David Pomerenke
commited on
Commit
·
0e5691e
1
Parent(s):
50128d8
Improve language and script names and speaker data
Browse files- .gitignore +2 -0
- README.md +1 -1
- data.txt +4 -0
- index.html +8 -1
- languagebench.py +63 -25
- languages.rq +1 -2
- languages.tsv +0 -0
- results.json +200 -120
- results_summary.json +34 -0
.gitignore
CHANGED
|
@@ -1,4 +1,6 @@
|
|
| 1 |
floresp-*
|
|
|
|
|
|
|
| 2 |
.cache
|
| 3 |
.env
|
| 4 |
|
|
|
|
| 1 |
floresp-*
|
| 2 |
+
LanguageCodes.tab
|
| 3 |
+
ScriptCodes.csv
|
| 4 |
.cache
|
| 5 |
.env
|
| 6 |
|
README.md
CHANGED
|
@@ -6,4 +6,4 @@ Sources:
|
|
| 6 |
|
| 7 |
1. For AI models: [OpenRouter](https://openrouter.ai/)
|
| 8 |
2. For language benchmarks: [FLORES+](https://github.com/openlanguagedata/flores)
|
| 9 |
-
3. For language statistics: [Wikidata](https://gist.github.com/unhammer/3e8f2e0f79972bf5008a4c970081502d) (
|
|
|
|
| 6 |
|
| 7 |
1. For AI models: [OpenRouter](https://openrouter.ai/)
|
| 8 |
2. For language benchmarks: [FLORES+](https://github.com/openlanguagedata/flores)
|
| 9 |
+
3. For language statistics: [Wikidata](https://gist.github.com/unhammer/3e8f2e0f79972bf5008a4c970081502d) (And [Ethnologue](https://www.ethnologue.com/browse/names/) for additional language names)
|
data.txt
ADDED
|
@@ -0,0 +1,4 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
floresp-v2.0-rc.3: https://github.com/openlanguagedata/flores
|
| 2 |
+
languages.csv: generated from https://query.wikidata.org/ using the languages.rq query
|
| 3 |
+
LanguageCodes.tab: https://www.ethnologue.com/
|
| 4 |
+
ScriptCodes.csv: https://www.unicode.org/iso15924/iso15924-codes.html
|
index.html
CHANGED
|
@@ -33,6 +33,8 @@
|
|
| 33 |
import * as Plot from "https://cdn.jsdelivr.net/npm/@observablehq/plot@0.6/+esm";
|
| 34 |
|
| 35 |
async function init() {
|
|
|
|
|
|
|
| 36 |
const response = await fetch('results.json');
|
| 37 |
const results = await response.json();
|
| 38 |
|
|
@@ -70,6 +72,11 @@
|
|
| 70 |
|
| 71 |
const languageData = results.filter(r => r.target_language_name === language);
|
| 72 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 73 |
// Create plot using the more idiomatic Observable Plot approach
|
| 74 |
const plot = Plot.plot({
|
| 75 |
width: 400,
|
|
@@ -81,7 +88,7 @@
|
|
| 81 |
},
|
| 82 |
marks: [
|
| 83 |
Plot.barY(languageData, {
|
| 84 |
-
x: d => d.model
|
| 85 |
y: "bleu"
|
| 86 |
})
|
| 87 |
]
|
|
|
|
| 33 |
import * as Plot from "https://cdn.jsdelivr.net/npm/@observablehq/plot@0.6/+esm";
|
| 34 |
|
| 35 |
async function init() {
|
| 36 |
+
const summary = await fetch('results_summary.json');
|
| 37 |
+
|
| 38 |
const response = await fetch('results.json');
|
| 39 |
const results = await response.json();
|
| 40 |
|
|
|
|
| 72 |
|
| 73 |
const languageData = results.filter(r => r.target_language_name === language);
|
| 74 |
|
| 75 |
+
const descriptor = code => {
|
| 76 |
+
let [org, model] = code.split("/")
|
| 77 |
+
return model.split("-")[0]
|
| 78 |
+
}
|
| 79 |
+
|
| 80 |
// Create plot using the more idiomatic Observable Plot approach
|
| 81 |
const plot = Plot.plot({
|
| 82 |
width: 400,
|
|
|
|
| 88 |
},
|
| 89 |
marks: [
|
| 90 |
Plot.barY(languageData, {
|
| 91 |
+
x: d => descriptor(d.model),
|
| 92 |
y: "bleu"
|
| 93 |
})
|
| 94 |
]
|
languagebench.py
CHANGED
|
@@ -1,6 +1,7 @@
|
|
| 1 |
import asyncio
|
| 2 |
import json
|
| 3 |
import os
|
|
|
|
| 4 |
from os import getenv
|
| 5 |
|
| 6 |
import evaluate
|
|
@@ -9,6 +10,7 @@ from dotenv import load_dotenv
|
|
| 9 |
from joblib.memory import Memory
|
| 10 |
from openai import AsyncOpenAI
|
| 11 |
from tqdm.asyncio import tqdm_asyncio
|
|
|
|
| 12 |
|
| 13 |
# config
|
| 14 |
models = [
|
|
@@ -21,15 +23,17 @@ models = [
|
|
| 21 |
# models = ["gpt-4o-mini"]
|
| 22 |
original_language = "eng_Latn"
|
| 23 |
