gpt-5.2 v1 Evaluation

Generated: 2026-09-20 11:21:25 UTC

Pairs100
Distinct lemmas100
Slope1.010
R^20.983
BLEU-423.4%
chrF++53.9%
METEOR58.8%
ROUGE-L58.5%
BERTScore91.9%
COMET71.5%
BLEURT63.6%

Profile version id: 2331

First translation using this version: 2026-02-14

Total usable AI runs for this version: 113 runs across 109 lemmas

Created: 2026-07-15 12:28:27.643960+10:00

Active: True

Notes: Model timeline evaluation: GPT-5.2 using prompt text from gpt-5.5 v1; approved-human-only for the 100-row Kappa corpus.

Prompt length scatter plot

Length Regression

FormulaAI words = intercept + slope * human words
Slope1.010
Intercept5.115
R^20.983
P-value9.43e-89
Pearson r0.992
Mean source words30.400
Length fallback rows0
Mean human words44.150
Mean AI words49.720
Mean absolute length residual3.984
Mean absolute percent error19.9%

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: negative; r = -0.088, R^2 = 0.008, p = 0.3826.
chrF++ vs passage length
chrF++: negative; r = -0.107, R^2 = 0.011, p = 0.2891.
METEOR vs passage length
METEOR: negative; r = -0.177, R^2 = 0.031, p = 0.0782.
ROUGE-L vs passage length
ROUGE-L: negative; r = -0.098, R^2 = 0.010, p = 0.3315.
BERTScore vs passage length
BERTScore: significant negative; r = -0.287, R^2 = 0.082, p = 0.0038.
COMET vs passage length
COMET: negative; r = -0.007, R^2 = 4.51e-05, p = 0.9471.
BLEURT vs passage length
BLEURT: significant negative; r = -0.640, R^2 = 0.410, p = 7.43e-13.
Trigram precision vs passage length
Trigram precision: positive; r = 0.024, R^2 = 5.96e-04, p = 0.8094.
Trigram recall vs passage length
Trigram recall: negative; r = -0.035, R^2 = 0.001, p = 0.7286.
Trigram F1 vs passage length
Trigram F1: negative; r = -5.81e-04, R^2 = 3.38e-07, p = 0.9954.
Trigram Jaccard vs passage length
Trigram Jaccard: negative; r = -0.019, R^2 = 3.63e-04, p = 0.8508.
Metric Rows Pattern Pearson r R^2 P-value Slope Status
BLEU-4 100 negative -0.088 0.008 0.3826 -0.00040 ok
chrF++ 100 negative -0.107 0.011 0.2891 -0.00035 ok
METEOR 100 negative -0.177 0.031 0.0782 -0.00085 ok
ROUGE-L 100 negative -0.098 0.010 0.3315 -0.00041 ok
BERTScore 100 significant negative -0.287 0.082 0.0038 -0.00020 ok
COMET 100 negative -0.007 4.51e-05 0.9471 -0.00001 ok
BLEURT 100 significant negative -0.640 0.410 7.43e-13 -0.00197 ok
Trigram precision 100 positive 0.024 5.96e-04 0.8094 0.00011 ok
Trigram recall 100 negative -0.035 0.001 0.7286 -0.00017 ok
Trigram F1 100 negative -5.81e-04 3.38e-07 0.9954 -2.69e-06 ok
Trigram Jaccard 100 negative -0.019 3.63e-04 0.8508 -0.00006 ok

Translation Similarity Metrics

Mean BLEU-423.4%
Mean chrF++53.9%
Mean METEOR58.8%
Mean ROUGE-L58.5%
Mean BERTScore91.9%
Mean COMET71.5%
Mean BLEURT63.6%
Legacy smoothed corpus BLEU24.5%
Unigram F161.8%
Bigram F132.8%
Trigram precision18.2%
Trigram recall20.6%
Trigram F119.3%
Trigram Jaccard10.7%
4-gram F111.6%

