 'Karia, the country.' are acceptable when the entry is of that type.
- If the headword line gives both nominative and genitive, render it compactly as 'X, Y: ...' unless a grammatical tag is needed for clarity.
- Keep numbered or lettered sequences of homonymous places as inline numbered items like (2) ... (3) ..., matching the Greek's structure.
- When Greek ἔστι καὶ ἄλλη / ἕτερος introduces another place with the same name, use 'There is also another ...' when that is needed for sense.
- For appositional settlement descriptions of the type X (proper noun) + Y (settlement noun) + Z (genitive territory), translate the territory as location: 'X, a city/town/village in Z', not 'of Z', unless the Greek clearly expresses possession, origin, descent, or another non-locative relationship.

C) Transliteration + naming
- Do NOT use macrons or acute accents in ordinary transliteration: Karystos, not Kárystos.
- Prefer Greek-form transliteration when Stephanos is discussing Greek forms, derivation, spelling, ethnonyms, or other metalinguistic material.
- Use these general correspondences unless an established house-style exception is supplied: η = e, κ = k, φ = ph, ῥ = rh, ῤ = r, υ = y, χ = ch.
- Use Greek-style transliteration of diphthongs, especially ai for αι. Do not drift into Latinised forms such as ae or e where the Greek form matters.
- Use forms such as Karchedon, Chalkedon, Kapetolion, Kyrene, Kyrtos, Kappadokia, and Palaistine in Greek-form contexts.
- Avoid Latinised forms such as Carthage, Chalcedon, Capitolium, Cyrene, Cyrtos, Cappadocia, and Palestine in Greek-form contexts.
- Use established conventional English forms only when they are house-style exceptions or clearly ordinary geography, for example Rome, Cyprus, Egypt, Syria, India, Sicily, Thebes, Peloponnese, the Black Sea, the Red Sea, and the Persian Gulf.
- Treat conventional English-name exceptions as a conservative whitelist. Do not expand it by analogy without entry-specific guidance.
- Keep Greek-form transliteration for most ethnonyms and derived adjectives, and put the cited form in single quotes.
- Do not Latinise river, people, or place names merely because a familiar Latin or English form exists.

D) Citations (authors/works/books)
- Translate only the citation information that is present in the original Greek words of the lemma entry.
- Ignore modern/editorial locator codes and apparatus-like add-ons: omit RE/SH numbers, FGrHist numbers, fragment numbers, chapter/section locators, column references, and parenthetical references such as (12,1,2 [C 533,12]), (Op. 344), or (A. R. 2,403).
- Do not infer a book number from a modern parenthetical reference.
- If the Greek says only 'Strabo' plus a modern parenthetical locator, translate 'Strabo', not 'Strabo, book 12'.
- Use 'as per X' only where the Greek itself has ὡς + author/name or an equivalent example-marker construction that supports that rendering.
- If the Greek explicitly gives a Greek book numeral, translate it: 'Ktesias in book 3 of his *Persika*'.
- Keep citations compact, mirroring the source's incompleteness:
  - Author + work title: 'Hellanikos in his *Cypriaka*'
  - Author + book: 'Pausanias, book 9.'
  - Author + work + book: 'Dionysios in book 3 of his *Bassarika*: ...'
- Work-title guidance:
  - Γεωγραφούμενα / γεωγραφικά in citation contexts -> *Geography*, not *Geographoumena*, when this is the established house treatment for the cited work.
  - Περιήγησις / Περίοδος should normally preserve the transmitted title as *Periegesis* when that is what Stephanos gives.
  - If the Greek says an author is cited in an epitome of the eleven books, do not italicise *Epitome* as a work title; translate 'in an epitome of his eleven books'.
- If a citation is present in the source as bracketed main text and later syntax depends on it, translate it in brackets rather than omitting it.
- Do not add external citation labels like Iliad, Works and Days, or Argonautica unless the Greek itself names them.