dataset = "floresp-v2.0-rc.3/dev"
|
| 24 |
-
|
| 25 |
-
target_languages = [
|
| 26 |
-
|
| 27 |
-
|
| 28 |
-
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
|
|
|
|
|
|
|
| 33 |
|
| 34 |
# setup
|
| 35 |
load_dotenv()
|
|
@@ -40,17 +44,33 @@ client = AsyncOpenAI(
|
|
| 40 |
)
|
| 41 |
cache = Memory(location=".cache", verbose=0).cache
|
| 42 |
bleu = evaluate.load("sacrebleu")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 43 |
language_stats = pd.read_csv("languages.tsv", sep="\t")
|
|
|
|
| 44 |
|
| 45 |
|
| 46 |
@cache
|
| 47 |
-
async def translate(model, target_language, sentence):
|
| 48 |
-
reply = await
|
| 49 |
model=model,
|
| 50 |
messages=[
|
| 51 |
{
|
| 52 |
"role": "user",
|
| 53 |
-
"content": f"Translate the following text
|
| 54 |
}
|
| 55 |
],
|
| 56 |
temperature=0,
|
|
@@ -61,9 +81,15 @@ async def translate(model, target_language, sentence):
|
|
| 61 |
def get_language_stats(language_code):
|
| 62 |
lang, script = language_code.split("_")
|
| 63 |
stats = language_stats[language_stats["iso639_3"] == lang]
|
| 64 |
-
if stats.empty:
|
| 65 |
-
|
| 66 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 67 |
|
| 68 |
|
| 69 |
async def main():
|
|
@@ -71,31 +97,43 @@ async def main():
|
|
| 71 |
results = []
|
| 72 |
original_sentences = open(f"{dataset}/dev.{original_language}").readlines()
|
| 73 |
for target_language in target_languages:
|
|
|
|
|
|
|
| 74 |
target_sentences = open(f"{dataset}/dev.{target_language}").readlines()
|
| 75 |
for model in models:
|
| 76 |
-
|
| 77 |
-
|
| 78 |
-
|
| 79 |
-
|
| 80 |
-
|
| 81 |
-
|
| 82 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 83 |
metrics = bleu.compute(
|
| 84 |
-
predictions=predictions,
|
|
|
|
|
|
|
| 85 |
)
|
| 86 |
-
|
| 87 |
results.append(
|
| 88 |
{
|
| 89 |
"model": model,
|
| 90 |
"original_language": original_language,
|
| 91 |
"target_language": target_language,
|
| 92 |
-
"target_language_name": stats
|
| 93 |
"speakers": stats.get("maxSpeakers"),
|
| 94 |
"bleu": metrics["score"],
|
| 95 |
}
|
| 96 |
)
|
| 97 |
with open("results.json", "w") as f:
|
| 98 |
json.dump(results, f, indent=2, ensure_ascii=False)
|
|
|
|
|
|
|
|
|
|
|
|
|
| 99 |
|
| 100 |
|
| 101 |
if __name__ == "__main__":
|
|
|
|
| 1 |
import asyncio
|
| 2 |
import json
|
| 3 |
import os
|
| 4 |
+
import random
|
| 5 |
from os import getenv
|
| 6 |
|
| 7 |
import evaluate
|
|
|
|
| 10 |
from joblib.memory import Memory
|
| 11 |
from openai import AsyncOpenAI
|
| 12 |
from tqdm.asyncio import tqdm_asyncio
|
| 13 |
+
from tqdm.auto import tqdm
|
| 14 |
|
| 15 |
# config
|
| 16 |
models = [
|
|
|
|
| 23 |
# models = ["gpt-4o-mini"]
|
| 24 |
original_language = "eng_Latn"
|
| 25 |
dataset = "floresp-v2.0-rc.3/dev"
|
| 26 |
+
random.seed(42)
|
| 27 |
+
target_languages = [f.split(".")[1] for f in os.listdir(dataset)]
|
| 28 |
+
target_languages = random.choices(target_languages, k=10)
|
| 29 |
+
# target_languages = [
|
| 30 |
+
# "eng_Latn",
|
| 31 |
+
# "deu_Latn",
|
| 32 |
+
# "fra_Latn",
|
| 33 |
+
# "spa_Latn",
|
| 34 |
+
# "cmn_Hans",
|
| 35 |
+
# "cmn_Hant",
|
| 36 |
+
# ]
|
| 37 |
|
| 38 |
# setup
|
| 39 |
load_dotenv()
|
|
|
|
| 44 |
)
|
| 45 |
cache = Memory(location=".cache", verbose=0).cache
|
| 46 |
bleu = evaluate.load("sacrebleu")
|
| 47 |
+
|
| 48 |
+
|
| 49 |
+
@cache
|
| 50 |
+
async def complete(**kwargs):
|
| 51 |
+
return await client.chat.completions.create(**kwargs)
|
| 52 |
+
|
| 53 |
+
|
| 54 |
+
def reorder(language_name):
|
| 55 |
+
if "," in language_name and "(" not in language_name:
|
| 56 |
+
return language_name.split(",")[1] + " " + language_name.split(",")[0]
|
| 57 |
+
return language_name
|
| 58 |
+
|
| 59 |
+
|
| 60 |
+
language_names = pd.read_csv("LanguageCodes.tab", sep="\t")
|
| 61 |
+
language_names["Name"] = language_names["Name"].apply(reorder)