Largest Length Residuals

Lemma ID Headword Human words AI words Residual BLEU-4 chrF++ METEOR ROUGE-L BERTScore COMET BLEURT Human Run Model
2468 Καπετώλιον 138 159 14.462 13.9% 44.2% 41.0% 47.8% 86.0% 65.6% 43.6% reviewed/approved 11473 gpt-5.2-2025-12-11
2055 Καβασσός 109 102 -13.239 16.2% 46.5% 39.8% 44.5% 90.9% 70.6% 42.2% reviewed/approved 11339 gpt-5.2-2025-12-11
2626 Κατάνη 95 89 -12.094 26.8% 51.4% 51.4% 53.3% 91.5% 70.8% 52.3% reviewed/approved 11530 gpt-5.2-2025-12-11
3288 Κορώνεια 110 128 11.751 15.2% 51.9% 50.6% 48.5% 90.6% 73.3% 50.3% reviewed/approved 11545 gpt-5.2-2025-12-11
2081 Καλάθη 49 43 -11.620 33.7% 61.9% 67.3% 80.4% 93.7% 67.9% 67.0% reviewed/approved 11386 gpt-5.2-2025-12-11
2057 Καβελλιών 34 51 11.535 10.9% 47.5% 51.8% 44.7% 89.1% 66.9% 52.7% reviewed/approved 11345 gpt-5.2-2025-12-11
2056 Καβειρία 112 107 -11.270 21.4% 55.7% 63.2% 67.6% 92.5% 73.0% 58.5% reviewed/approved 11342 gpt-5.2-2025-12-11
7245 Κύρης 13 28 9.751 4.8% 32.2% 36.7% 39.0% 87.9% 56.8% 60.7% reviewed/approved 11565 gpt-5.2-2025-12-11
7243 Κύρη 22 37 9.658 16.5% 49.2% 45.1% 47.5% 90.0% 72.6% 68.6% reviewed/approved 11560 gpt-5.2-2025-12-11
2079 Καισάρεια 39 35 -9.517 30.2% 56.3% 67.6% 70.3% 91.9% 70.7% 60.0% reviewed/approved 11380 gpt-5.2-2025-12-11
2328 Καλλίπολις 43 58 9.442 8.2% 44.4% 40.7% 47.5% 89.3% 72.0% 57.1% reviewed/approved 11434 gpt-5.2-2025-12-11
2605 Κάσιον 56 71 9.308 17.9% 53.5% 62.5% 50.4% 90.4% 70.0% 57.8% reviewed/approved 11500 gpt-5.2-2025-12-11
7267 Κωνώπη 64 79 9.225 26.4% 51.8% 52.6% 55.9% 90.7% 63.6% 62.3% reviewed/approved 11630 gpt-5.2-2025-12-11
7249 Κύρρος 45 59 8.421 4.9% 41.0% 42.5% 46.2% 89.4% 64.7% 49.3% reviewed/approved 11577 gpt-5.2-2025-12-11
7255 Κυτέριον 22 35 7.658 21.6% 57.4% 65.8% 56.1% 91.5% 68.5% 65.9% reviewed/approved 11594 gpt-5.2-2025-12-11
2597 Καρπασία 112 111 -7.270 13.5% 50.6% 47.4% 49.3% 92.6% 73.3% 52.7% reviewed/approved 11485 gpt-5.2-2025-12-11
2115 Καλὴ ἀκτή 64 77 7.225 16.0% 42.6% 47.9% 51.1% 89.1% 63.6% 45.3% reviewed/approved 11410 gpt-5.2-2025-12-11
2119 Κάλλατις 65 78 7.215 16.0% 49.0% 50.0% 47.6% 90.0% 62.5% 60.7% reviewed/approved 11422 gpt-5.2-2025-12-11
7266 Κώμη 79 92 7.071 5.4% 36.4% 31.7% 31.6% 88.8% 65.4% 43.5% reviewed/approved 11627 gpt-5.2-2025-12-11
2802 Κάδρεμα 24 23 -6.362 24.5% 53.6% 59.6% 66.7% 93.7% 79.6% 75.4% reviewed/approved 11360 gpt-5.2-2025-12-11
2330 Καλύβη 24 23 -6.362 19.1% 46.7% 66.7% 68.1% 92.6% 73.6% 67.1% reviewed/approved 11440 gpt-5.2-2025-12-11
2078 Καινύς 22 21 -6.342 60.1% 71.9% 76.6% 74.4% 95.4% 82.7% 74.7% reviewed/approved 11375 gpt-5.2-2025-12-11
2061 Καδούσιοι 14 13 -6.259 54.2% 65.4% 77.2% 74.1% 95.5% 74.9% 59.2% reviewed/approved 11357 gpt-5.2-2025-12-11
7253 Κυρτώνιος 18 29 5.700 7.8% 45.8% 49.6% 46.8% 91.4% 72.0% 73.2% reviewed/approved 11588 gpt-5.2-2025-12-11
2335 Κάλυτις 26 37 5.617 3.9% 42.4% 43.2% 34.9% 91.1% 70.1% 62.4% reviewed/approved 11455 gpt-5.2-2025-12-11
2484 Καρία 270 273 -4.899 18.4% 47.4% 42.6% 49.0% 88.4% 74.4% 50.2% reviewed/approved 11482 gpt-5.2-2025-12-11
2345 Κάναθα 17 27 4.710 22.8% 57.0% 70.5% 59.1% 90.9% 74.1% 60.2% reviewed/approved 11461 gpt-5.2-2025-12-11
2621 Κάστνιον 25 35 4.627 18.1% 49.0% 62.0% 53.3% 92.3% 59.8% 67.7% reviewed/approved 11512 gpt-5.2-2025-12-11
2326 Καλλίαρος 47 57 4.401 28.4% 54.8% 59.9% 59.6% 93.5% 70.8% 63.9% reviewed/approved 11428 gpt-5.2-2025-12-11
3258 Κεκρυφάλεια 26 27 -4.383 52.0% 74.7% 79.9% 79.2% 96.1% 72.3% 71.2% reviewed/approved 11542 gpt-5.2-2025-12-11

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: 4.125; cross-validated R^2: -0.242.

