SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2202.08479 · EMNLP · 2022

On the Evaluation Metrics for Paraphrase Generation

Lemao Liu, Shuming Shi, Haiyun Jiang, Lingfeng Shen

arXiv · PDF · Open in the Atlas

Code that ran

10 of 14 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; 2 failures 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/2202.08479.json · badge: /api/ran/2202.08479.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
shadowkiller33/parascore canonical (classifier label)
classifier basis not recorded on the link
10 ran · 4 no recorded run
FunctionStatusWhere it lives
average Ran
ran on a synthesized input
shadowkiller33/parascore/src/utils.py
code served (permissive licence) · get_code("5e189add31fe0b72")
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. 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 torch
import numpy as np

torch.manual_seed(0)

# Analyzing the function:
# - It takes `lists` as parameter
# - It iterates through lists[i] with range(len(lists))
# - For each lists[i], it iterates through `lst` in lists[i]
# - For each lst, it computes sum(lst) / len(lst)
# So lists is a 3-level nested structure: list of list of lists
# The innermost lists should be numeric sequences (they get sum() and len() applied)

# Create a 3-level nested list structure
# lists[i] contains multiple lists
# Each innermost list contains numbers

lists = [
    [[1, 2, 3], [4, 5], [6, 7, 8, 9]],  # First outer list
    [[10, 20], [30, 40, 50]],             # Second outer list
    [[100, 200, 300, 400], [500]]         # Third outer list
]

ARGS = {
    "lists": lists
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list — a list 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
edit Ran
ran on a synthesized input
shadowkiller33/parascore/src/utils.py
code served (permissive licence) · get_code("3fede4adffad7e69")
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 nltk
import torch
import torch.nn as nn

# Set seed for reproducibility
torch.manual_seed(0)

# Analyze the function:
# - x: a string (has .replace() method called on it)
# - y: a string (has .replace() method called on it)
# - chinese: either 0 (default) or 'bq' (a string for comparison)
# The function computes edit distance between strings and normalizes by max length

ARGS = {
    'x': 'hello',
    'y': 'hallo',
    'chinese': 0
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
nltk, 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
get_tokenizer Ran
ran on a synthesized input
shadowkiller33/parascore/bert_score/utils.py
code served (permissive licence) · get_code("0789b5ae053b6b74")
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 MagicMock. 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 os
import tempfile
import json
from unittest.mock import patch, MagicMock

# Create a temporary directory for cache
temp_cache = tempfile.mkdtemp()

# Mock the AutoTokenizer.from_pretrained to avoid actual downloads
def mock_from_pretrained(model_type, use_fast=False):
    mock_tokenizer = MagicMock()
    mock_tokenizer.encode = MagicMock(return_value=[101, 2054, 2003, 102])
    mock_tokenizer.decode = MagicMock(return_value="[CLS] what is [SEP]")
    return mock_tokenizer

# For non-scibert models, we can pass a simple string
# For scibert models, we need to handle the caching logic
# Since we want to avoid actual downloads, we'll use a non-scibert model

ARGS = {
    'model_type': 'bert-base-uncased'
}

# Patch AutoTokenizer.from_pretrained to avoid network calls
import transformers
original_from_pretrained = transformers.AutoTokenizer.from_pretrained
transformers.AutoTokenizer.from_pretrained = mock_from_pretrained
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
distutils, transformers — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type MagicMock — a MagicMock output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a MagicMock output the recorded ndim 1 and dtype family None 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
isChineseChar Ran
ran on a synthesized input
shadowkiller33/parascore/src/tokenizeChinese.py
code served (permissive licence) · get_code("a90c4324948923f4")
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 torch

# The function isChineseChar takes a single parameter: uchar
# It expects a unicode character (string of length 1)
# The function checks if the character falls within various Chinese character Unicode ranges

# Test with a valid Chinese character
ARGS = {
    "uchar": u'中'  # A common Chinese character that should return True
}
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 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
max_pooling Ran
ran on a synthesized input
shadowkiller33/parascore/src/utils.py
code served (permissive licence) · get_code("10e7ac066353d8e1")
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 Tensor, ndim 2, dtype family float, shape (2, 4) 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)

# Build model_output as a tuple where first element is token_embeddings
# Shape: (batch_size=2, sequence_length=3, embedding_dim=4)
token_embeddings = torch.randn(2, 3, 4)
model_output = (token_embeddings,)

# Build attention_mask
# Shape: (batch_size=2, sequence_length=3)
attention_mask = torch.tensor([
    [1, 1, 0],
    [1, 0, 0]
], dtype=torch.long)