|
| 62 |
language_stats = pd.read_csv("languages.tsv", sep="\t")
|
| 63 |
+
script_names = pd.read_csv("ScriptCodes.csv")
|
| 64 |
|
| 65 |
|
| 66 |
@cache
|
| 67 |
+
async def translate(model, target_language, target_script, sentence):
|
| 68 |
+
reply = await complete(
|
| 69 |
model=model,
|
| 70 |
messages=[
|
| 71 |
{
|
| 72 |
"role": "user",
|
| 73 |
+
"content": f"Translate the following text to {target_language} (script: {target_script}):\n\n{sentence}",
|
| 74 |
}
|
| 75 |
],
|
| 76 |
temperature=0,
|
|
|
|
| 81 |
def get_language_stats(language_code):
|
| 82 |
lang, script = language_code.split("_")
|
| 83 |
stats = language_stats[language_stats["iso639_3"] == lang]
|
| 84 |
+
if not stats.empty:
|
| 85 |
+
stats = stats.iloc[0].to_dict()
|
| 86 |
+
else:
|
| 87 |
+
stats = dict()
|
| 88 |
+
stats["script"] = script_names[script_names["Code"] == script]["English Name"].iloc[
|
| 89 |
+
0
|
| 90 |
+
]
|
| 91 |
+
stats["name"] = language_names[language_names["LangID"] == lang]["Name"].iloc[0]
|
| 92 |
+
return stats
|
| 93 |
|
| 94 |
|
| 95 |
async def main():
|
|
|
|
| 97 |
results = []
|
| 98 |
original_sentences = open(f"{dataset}/dev.{original_language}").readlines()
|
| 99 |
for target_language in target_languages:
|
| 100 |
+
if target_language == original_language:
|
| 101 |
+
continue
|
| 102 |
target_sentences = open(f"{dataset}/dev.{target_language}").readlines()
|
| 103 |
for model in models:
|
| 104 |
+
stats = get_language_stats(target_language)
|
| 105 |
+
print(f"{model} -> {stats['name']}")
|
| 106 |
+
# predictions = [
|
| 107 |
+
# await translate(model, stats["name"], stats["script"], sentence)
|
| 108 |
+
# for sentence in tqdm(original_sentences[:n])
|
| 109 |
+
# ]
|
| 110 |
+
predictions = [
|
| 111 |
+
translate(model, stats["name"], stats["script"], sentence)
|
| 112 |
+
for sentence in tqdm(original_sentences[:n])
|
| 113 |
+
]
|
| 114 |
+
predictions = await tqdm_asyncio.gather(*predictions)
|
| 115 |
metrics = bleu.compute(
|
| 116 |
+
predictions=predictions,
|
| 117 |
+
references=target_sentences[:n],
|
| 118 |
+
tokenize="char",
|
| 119 |
)
|
| 120 |
+
|
| 121 |
results.append(
|
| 122 |
{
|
| 123 |
"model": model,
|
| 124 |
"original_language": original_language,
|
| 125 |
"target_language": target_language,
|
| 126 |
+
"target_language_name": stats["name"],
|
| 127 |
"speakers": stats.get("maxSpeakers"),
|
| 128 |
"bleu": metrics["score"],
|
| 129 |
}
|
| 130 |
)
|
| 131 |
with open("results.json", "w") as f:
|
| 132 |
json.dump(results, f, indent=2, ensure_ascii=False)
|
| 133 |
+
# compute mean bleu for each target language
|
| 134 |
+
pd.DataFrame(results).groupby("target_language_name").agg(
|
| 135 |
+
{"bleu": "mean"}
|
| 136 |
+
).reset_index().to_json("results_summary.json", indent=2, orient="records")
|
| 137 |
|
| 138 |
|
| 139 |
if __name__ == "__main__":
|
languages.rq
CHANGED
|
@@ -1,8 +1,7 @@
|
|
| 1 |
# https://query.wikidata.org/
|
| 2 |
SELECT DISTINCT ?item (MAX(?numberOfSpeakers) AS ?maxSpeakers) (MAX(?speakersTime) AS ?mostRecentTime) ?iso639_1 ?iso639_3 ?itemLabel ?itemLabel_en
|
| 3 |
WHERE {
|
| 4 |
-
?item wdt:
|
| 5 |
-
wdt:P220 ?iso639_3. # Language with ISO 639-3 code
|
| 6 |
|
| 7 |
?item p:P1098 ?numberOfSpeakersStatement.
|
| 8 |
?numberOfSpeakersStatement ps:P1098 ?numberOfSpeakers.
|
|
|
|
| 1 |
# https://query.wikidata.org/
|
| 2 |
SELECT DISTINCT ?item (MAX(?numberOfSpeakers) AS ?maxSpeakers) (MAX(?speakersTime) AS ?mostRecentTime) ?iso639_1 ?iso639_3 ?itemLabel ?itemLabel_en
|
| 3 |
WHERE {
|
| 4 |
+
?item wdt:P220 ?iso639_3. # Language with ISO 639-3 code
|
|
|
|
| 5 |
|
| 6 |
?item p:P1098 ?numberOfSpeakersStatement.
|
| 7 |
?numberOfSpeakersStatement ps:P1098 ?numberOfSpeakers.
|
languages.tsv
CHANGED
|
The diff for this file is too large to render.