Positive termCoefficientDocsMean presentMean absent
δια του 1.9363 10 4.004 -0.445
εν 1.7707 37 1.708 -1.003
ως 1.6397 52 0.672 -0.728
εν τη 1.4973 7 4.675 -0.352
του 1.4829 37 0.897 -0.527
δια 1.4712 16 2.324 -0.443
δευτερα 1.3494 3 8.937 -0.276
πολυβιος 1.2791 5 4.537 -0.239
την 1.2507 13 1.609 -0.240
γαρ 1.2228 9 1.386 -0.137
χωρας 1.1070 4 3.870 -0.161
παρα 1.1027 7 2.697 -0.203
κωμη 1.0882 3 5.579 -0.173
της χωρας 0.9983 3 4.610 -0.143
εν πρωτη 0.9599 3 3.679 -0.114
πρωτη 0.9599 3 3.679 -0.114
εξ 0.9181 3 7.370 -0.228
δευτερω 0.8942 3 4.060 -0.126
ος 0.8475 6 4.459 -0.285
πολιχνιον 0.8295 3 4.916 -0.152
Negative termCoefficientDocsMean presentMean absent
καππαδοκιας -1.6475 3 -8.039 0.249
εθνος -1.4421 8 -3.807 0.331
και του -1.4420 4 -7.610 0.317
λυκιας -1.2513 3 -5.967 0.185
προς -1.2383 15 -1.255 0.221
φησιν -1.2015 10 -2.136 0.237
εστι και -1.0796 20 -1.461 0.365
ουτως -0.9812 5 -5.000 0.263
ως το -0.9304 3 -8.101 0.251
μετα -0.8708 3 -6.833 0.211
της -0.8677 35 -0.123 0.066
πλησιον -0.8533 8 -2.273 0.198
υπο -0.8527 4 -1.867 0.078
αυτην -0.8342 9 -2.100 0.208
των -0.7959 17 -1.512 0.310
εισι -0.7906 5 -3.870 0.204
δεκατω -0.7800 3 -4.029 0.125
θρακης -0.7795 5 -2.251 0.118
εκαλειτο -0.7541 7 -3.147 0.237
θηλυκον και -0.7311 3 -5.376 0.166

Greek source terms predicting large absolute residuals

Features: 176; ridge alpha: 11.938; cross-validated R^2: -0.047.

Positive termCoefficientDocsMean presentMean absent
δε 0.6605 46 4.845 3.251
γαρ 0.5312 9 6.400 3.746
της 0.4548 35 5.472 3.183
την 0.4392 13 6.036 3.678
παρα 0.4099 7 7.316 3.734
τον 0.3992 10 6.325 3.724
κατα 0.3967 10 5.423 3.825
εκαταιος ευρωπη 0.3880 5 9.151 3.713
ευρωπη 0.3880 5 9.151 3.713
εν 0.3799 37 4.734 3.544
τριτη 0.3402 3 10.237 3.791
δυναται 0.3274 3 10.883 3.771
ποντου 0.2894 3 8.365 3.849
του ποντου 0.2894 3 8.365 3.849
δια του 0.2886 10 5.378 3.830
καππαδοκιας 0.2740 3 8.039 3.859
τα 0.2691 12 6.578 3.631
δευτερα 0.2685 3 8.937 3.831
εστι 0.2577 24 5.251 3.585
βοιωτιας 0.2510 3 8.876 3.833
Negative termCoefficientDocsMean presentMean absent
εκ -0.3339 6 1.758 4.127
ηροδοτος -0.3286 9 2.838 4.098
θεοπομπος -0.2822 5 2.986 4.037
πολις και -0.2743 6 1.849 4.121
παιδος -0.2572 4 1.410 4.092
εκαταιος ασια -0.2349 5 2.219 4.077
ασια -0.2349 5 2.219 4.077
απο -0.1950 27 4.054 3.959
τοπος -0.1941 4 1.706 4.079
τινες -0.1896 6 2.412 4.085
θαλασση -0.1888 3 1.371 4.065
οι -0.1837 11 3.601 4.032
το εθνικον -0.1658 56 3.839 4.169
εθνικον -0.1658 56 3.839 4.169
οικητωρ -0.1650 6 3.648 4.006
και το -0.1641 9 3.601 4.022
διονυσιος -0.1582 4 3.168 4.019
λιβυης -0.1526 4 2.404 4.050
πλησιον -0.1518 8 2.884 4.080
μια -0.1486 3 2.035 4.045

AI English terms predicting large absolute residuals

Features: 404; ridge alpha: 17.013; cross-validated R^2: -0.051.

