SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2308.07286 · 2023

The Devil is in the Errors: Leveraging Large Language Models for Fine-grained Machine Translation Evaluation

arXiv · PDF · Open in the Atlas

Code that ran

9 of 13 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 8 have no contract verdict; 1 failure of our instrument; 4 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2308.07286.json · badge: /api/ran/2308.07286.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
xuuhuang/lost_in_the_src community (classifier label)
classifier basis not recorded on the link
9 ran · 4 no recorded run
FunctionStatusWhere it lives
check_icl_set Ran
ran on a synthesized input
xuuhuang/lost_in_the_src/automqm/rejection_sample.py
code served (permissive licence) · get_code("ceffa1ba9496bac6")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned bool, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import pandas as pd

# Create a valid DataFrame that passes all checks in check_icl_set
examples = pd.DataFrame({
    'span': [
        ['error1', 'error2', 'error3'],  # 3 errors
        ['error4']  # 1 error
    ],
    'severity': [
        ['major', 'minor', 'major'],  # 2 major, 1 minor
        ['minor']  # 0 major, 1 minor
    ],
    'category': [
        ['grammar/subject', 'spelling/word', 'punctuation/comma'],
        ['grammar/verb']
    ],
    'source': [
        'The cat are sleeping on the mat today',  # 37 chars
        'He go to school'  # 15 chars - will be min
    ],
    'reference': [
        'The cat is sleeping on the mat today',  # 37 chars
        'He goes to school'  # 17 chars
    ],
    'candidate': [
        'The cat are sleeping on the mat today',  # 37 chars
        'He go to school'  # 15 chars - will be min, needs >= 20
    ]
})

# Adjust candidate to meet min_clen requirement
examples.loc[1, 'candidate'] = 'He go to the school today'  # 25 chars

ARGS = {
    'examples': examples,
    'min_errors': 3,
    'majmin_threshold': 2,
    'cat_diversity': 2,
    'min_clen': 20,
    'max_clen': 400
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
pandas — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type bool, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
check_icl_set_uniform Ran
ran on a synthesized input
xuuhuang/lost_in_the_src/automqm/rejection_sample.py
code served (permissive licence) · get_code("ca6b7c2e99c13a8b")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned bool, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import pandas as pd

# Create a DataFrame with the required structure
examples = pd.DataFrame({
    'severity': [
        ['major', 'minor'],
        ['major', 'minor'],
        ['minor', 'minor']
    ],
    'category': [
        ['grammar/subject-verb', 'spelling/typo'],
        ['punctuation/comma', 'grammar/article'],
        ['style/word-choice', 'grammar/tense']
    ],
    'source': [
        'The quick brown fox jumps over the lazy dog',
        'She go to the store yesterday',
        'I is happy today'
    ],
    'reference': [
        'The quick brown fox jumps over the lazy dog',
        'She goes to the store yesterday',
        'I am happy today'
    ],
    'candidate': [
        'The quick brown fox jump over the lazy dog',
        'She go to store yesterday',
        'I is happy to day'
    ]
})

ARGS = {
    'examples': examples,
    'min_errors': 3,
    'majmin_threshold': 2,
    'cat_diversity': 5,
    'min_clen': 20,
    'max_clen': 400
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
pandas — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type bool, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
format_examples Ran
ran on a synthesized input
xuuhuang/lost_in_the_src/automqm/automqm.py
code served (permissive licence) · get_code("d521942f2e0519a5")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import pandas as pd
import re

# Create a sample DataFrame with the required columns
examples_df = pd.DataFrame({
    'source': ['Hello world', 'Good morning'],
    'candidate': ['Hola mundo', 'Buenos días'],
    'reference': ['Hola mundo', 'Buenos días'],
    'translation_m': ['Hola mundo alt', 'Buenos días alt'],
    'span': [
        ['<v>world</v>', '<v>morning</v>'],
        ['<v>Good</v>']
    ],
    'category': [
        ['terminology/noun', 'style/greeting'],
        ['grammar/verb']
    ],
    'severity': [
        ['medium', 'low'],
        ['high']
    ]
})

# Add a row with no errors (category is None/empty)
examples_df_no_error = pd.DataFrame({
    'source': ['Perfect sentence'],
    'candidate': ['Oración perfecta'],
    'reference': ['Oración perfecta'],
    'translation_m': ['Oración perfecta alt'],
    'span': [None],
    'category': [None],
    'severity': [None]
})

examples_df = pd.concat([examples_df, examples_df_no_error], ignore_index=True)

ARGS = {
    'examples': examples_df,
    'src_lang': 'en',
    'tgt_lang': 'es',
    'has_reference': True,
    'has_source': True,
    'translation_as_ref': False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
pandas — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
format_prefix Ran
ran on a synthesized input
xuuhuang/lost_in_the_src/automqm/llm_ppl.py
code served (permissive licence) · get_code("1790c7db877d5562")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple, ndim 2, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

# Create a mock tokenizer that implements the __call__ method
class MockTokenizer:
    def __call__(self, texts):
        # Return a mock object with input_ids attribute
        # input_ids should be a list of lists with varying lengths
        if isinstance(texts, str):
            texts = [texts]
        # Create mock token ids with lengths proportional to text length
        return MockTokenizerOutput([len(text.split()) + 5 for text in texts])

class MockTokenizerOutput:
    def __init__(self, lengths):
        self.input_ids = [[i for i in range(length)] for length in lengths]

# Define test data
prefix_type = 'ins_S-R-T'
lang_pair = 'en-zh'
tokenizer = MockTokenizer()
sources = ['Hello world', 'Good morning']
references = ['你好世界', '早上好']
chat_tag = ''