See raw diff
|
|
|
results.json
CHANGED
|
@@ -2,241 +2,321 @@
|
|
| 2 |
{
|
| 3 |
"model": "openai/gpt-4o-mini",
|
| 4 |
"original_language": "eng_Latn",
|
| 5 |
-
"target_language": "
|
| 6 |
-
"target_language_name": "
|
| 7 |
-
"speakers":
|
| 8 |
-
"bleu":
|
| 9 |
},
|
| 10 |
{
|
| 11 |
"model": "google/gemini-flash-1.5",
|
| 12 |
"original_language": "eng_Latn",
|
| 13 |
-
"target_language": "
|
| 14 |
-
"target_language_name": "
|
| 15 |
-
"speakers":
|
| 16 |
-
"bleu":
|
| 17 |
},
|
| 18 |
{
|
| 19 |
"model": "anthropic/claude-3.5-sonnet",
|
| 20 |
"original_language": "eng_Latn",
|
| 21 |
-
"target_language": "
|
| 22 |
-
"target_language_name": "
|
| 23 |
-
"speakers":
|
| 24 |
-
"bleu":
|
| 25 |
},
|
| 26 |
{
|
| 27 |
"model": "qwen/qwen-2.5-72b-instruct",
|
| 28 |
"original_language": "eng_Latn",
|
| 29 |
-
"target_language": "
|
| 30 |
-
"target_language_name": "
|
| 31 |
-
"speakers":
|
| 32 |
-
"bleu":
|
| 33 |
},
|
| 34 |
{
|
| 35 |
"model": "meta-llama/llama-3.1-8b-instruct",
|
| 36 |
"original_language": "eng_Latn",
|
| 37 |
-
"target_language": "
|
| 38 |
-
"target_language_name": "
|
| 39 |
-
"speakers":
|
| 40 |
-
"bleu":
|
| 41 |
},
|
| 42 |
{
|
| 43 |
"model": "openai/gpt-4o-mini",
|
| 44 |
"original_language": "eng_Latn",
|
| 45 |
-
"target_language": "
|
| 46 |
-
"target_language_name": "
|
| 47 |
-
"speakers":
|
| 48 |
-
"bleu":
|
| 49 |
},
|
| 50 |
{
|
| 51 |
"model": "google/gemini-flash-1.5",
|
| 52 |
"original_language": "eng_Latn",
|
| 53 |
-
"target_language": "
|
| 54 |
-
"target_language_name": "
|
| 55 |
-
"speakers":
|
| 56 |
-
"bleu":
|
| 57 |
},
|
| 58 |
{
|
| 59 |
"model": "anthropic/claude-3.5-sonnet",
|
| 60 |
"original_language": "eng_Latn",
|
| 61 |
-
"target_language": "
|
| 62 |
-
"target_language_name": "
|
| 63 |
-
"speakers":
|
| 64 |
-
"bleu":
|
| 65 |
},
|
| 66 |
{
|
| 67 |
"model": "qwen/qwen-2.5-72b-instruct",
|
| 68 |
"original_language": "eng_Latn",
|
| 69 |
-
"target_language": "
|
| 70 |
-
"target_language_name": "
|
| 71 |
-
"speakers":
|
| 72 |
-
"bleu":
|
| 73 |
},
|
| 74 |
{
|
| 75 |
"model": "meta-llama/llama-3.1-8b-instruct",
|
| 76 |
"original_language": "eng_Latn",
|
| 77 |
-
"target_language": "
|
| 78 |
-
"target_language_name": "
|
| 79 |
-
"speakers":
|
| 80 |
-
"bleu":
|
| 81 |
},
|
| 82 |
{
|
| 83 |
"model": "openai/gpt-4o-mini",
|
| 84 |
"original_language": "eng_Latn",
|
| 85 |
-
"target_language": "
|
| 86 |
-
"target_language_name": "
|
| 87 |
-
"speakers":
|
| 88 |
-
"bleu":
|
| 89 |
},
|
| 90 |
{
|
| 91 |
"model": "google/gemini-flash-1.5",
|
| 92 |
"original_language": "eng_Latn",
|
| 93 |
-
"target_language": "
|
| 94 |
-
"target_language_name": "
|
| 95 |
-
"speakers":
|
| 96 |
-
"bleu":
|
| 97 |
},
|
| 98 |
{
|
| 99 |
"model": "anthropic/claude-3.5-sonnet",
|
| 100 |
"original_language": "eng_Latn",
|
| 101 |
-
"target_language": "
|
| 102 |
-
"target_language_name": "
|
| 103 |
-
"speakers":
|
| 104 |
-
"bleu":
|
| 105 |
},
|
| 106 |
{
|
| 107 |
"model": "qwen/qwen-2.5-72b-instruct",
|
| 108 |
"original_language": "eng_Latn",
|
| 109 |
-
"target_language": "
|
| 110 |
-
"target_language_name": "
|
| 111 |
-
"speakers":
|
| 112 |
-
"bleu":
|
| 113 |
},
|
| 114 |
{
|
| 115 |
"model": "meta-llama/llama-3.1-8b-instruct",
|
| 116 |
"original_language": "eng_Latn",
|
| 117 |
-
"target_language": "
|
| 118 |
-
"target_language_name": "
|
| 119 |
-
"speakers":
|
| 120 |
-
"bleu":
|
| 121 |
},
|
| 122 |
{
|
| 123 |
"model": "openai/gpt-4o-mini",
|
| 124 |
"original_language": "eng_Latn",
|
| 125 |
-
"target_language": "
|
| 126 |
-
"target_language_name": "
|
| 127 |
-
"speakers":
|
| 128 |
-
"bleu":
|
| 129 |
},
|
| 130 |
{
|
| 131 |
"model": "google/gemini-flash-1.5",
|
| 132 |
"original_language": "eng_Latn",
|
| 133 |
-
"target_language": "
|
| 134 |
-
"target_language_name": "