Positive termCoefficientDocsMean presentMean absent
i 0.3448 12 6.302 3.668
with 0.3188 17 5.486 3.677
for 0.2846 11 5.216 3.832
the 0.2420 97 4.051 1.845
to the 0.2045 6 6.770 3.807
it 0.1963 51 4.590 3.354
on the 0.1960 9 6.230 3.762
or 0.1872 13 6.338 3.633
on 0.1817 16 4.846 3.820
formed 0.1793 4 8.966 3.777
but 0.1695 19 5.586 3.609
is a 0.1664 9 6.337 3.752
e 0.1631 10 4.915 3.881
cappadocia 0.1608 4 6.325 3.887
sicily 0.1579 3 9.293 3.820
a third 0.1569 3 10.237 3.791
according 0.1563 5 6.493 3.852
according to 0.1563 5 6.493 3.852
i e 0.1544 8 5.725 3.833
it is 0.1537 24 4.793 3.729
Negative termCoefficientDocsMean presentMean absent
the ethnic gentilic -0.2037 21 3.226 4.186
gentilic -0.2037 21 3.226 4.186
ethnic gentilic -0.2037 21 3.226 4.186
book of the -0.1921 8 2.570 4.107
of the -0.1812 47 3.614 4.313
gentilic is -0.1609 17 3.156 4.154
ethnic gentilic is -0.1609 17 3.156 4.154
also called -0.1481 14 2.580 4.213
and -0.1461 55 4.012 3.950
the ethnic -0.1453 54 3.755 4.254
an -0.1383 19 3.180 4.173
the ethnic name -0.1382 5 1.332 4.124
ethnic name -0.1382 5 1.332 4.124
ethnic -0.1369 58 3.837 4.188
herodotus -0.1268 9 2.838 4.098
named from -0.1167 5 1.458 4.117
named -0.1120 16 3.163 4.141
named after -0.1075 5 2.008 4.088
its -0.1064 9 3.308 4.051
in the feminine -0.1042 4 1.851 4.073

Human English terms predicting large absolute residuals

Features: 335; ridge alpha: 10000.000; cross-validated R^2: -0.053.

Positive termCoefficientDocsMean presentMean absent
form 0.0007 11 7.263 3.579
the 0.0006 90 4.178 2.244
with 0.0006 16 5.208 3.751
on 0.0005 16 5.133 3.766
the form 0.0004 9 7.556 3.631
is 0.0004 87 4.042 3.598
on the 0.0004 10 6.481 3.707
of 0.0004 66 4.301 3.371
has 0.0004 6 9.460 3.635
his europe 0.0004 5 9.151 3.713
in his europe 0.0004 5 9.151 3.713
a 0.0004 91 4.140 2.416
from 0.0004 37 4.597 3.625
as in 0.0004 29 4.431 3.802
in 0.0004 87 4.147 2.895
europe 0.0003 6 8.129 3.720
there 0.0003 34 5.181 3.368
one 0.0003 12 6.961 3.579
sicily 0.0003 3 9.293 3.820
is also a 0.0003 14 5.480 3.741
Negative termCoefficientDocsMean presentMean absent
herodotos -0.0003 9 2.838 4.098
the ethnonym is -0.0002 52 3.522 4.486
ethnonym is -0.0002 52 3.522 4.486
inhabitant is a -0.0002 6 2.691 4.067
in his asia -0.0002 5 2.219 4.077
his asia -0.0002 5 2.219 4.077
libya -0.0002 5 2.059 4.086
in libya -0.0002 5 2.059 4.086
book the -0.0002 10 2.955 4.099
theopompos in book -0.0002 4 1.915 4.071
theopompos in -0.0002 4 1.915 4.071
book the ethnonym -0.0002 7 2.751 4.077
an -0.0002 22 3.604 4.092
lydia -0.0002 3 2.021 4.045
some -0.0002 10 2.991 4.095
is the -0.0002 3 0.905 4.080
inhabitant is -0.0002 7 3.317 4.035
inhabitant -0.0002 7 3.317 4.035
an inhabitant is -0.0002 7 3.317 4.035
an inhabitant -0.0002 7 3.317 4.035

Prompt Text

Show prompt text
You are an expert classical philologist and translator specializing in Byzantine Greek geographical texts.
You will receive Greek text from a lemma entry in Stephanos of Byzantium's Ethnika.
Translate the Greek text into clear, scholarly English.
Preserve technical terminology and place names appropriately.