ARGS = {
    'prefix_type': prefix_type,
    'lang_pair': lang_pair,
    'tokenizer': tokenizer,
    'sources': sources,
    'references': references,
    'chat_tag': chat_tag,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 2, dtype family float — a tuple output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
load_file Ran
ran on a synthesized input
xuuhuang/lost_in_the_src/automqm/llm_ppl.py
code served (permissive licence) · get_code("4fa0d7896f0984c2")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import tempfile
import os

# Create a temporary file with some content
temp_dir = tempfile.mkdtemp()
temp_file = os.path.join(temp_dir, "test.txt")

with open(temp_file, 'w') as f:
    f.write("line one\n")
    f.write("line two\n")
    f.write("  line three with spaces  \n")
    f.write("line four\n")

ARGS = {
    "path": temp_file
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
preprocess Ran
ran; our contract draft was wrong, not the code
xuuhuang/lost_in_the_src/automqm/finetune_llama.py
pointer only (licence: MIT); the stored copy get_code serves, from GCYZSL/O3-LLM-UNLEARNING, is pointer only (licence: NONE) · get_code("817d5f759487007d")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from GCYZSL/O3-LLM-UNLEARNING, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

process_type Ran
ran on a synthesized input
xuuhuang/lost_in_the_src/automqm/evaluate.py
code served (permissive licence) · get_code("859ef0e2fe47e47b")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import re

def process_type(output, type):
    if type == 'polish':
        p = re.search('\\nErrors:', output)
        if p is not None:
            r = output[p.end(0):].strip()
        else:
            print("No 'Errors:' in the output")
            r = 'None'
        return r
    else:
        return output


# Setup code
output_with_errors = "Some output\nErrors: This is an error message\nMore details"
output_without_errors = "Some output without errors section"
type_polish = 'polish'
type_other = 'other'

ARGS = {
    'output': output_with_errors,
    'type': type_polish
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
score_label Ran
ran on a synthesized input
xuuhuang/lost_in_the_src/automqm/postprocess_inference.py
code served (permissive licence) · get_code("be66b9284a0d7bbd")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned float, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# The function score_label takes a single parameter: text (a string)
# It counts occurrences of specific substrings and calculates a score based on those counts
# We need to construct a text string that will be passed to the function

text = "Major/minor: Minor and Major/minor: Major and Major/minor: Major"

ARGS = {
    "text": text
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type float, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
span_label Ran
ran on a synthesized input
xuuhuang/lost_in_the_src/automqm/evaluate.py
code served (permissive licence) · get_code("29de5dbe6b96b9f4")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

class Sample:
    def __init__(self, span, category, severity, source, candidate):
        self.span = span
        self.category = category
        self.severity = severity
        self.source = source
        self.candidate = candidate

sample = Sample(
    span=['The <v>quick</v> brown fox', 'jumps <v>over</v> the lazy dog'],
    category=['Accuracy/Omission', 'Fluency'],
    severity=['major', 'minor'],
    source='The quick brown fox jumps over the lazy dog',
    candidate='The fast brown fox jumps over the lazy dog'
)

ARGS = {
    'sample': sample,
    'major': False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
format_template Not yet run
unverified (harvested, no recorded run)
xuuhuang/lost_in_the_src/automqm/automqm.py
code served (permissive licence) · get_code("e9ffc8aced13263c")
How it ran

no invocation recorded.

initialize Not yet run
unverified (harvested, no recorded run)
xuuhuang/lost_in_the_src/automqm/llm_ppl.py
code served (permissive licence) · get_code("5d2cdb987c833aa0")
How it ran

no invocation recorded.

make_supervised_data_module Not yet run
unverified (harvested, no recorded run)
xuuhuang/lost_in_the_src/automqm/finetune_llama.py
code served (permissive licence) · get_code("78553e11ab537e05")
How it ran

no invocation recorded.

preprocess_results Not yet run
unverified (harvested, no recorded run)
xuuhuang/lost_in_the_src/automqm/evaluate.py
code served (permissive licence) · get_code("8d0012e897a9cfff")
How it ran

no invocation recorded.

Also present in other papers' code

Of 13 samples lifted from this paper's linked repositories, 13 carry a function identity and 0 do not. Of the 13 distinct functions with an identity, 1 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 1 the same month, 1 later) and 12 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

FunctionAlso present inEarliest-datedOthers named
preprocess 23 other papers
2 dated earlier, 1 the same month, 20 later
2305.12870 (2023-05)
Lion: Adversarial Distillation of Proprietary Large Language Models
2306.17194 (2023-06)
On the Exploitability of Instruction Tuning
2308.09583 (2023-08)
WizardMath: EMPOWERING MATHEMATICAL REASON-ING FOR LARGE LANGUAGE MODELS VIA Reinforced Evol-Instruct
and 20 more not named here

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Automatic evaluation of machine translation (MT) is a critical tool driving the rapid iterative development of MT systems. While considerable progress has been made on estimating a single scalar quality score, current metrics lack the informativeness of more detailed schemes that annotate individual errors, such as Multidimensional Quality Metrics (MQM). In this paper, we help fill this gap by proposing AutoMQM, a prompting technique which leverages the reasoning and in-context learning capabilities of large language models (LLMs) and asks them to identify and categorize errors in translations. We start by evaluating recent LLMs, such as PaLM and PaLM-2, through simple score prediction prompting, and we study the impact of labeled data through in-context learning and finetuning. We then evaluate AutoMQM with PaLM-2 models, and we find that it improves performance compared to just prompting for scores (with particularly large gains for larger models) while providing interpretability through error spans that align with human annotations.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2308.07286")
get_code_for_paper("2308.07286")
have("2308.07286")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2308.07286.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2308.07286.svg)](https://syntology.ai/paper/2308.07286)

Connect an agent — have() is free.