|
| 135 |
-
"speakers":
|
| 136 |
-
"bleu":
|
| 137 |
},
|
| 138 |
{
|
| 139 |
"model": "anthropic/claude-3.5-sonnet",
|
| 140 |
"original_language": "eng_Latn",
|
| 141 |
-
"target_language": "
|
| 142 |
-
"target_language_name": "
|
| 143 |
-
"speakers":
|
| 144 |
-
"bleu":
|
| 145 |
},
|
| 146 |
{
|
| 147 |
"model": "qwen/qwen-2.5-72b-instruct",
|
| 148 |
"original_language": "eng_Latn",
|
| 149 |
-
"target_language": "
|
| 150 |
-
"target_language_name": "
|
| 151 |
-
"speakers":
|
| 152 |
-
"bleu":
|
| 153 |
},
|
| 154 |
{
|
| 155 |
"model": "meta-llama/llama-3.1-8b-instruct",
|
| 156 |
"original_language": "eng_Latn",
|
| 157 |
-
"target_language": "
|
| 158 |
-
"target_language_name": "
|
| 159 |
-
"speakers":
|
| 160 |
-
"bleu":
|
| 161 |
},
|
| 162 |
{
|
| 163 |
"model": "openai/gpt-4o-mini",
|
| 164 |
"original_language": "eng_Latn",
|
| 165 |
-
"target_language": "
|
| 166 |
-
"target_language_name": "
|
| 167 |
-
"speakers":
|
| 168 |
-
"bleu":
|
| 169 |
},
|
| 170 |
{
|
| 171 |
"model": "google/gemini-flash-1.5",
|
| 172 |
"original_language": "eng_Latn",
|
| 173 |
-
"target_language": "
|
| 174 |
-
"target_language_name": "
|
| 175 |
-
"speakers":
|
| 176 |
-
"bleu":
|
| 177 |
},
|
| 178 |
{
|
| 179 |
"model": "anthropic/claude-3.5-sonnet",
|
| 180 |
"original_language": "eng_Latn",
|
| 181 |
-
"target_language": "
|
| 182 |
-
"target_language_name": "
|
| 183 |
-
"speakers":
|
| 184 |
-
"bleu":
|
| 185 |
},
|
| 186 |
{
|
| 187 |
"model": "qwen/qwen-2.5-72b-instruct",
|
| 188 |
"original_language": "eng_Latn",
|
| 189 |
-
"target_language": "
|
| 190 |
-
"target_language_name": "
|
| 191 |
-
"speakers":
|
| 192 |
-
"bleu":
|
| 193 |
},
|
| 194 |
{
|
| 195 |
"model": "meta-llama/llama-3.1-8b-instruct",
|
| 196 |
"original_language": "eng_Latn",
|
| 197 |
-
"target_language": "
|
| 198 |
-
"target_language_name": "
|
| 199 |
-
"speakers":
|
| 200 |
-
"bleu": 0.
|
| 201 |
},
|
| 202 |
{
|
| 203 |
"model": "openai/gpt-4o-mini",
|
| 204 |
"original_language": "eng_Latn",
|
| 205 |
-
"target_language": "
|
| 206 |
-
"target_language_name": "
|
| 207 |
-
"speakers":
|
| 208 |
-
"bleu":
|
| 209 |
},
|
| 210 |
{
|
| 211 |
"model": "google/gemini-flash-1.5",
|
| 212 |
"original_language": "eng_Latn",
|
| 213 |
-
"target_language": "
|
| 214 |
-
"target_language_name": "
|
| 215 |
-
"speakers":
|
| 216 |
-
"bleu":
|
| 217 |
},
|
| 218 |
{
|
| 219 |
"model": "anthropic/claude-3.5-sonnet",
|
| 220 |
"original_language": "eng_Latn",
|
| 221 |
-
"target_language": "
|
| 222 |
-
"target_language_name": "
|
| 223 |
-
"speakers":
|
| 224 |
-
"bleu":
|
| 225 |
},
|
| 226 |
{
|
| 227 |
"model": "qwen/qwen-2.5-72b-instruct",
|
| 228 |
"original_language": "eng_Latn",
|
| 229 |
-
"target_language": "
|
| 230 |
-
"target_language_name": "
|
| 231 |
-
"speakers":
|
| 232 |
-
"bleu":
|
| 233 |
},
|
| 234 |
{
|
| 235 |
"model": "meta-llama/llama-3.1-8b-instruct",
|
| 236 |
"original_language": "eng_Latn",
|
| 237 |
-
"target_language": "
|
| 238 |
-
"target_language_name": "
|
| 239 |
-
"speakers":
|
| 240 |
-
"bleu":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 241 |
}
|
| 242 |
]
|
|
|
|
| 2 |
{
|
| 3 |
"model": "openai/gpt-4o-mini",
|
| 4 |
"original_language": "eng_Latn",
|
| 5 |
+
"target_language": "isl_Latn",
|
| 6 |
+
"target_language_name": "Icelandic",
|
| 7 |
+
"speakers": "358000",
|
| 8 |
+
"bleu": 61.64765463591684
|
| 9 |
},
|
| 10 |
{
|
| 11 |
"model": "google/gemini-flash-1.5",
|
| 12 |
"original_language": "eng_Latn",
|
| 13 |
+
"target_language": "isl_Latn",
|
| 14 |
+
"target_language_name": "Icelandic",
|
| 15 |
+
"speakers": "358000",
|
| 16 |
+
"bleu": 58.800114937175536
|
| 17 |
},
|
| 18 |
{
|
| 19 |
"model": "anthropic/claude-3.5-sonnet",
|
| 20 |
"original_language": "eng_Latn",
|
| 21 |
+
"target_language": "isl_Latn",
|
| 22 |