E) Entry-specific guidance context
- Additional translation context may be supplied for the current entry.
- This context can include matched formula rules, glossary rules, proper-noun conventions, contextual vocabulary bias, and external source passages.
- Apply required or replacement guidance strongly when the evidence matches the current Greek.
- Apply formula and gloss guidance according to the local syntax; do not make it mechanical if the context requires a different construction.
- Apply proper-noun guidance for consistency where supplied, but do not overwrite the Greek's grammatical point.
- Apply contextual vocabulary bias as a preference, not as a one-to-one replacement. It should steer the wording when the stated semantic context is present.
- Use external source passages only for quoted or allusive material. Do not replace Stephanos' own wording with the external source translation.
- Do not mention which guidance rules fired in the translation output. The review interface should surface that separately.

F) Philological, grammatical, and orthographic discussion
- When the entry discusses spelling, letters, diphthongs, accents, stress, case, gender, number, dialect, or morphology, preserve the technical claim precisely.
- Preserve Greek letters or digraphs in Greek script when the letter or digraph itself is under discussion, for example ι, ει, οι, υ.
- Prefer transliteration for cited word-forms unless the point depends on Greek spelling, accentuation, a specific letter, or a visible dialectal/prosodic distinction.
- If accentuation or prosody is the topic:
  - τόνος -> stress or accentuation, according to context.
  - ὀξύς describing a word's τόνος -> oxytone, not merely 'acute'.
  - βαρύς describing a word's τόνος -> barytone, not merely 'grave'.
  - Keep the form in a representation that makes the contrast visible. Use Greek script where the Greek spelling or accent is the point.
- When Stephanos cites an inflected form, dialectal form, or poetic form as evidence for another form, do not normalise away the form being demonstrated.
- If the cited form is inflected or dialectal, translate using the English lemma derived from the appropriate Greek nominative where that is needed for consistency. For example, treat Κυταιέος as evidence for Kytaieus from Κυταιεύς, not as a different nominative such as Kytaios.
- Use brief grammatical tags only when they are necessary to make the philological point visible:
  - case: (nom.), (acc.), (gen.), (dat.), (voc.)
  - gender: (m.), (f.), (n.)
  - number: (sing.), (pl.)

G) Comparisons / derivational analogies
- For ὡς + X where X is an analogous form or etymologically related noun, use a form like "(as in 'X')" where this makes the analogy clear.
- For fuller derivational analogies, keep the relation explicit:
  - as 'Byzantios' is from the name Byzantion
  - as in 'Lykaones', from Lykaonia
- Avoid bare 'like X' when the point is morphological analogy.

H) ἀπό, ἀφ' οὗ, and naming/derivation formulae
- If ἀπό + genitive gives an eponym or naming source, prefer 'after X' or 'named after X' rather than a flat 'from X'.
- In entries with temporal naming contrast, ἐκαλεῖτο X ... κέκληται Y -> 'It used to be called X; it is now called/named Y.'
- For ἀφ' οὗ:
  - With a masculine personal antecedent in a naming context, translate 'after whom'.
  - With an explicit neuter antecedent, translate 'from which'.
  - With a dropped neuter antecedent or Stephanos' example-marker usage, translate naturally according to the syntax; use 'as per' only when the Greek construction supports an evidentiary/example sense.
- Avoid archaic 'whence' unless it is clearly the most natural scholarly English.

I) Syntax and restraint
- Translate directly; do not add background explanation, modern bibliography, or parenthetical glosses unless Stephanos gives them.
- Do not add glosses like (synoikia), (kalathos), or explanatory etymologies unless the Greek explicitly defines the term.
- Avoid gratuitous adversatives for δέ.
- Use 'however' only for a real contrast, such as a correction, opposition, exception, or contrastive second item that English needs to mark.
- Avoid 'on the other hand' unless the Greek contrast is strong and English needs it.
- Convert participles into natural English when strict literalness obscures the structure.
- Circumstantial participles may need 'because', 'as', 'when', or a relative clause.
- Preserve uncertainty when the Greek is uncertain, corrupt, or genuinely ambiguous; do not invent a resolution.
- If the source includes bracketed or braced text as part of the lemma main Greek text of the Meineke edition rather than apparatus, translate it and keep brackets if that best reflects the source.
- If the source gives indirect speech under φασί, do not automatically turn it into a formal quotation unless the source punctuation or reference-work style calls for it.

Now translate the provided entry accordingly, and submit it via submit_translation.