ARGS = {
    "model_output": model_output,
    "attention_mask": attention_mask,
}
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 Tensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) 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
read_file_to_list Ran
ran; our contract draft was wrong, not the code
shadowkiller33/parascore/BARTScore/utils.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from gersteinlab/Struc-Bench, is pointer only (licence: NONE) · get_code("8b3846aabf2bb7ef")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from gersteinlab/Struc-Bench, 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.

read_pickle Ran
ran; our contract draft was wrong, not the code
shadowkiller33/parascore/BARTScore/utils.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from gersteinlab/Struc-Bench, is pointer only (licence: NONE) · get_code("1ab147c8cefa1585")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from gersteinlab/Struc-Bench, 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.

score Ran
ran on a synthesized input
shadowkiller33/parascore/bert_score/score.py
code served (permissive licence) · get_code("8f1719ca230ca245")
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. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is 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
from collections import defaultdict
import pandas as pd

torch.manual_seed(0)

# Mock tokenizer
class MockTokenizer:
    def __init__(self):
        self.sep_token_id = 102
        self.cls_token_id = 101
        self.pad_token_id = 0
    
    def encode(self, text, add_special_tokens=True):
        return [101] + [1000, 1001] + [102]
    
    def batch_encode_plus(self, texts, padding=True, return_tensors=None):
        input_ids = []
        attention_mask = []
        for text in texts:
            ids = self.encode(text)
            input_ids.append(ids)
            attention_mask.append([1] * len(ids))
        
        max_len = max(len(ids) for ids in input_ids)
        for ids in input_ids:
            ids.extend([self.pad_token_id] * (max_len - len(ids)))
        for mask in attention_mask:
            mask.extend([0] * (max_len - len(mask)))
        
        result = {
            'input_ids': torch.tensor(input_ids),
            'attention_mask': torch.tensor(attention_mask)
        }
        return result

# Mock BERT model
class MockBertModel(nn.Module):
    def __init__(self, hidden_size=4, num_layers=2):
        super().__init__()
        self.hidden_size = hidden_size
        self.num_layers = num_layers
        self.embeddings = nn.Embedding(2000, hidden_size)
        self.layers = nn.ModuleList([
            nn.Linear(hidden_size, hidden_size) for _ in range(num_layers)
        ])
    
    def forward(self, input_ids, attention_mask=None, output_all_encoded_layers=False):
        x = self.embeddings(input_ids)
        all_layers_output = [x]
        
        for layer in self.layers:
            x = layer(x)
            all_layers_output.append(x)
        
        if output_all_encoded_layers:
            return all_layers_output
        else:
            return all_layers_output[-1]

# Mock functions that would be imported
def mock_get_tokenizer(model_type):
    return MockTokenizer()

def mock_get_model(model_type, num_layers, all_layers=False):
    model = MockBertModel(hidden_size=4, num_layers=num_layers)
    return model

def mock_bert_cos_score_idf(model, refs, cands, tokenizer, idf_dict, verbose=False, device='cpu', batch_size=64, all_layers=False):
    # Returns tensor of shape (len(refs), 3) for (P, R, F) scores
    num_pairs = len(cands)
    scores = torch.rand(num_pairs, 3)
    return scores

def mock_get_idf_dict(refs, tokenizer, nthreads=4):
    idf_dict = defaultdict(lambda: 1.0)
    idf_dict[tokenizer.sep_token_id] = 0
    idf_dict[tokenizer.cls_token_id] = 0
    return idf_dict

# Monkey-patch the functions in the module namespace
import sys
mock_module = sys.modules[__name__]
mock_module.get_tokenizer = mock_get_tokenizer
mock_module.get_model = mock_get_model
mock_module.bert_cos_score_idf = mock_bert_cos_score_idf
mock_module.get_idf_dict = mock_get_idf_dict
mock_module.lang2model = {'en': 'bert-base-uncased', 'zh': 'bert-base-chinese'}
mock_module.model2layers = {'bert-base-uncased': 9, 'bert-base-chinese': 9}