+
"target_language_name": "Icelandic",
|
| 23 |
+
"speakers": "358000",
|
| 24 |
+
"bleu": 63.29310272337262
|
| 25 |
},
|
| 26 |
{
|
| 27 |
"model": "qwen/qwen-2.5-72b-instruct",
|
| 28 |
"original_language": "eng_Latn",
|
| 29 |
+
"target_language": "isl_Latn",
|
| 30 |
+
"target_language_name": "Icelandic",
|
| 31 |
+
"speakers": "358000",
|
| 32 |
+
"bleu": 28.81229691079592
|
| 33 |
},
|
| 34 |
{
|
| 35 |
"model": "meta-llama/llama-3.1-8b-instruct",
|
| 36 |
"original_language": "eng_Latn",
|
| 37 |
+
"target_language": "isl_Latn",
|
| 38 |
+
"target_language_name": "Icelandic",
|
| 39 |
+
"speakers": "358000",
|
| 40 |
+
"bleu": 35.26511601216665
|
| 41 |
},
|
| 42 |
{
|
| 43 |
"model": "openai/gpt-4o-mini",
|
| 44 |
"original_language": "eng_Latn",
|
| 45 |
+
"target_language": "ind_Latn",
|
| 46 |
+
"target_language_name": "Indonesian",
|
| 47 |
+
"speakers": "198996550",
|
| 48 |
+
"bleu": 77.046561794683
|
| 49 |
},
|
| 50 |
{
|
| 51 |
"model": "google/gemini-flash-1.5",
|
| 52 |
"original_language": "eng_Latn",
|
| 53 |
+
"target_language": "ind_Latn",
|
| 54 |
+
"target_language_name": "Indonesian",
|
| 55 |
+
"speakers": "198996550",
|
| 56 |
+
"bleu": 73.29149877209336
|
| 57 |
},
|
| 58 |
{
|
| 59 |
"model": "anthropic/claude-3.5-sonnet",
|
| 60 |
"original_language": "eng_Latn",
|
| 61 |
+
"target_language": "ind_Latn",
|
| 62 |
+
"target_language_name": "Indonesian",
|
| 63 |
+
"speakers": "198996550",
|
| 64 |
+
"bleu": 71.2935146236652
|
| 65 |
},
|
| 66 |
{
|
| 67 |
"model": "qwen/qwen-2.5-72b-instruct",
|
| 68 |
"original_language": "eng_Latn",
|
| 69 |
+
"target_language": "ind_Latn",
|
| 70 |
+
"target_language_name": "Indonesian",
|
| 71 |
+
"speakers": "198996550",
|
| 72 |
+
"bleu": 71.24329964015463
|
| 73 |
},
|
| 74 |
{
|
| 75 |
"model": "meta-llama/llama-3.1-8b-instruct",
|
| 76 |
"original_language": "eng_Latn",
|
| 77 |
+
"target_language": "ind_Latn",
|
| 78 |
+
"target_language_name": "Indonesian",
|
| 79 |
+
"speakers": "198996550",
|
| 80 |
+
"bleu": 68.73551514908719
|
| 81 |
},
|
| 82 |
{
|
| 83 |
"model": "openai/gpt-4o-mini",
|
| 84 |
"original_language": "eng_Latn",
|
| 85 |
+
"target_language": "pol_Latn",
|
| 86 |
+
"target_language_name": "Polish",
|
| 87 |
+
"speakers": "40200000",
|
| 88 |
+
"bleu": 65.22066652766671
|
| 89 |
},
|
| 90 |
{
|
| 91 |
"model": "google/gemini-flash-1.5",
|
| 92 |
"original_language": "eng_Latn",
|
| 93 |
+
"target_language": "pol_Latn",
|
| 94 |
+
"target_language_name": "Polish",
|
| 95 |
+
"speakers": "40200000",
|
| 96 |
+
"bleu": 64.9142026045634
|
| 97 |
},
|
| 98 |
{
|
| 99 |
"model": "anthropic/claude-3.5-sonnet",
|
| 100 |
"original_language": "eng_Latn",
|
| 101 |
+
"target_language": "pol_Latn",
|
| 102 |
+
"target_language_name": "Polish",
|
| 103 |
+
"speakers": "40200000",
|
| 104 |
+
"bleu": 62.0169794404058
|
| 105 |
},
|
| 106 |
{
|
| 107 |
"model": "qwen/qwen-2.5-72b-instruct",
|
| 108 |
"original_language": "eng_Latn",
|
| 109 |
+
"target_language": "pol_Latn",
|
| 110 |
+
"target_language_name": "Polish",
|
| 111 |
+
"speakers": "40200000",
|
| 112 |
+
"bleu": 50.22137732799528
|
| 113 |
},
|
| 114 |
{
|
| 115 |
"model": "meta-llama/llama-3.1-8b-instruct",
|
| 116 |
"original_language": "eng_Latn",
|
| 117 |
+
"target_language": "pol_Latn",
|
| 118 |
+
"target_language_name": "Polish",
|
| 119 |
+
"speakers": "40200000",
|
| 120 |
+
"bleu": 54.39716369344653
|
| 121 |
},
|
| 122 |
{
|
| 123 |
"model": "openai/gpt-4o-mini",
|
| 124 |
"original_language": "eng_Latn",
|
| 125 |
+
"target_language": "kas_Deva",
|
| 126 |
+
"target_language_name": "Kashmiri",
|
| 127 |
+
"speakers": "6900000",
|
| 128 |
+
"bleu": 22.853844625462184
|
| 129 |
},
|
| 130 |
{
|
| 131 |
"model": "google/gemini-flash-1.5",
|
| 132 |
"original_language": "eng_Latn",
|
| 133 |
+
"target_language": "kas_Deva",
|
| 134 |
+
"target_language_name": "Kashmiri",
|
| 135 |
+
"speakers": "6900000",
|
| 136 |
+
"bleu": 1.6028583744267129
|
| 137 |
},
|
| 138 |
{
|
| 139 |
"model": "anthropic/claude-3.5-sonnet",
|
| 140 |
"original_language": "eng_Latn",
|
| 141 |
+
"target_language": "kas_Deva",
|
| 142 |
+
"target_language_name": "Kashmiri",
|
| 143 |
+
"speakers": "6900000",
|
| 144 |
+
"bleu": 10.24928534626491
|
| 145 |
},
|
| 146 |
{
|
| 147 |
"model": "qwen/qwen-2.5-72b-instruct",
|
| 148 |
"original_language": "eng_Latn",
|
| 149 |
+
"target_language": "kas_Deva",
|
| 150 |
+
"target_language_name": "Kashmiri",
|
| 151 |
+
"speakers": "6900000",
|
| 152 |
+
"bleu": 18.289957619904254
|
| 153 |
},
|
| 154 |
{
|
| 155 |
"model": "meta-llama/llama-3.1-8b-instruct",
|
| 156 |
"original_language": "eng_Latn",
|
| 157 |
+
"target_language": "kas_Deva",
|
| 158 |
+
"target_language_name": "Kashmiri",
|
| 159 |
+
"speakers": "6900000",
|
| 160 |
+
"bleu": 0.8669151440746464
|
| 161 |
},
|
| 162 |
{
|
| 163 |
"model": "openai/gpt-4o-mini",
|
| 164 |
"original_language": "eng_Latn",
|
| 165 |
+
"target_language": "lin_Latn",
|
| 166 |
+
"target_language_name": "Lingala",
|
| 167 |
+
"speakers": "20000000",
|
| 168 |
+
"bleu": 50.873508850595044
|
| 169 |
},
|
| 170 |
{
|
| 171 |
"model": "google/gemini-flash-1.5",
|
| 172 |
"original_language": "eng_Latn",
|
| 173 |
+
"target_language": "lin_Latn",
|
| 174 |
+
"target_language_name": "Lingala",
|
| 175 |
+
"speakers": "20000000",
|
| 176 |
+
"bleu": 6.943431977023627
|
| 177 |
},
|
| 178 |
{
|
| 179 |
"model": "anthropic/claude-3.5-sonnet",
|
| 180 |
"original_language": "eng_Latn",
|
| 181 |
+
"target_language": "lin_Latn",
|
| 182 |
+
"target_language_name": "Lingala",
|
| 183 |
+
"speakers": "20000000",
|
| 184 |
+
"bleu": 45.048992005653
|
| 185 |
},
|
| 186 |
{
|
| 187 |
"model": "qwen/qwen-2.5-72b-instruct",
|
| 188 |
"original_language": "eng_Latn",
|
| 189 |
+
"target_language": "lin_Latn",
|
| 190 |
+
"target_language_name": "Lingala",
|
| 191 |
+
"speakers": "20000000",
|
| 192 |
+
"bleu": 2.9809864732757902
|
| 193 |
},
|
| 194 |
{
|
| 195 |
"model": "meta-llama/llama-3.1-8b-instruct",
|
| 196 |
"original_language": "eng_Latn",
|
| 197 |
+
"target_language": "lin_Latn",
|
| 198 |
+
"target_language_name": "Lingala",
|
| 199 |
+
"speakers": "20000000",
|
| 200 |
+
"bleu": 0.5682271000080301
|
| 201 |
},
|
| 202 |
{
|
| 203 |
"model": "openai/gpt-4o-mini",
|
| 204 |
"original_language": "eng_Latn",
|
| 205 |
+
"target_language": "bjn_Latn",
|
| 206 |
+
"target_language_name": "Banjar",
|
| 207 |
+
"speakers": "3500000",
|
| 208 |
+
"bleu": 50.02000892713302
|
| 209 |
},
|
| 210 |
{
|
| 211 |
"model": "google/gemini-flash-1.5",
|
| 212 |
"original_language": "eng_Latn",
|
| 213 |
+
"target_language": "bjn_Latn",
|
| 214 |
+
"target_language_name": "Banjar",
|
| 215 |
+
"speakers": "3500000",
|
| 216 |
+
"bleu": 5.960932185623333
|
| 217 |
},
|
| 218 |
{
|
| 219 |
"model": "anthropic/claude-3.5-sonnet",
|
| 220 |
"original_language": "eng_Latn",
|
| 221 |
+
"target_language": "bjn_Latn",
|
| 222 |
+
"target_language_name": "Banjar",
|
| 223 |
+
"speakers": "3500000",
|
| 224 |
+
"bleu": 46.23236901760108
|
| 225 |
},
|
| 226 |
{
|
| 227 |
"model": "qwen/qwen-2.5-72b-instruct",
|
| 228 |
"original_language": "eng_Latn",
|
| 229 |
+
"target_language": "bjn_Latn",
|
| 230 |
+
"target_language_name": "Banjar",
|
| 231 |
+
"speakers": "3500000",
|
| 232 |
+
"bleu": 21.478973248564643
|
| 233 |
},
|
| 234 |
{
|
| 235 |
"model": "meta-llama/llama-3.1-8b-instruct",
|
| 236 |
"original_language": "eng_Latn",
|
| 237 |
+
"target_language": "bjn_Latn",
|
| 238 |
+
"target_language_name": "Banjar",
|
| 239 |
+
"speakers": "3500000",
|
| 240 |
+
"bleu": 22.371002216375594
|
| 241 |
+
},
|
| 242 |
+
{
|
| 243 |
+
"model": "openai/gpt-4o-mini",
|
| 244 |
+
"original_language": "eng_Latn",
|
| 245 |
+
"target_language": "bho_Deva",
|
| 246 |
+
"target_language_name": "Bhojpuri",
|
| 247 |
+
"speakers": "52200000",
|
| 248 |
+
"bleu": 43.46871320382143
|
| 249 |
+
},
|
| 250 |
+
{
|
| 251 |
+
"model": "google/gemini-flash-1.5",
|
| 252 |
+
"original_language": "eng_Latn",
|
| 253 |
+
"target_language": "bho_Deva",
|
| 254 |
+
"target_language_name": "Bhojpuri",
|
| 255 |
+
"speakers": "52200000",
|
| 256 |
+
"bleu": 18.253876118905147
|
| 257 |
+
},
|
| 258 |
+
{
|
| 259 |
+
"model": "anthropic/claude-3.5-sonnet",
|
| 260 |
+
"original_language": "eng_Latn",
|
| 261 |
+
"target_language": "bho_Deva",
|
| 262 |
+
"target_language_name": "Bhojpuri",
|
| 263 |
+
"speakers": "52200000",
|
| 264 |
+
"bleu": 23.447356116551486
|
| 265 |
+
},
|
| 266 |
+
{
|
| 267 |
+
"model": "qwen/qwen-2.5-72b-instruct",
|
| 268 |
+
"original_language": "eng_Latn",
|
| 269 |
+
"target_language": "bho_Deva",
|
| 270 |
+
"target_language_name": "Bhojpuri",
|
| 271 |
+
"speakers": "52200000",
|
| 272 |
+
"bleu": 29.81946758376717
|
| 273 |
+
},
|
| 274 |
+
{
|
| 275 |
+
"model": "meta-llama/llama-3.1-8b-instruct",
|
| 276 |
+
"original_language": "eng_Latn",
|
| 277 |
+
"target_language": "bho_Deva",
|
| 278 |
+
"target_language_name": "Bhojpuri",
|
| 279 |
+
"speakers": "52200000",
|
| 280 |
+
"bleu": 3.2486895985868474
|
| 281 |
+
},
|
| 282 |
+
{
|
| 283 |
+
"model": "openai/gpt-4o-mini",
|
| 284 |
+
"original_language": "eng_Latn",
|
| 285 |
+
"target_language": "ces_Latn",
|
| 286 |
+
"target_language_name": "Czech",
|
| 287 |
+
"speakers": "10700000",
|
| 288 |
+
"bleu": 69.01583822081993
|
| 289 |
+
},
|
| 290 |
+
{
|
| 291 |
+
"model": "google/gemini-flash-1.5",
|
| 292 |
+
"original_language": "eng_Latn",
|
| 293 |
+
"target_language": "ces_Latn",
|
| 294 |
+
"target_language_name": "Czech",
|
| 295 |
+
"speakers": "10700000",
|
| 296 |
+
"bleu": 69.4260447999661
|
| 297 |
+
},
|
| 298 |
+
{
|
| 299 |
+
"model": "anthropic/claude-3.5-sonnet",
|
| 300 |
+
"original_language": "eng_Latn",
|
| 301 |
+
"target_language": "ces_Latn",
|
| 302 |
+
"target_language_name": "Czech",
|
| 303 |
+
"speakers": "10700000",
|
| 304 |
+
"bleu": 68.6109083634317
|
| 305 |
+
},
|
| 306 |
+
{
|
| 307 |
+
"model": "qwen/qwen-2.5-72b-instruct",
|
| 308 |
+
"original_language": "eng_Latn",
|
| 309 |
+
"target_language": "ces_Latn",
|
| 310 |
+
"target_language_name": "Czech",
|
| 311 |
+
"speakers": "10700000",
|
| 312 |
+
"bleu": 59.72501366200287
|
| 313 |
+
},
|
| 314 |
+
{
|
| 315 |
+
"model": "meta-llama/llama-3.1-8b-instruct",
|
| 316 |
+
"original_language": "eng_Latn",
|
| 317 |
+
"target_language": "ces_Latn",
|
| 318 |
+
"target_language_name": "Czech",
|
| 319 |
+
"speakers": "10700000",
|
| 320 |
+
"bleu": 60.25088578142904
|
| 321 |
}
|
| 322 |
]
|
results_summary.json
ADDED
|
@@ -0,0 +1,34 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"target_language_name":"Banjar",
|
| 4 |
+
"bleu":29.2126571191
|
| 5 |
+
},
|
| 6 |
+
{
|
| 7 |
+
"target_language_name":"Bhojpuri",
|
| 8 |
+
"bleu":23.6476205243
|
| 9 |
+
},
|
| 10 |
+
{
|
| 11 |
+
"target_language_name":"Czech",
|
| 12 |
+
"bleu":65.4057381655
|
| 13 |
+
},
|
| 14 |
+
{
|
| 15 |
+
"target_language_name":"Icelandic",
|
| 16 |
+
"bleu":49.5636570439
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"target_language_name":"Indonesian",
|
| 20 |
+
"bleu":72.3220779959
|
| 21 |
+
},
|
| 22 |
+
{
|
| 23 |
+
"target_language_name":"Kashmiri",
|
| 24 |
+
"bleu":10.772572222
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"target_language_name":"Lingala",
|
| 28 |
+
"bleu":21.2830292813
|
| 29 |
+
},
|
| 30 |
+
{
|
| 31 |
+
"target_language_name":"Polish",
|
| 32 |
+
"bleu":59.3540779188
|
| 33 |
+
}
|
| 34 |
+
]
|