# Test data
ARGS = {
    'cands': [
        'This is a candidate sentence.',
        'Another candidate sentence here.'
    ],
    'refs': [
        'This is a reference sentence.',
        'Another reference sentence here.'
    ],
    'model_type': 'bert-base-uncased',
    'num_layers': 9,
    'verbose': False,
    'idf': False,
    'device': 'cpu',
    'batch_size': 2,
    'nthreads': 4,
    'all_layers': False,
    'lang': 'en',
    'return_hash': False,
    'rescale_with_baseline': False,
    'baseline_path': None
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
pandas, 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
tokenizeSentence Ran
ran on a synthesized input
shadowkiller33/parascore/src/tokenizeChinese.py
code served (permissive licence) · get_code("39127f87f72d1142")
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 re

def isChineseChar(uchar):
    if uchar >= u'㐀' and uchar <= u'䶵':
        return True
    elif uchar >= u'一' and uchar <= u'龥':
        return True
    elif uchar >= u'龦' and uchar <= u'龻':
        return True
    elif uchar >= u'豈' and uchar <= u'鶴':
        return True
    elif uchar >= u'侮' and uchar <= u'頻':
        return True
    elif uchar >= u'並' and uchar <= u'龎':
        return True
    elif uchar >= u'\u20000' and uchar <= u'⩭6':
        return True
    elif uchar >= u'⾀0' and uchar <= u'⾡d':
        return True
    elif uchar >= u'\uff00' and uchar <= u'\uffef':
        return True
    elif uchar >= u'⺀' and uchar <= u'\u2eff':
        return True
    elif uchar >= u'\u3000' and uchar <= u'〿':
        return True
    elif uchar >= u'㇀' and uchar <= u'㇯':
        return True
    elif uchar >= u'⼀' and uchar <= u'\u2fdf':
        return True
    elif uchar >= u'⿰' and uchar <= u'⿿':
        return True
    elif uchar >= u'\u3100' and uchar <= u'ㄯ':
        return True
    elif uchar >= u'ㆠ' and uchar <= u'ㆿ':
        return True
    elif uchar >= u'︐' and uchar <= u'\ufe1f':
        return True
    elif uchar >= u'︰' and uchar <= u'﹏':
        return True
    elif uchar >= u'☀' and uchar <= u'⛿':
        return True
    elif uchar >= u'✀' and uchar <= u'➿':
        return True
    elif uchar >= u'㈀' and uchar <= u'㋿':
        return True
    elif uchar >= u'㌀' and uchar <= u'㏿':
        return True
    else:
        return False

def tokenizeString(sentence, lc=False):
    sentence = sentence.strip()
    sentence_in_chars = ''
    for c in sentence:
        if isChineseChar(c):
            sentence_in_chars += ' '
            sentence_in_chars += c
            sentence_in_chars += ' '
        else:
            sentence_in_chars += c
    sentence = sentence_in_chars
    if lc:
        sentence = sentence.lower()
    sentence = re.sub('([\\{-\\~\\[-\\` -\\&\\(-\\+\\:-\\@\\/])', ' \\1 ', sentence)
    sentence = re.sub('([^0-9])([\\.,])', '\\1 \\2 ', sentence)
    sentence = re.sub('([\\.,])([^0-9])', ' \\1 \\2', sentence)
    sentence = re.sub('([0-9])(-)', '\\1 \\2 ', sentence)
    sentence = re.sub('\\s+', ' ', sentence)
    sentence = re.sub('^\\s+', '', sentence)
    sentence = re.sub('\\s+$', '', sentence)
    return sentence

def tokenizeSentence(sentences):
    output = []
    for sentence in sentences:
        if sentence.startswith('<seg'):
            start = sentence.find('>') + 1
            end = sentence.rfind('<')
            new_sentence = sentence[:start] + tokenizeString(sentence[start:end]) + sentence[end:]
        else:
            new_sentence = tokenizeString(sentence)
        output.append(new_sentence)
    return output


ARGS = {
    "sentences": [
        "Hello world, this is a test.",
        "<seg id='1'>This is another test with punctuation!</seg>",
    ]
}
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
tokenizeString Ran
ran on a synthesized input
shadowkiller33/parascore/src/tokenizeChinese.py
code served (permissive licence) · get_code("3622ee6e7373fea5")
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 isChineseChar(uchar):
    if uchar >= u'㐀' and uchar <= u'䶵':
        return True
    elif uchar >= u'一' and uchar <= u'龥':
        return True
    elif uchar >= u'龦' and uchar <= u'龻':
        return True
    elif uchar >= u'豈' and uchar <= u'鶴':
        return True
    elif uchar >= u'侮' and uchar <= u'頻':
        return True
    elif uchar >= u'並' and uchar <= u'龎':
        return True
    elif uchar >= u'\u20000' and uchar <= u'⩭6':
        return True
    elif uchar >= u'⾀0' and uchar <= u'⾡d':
        return True
    elif uchar >= u'\uff00' and uchar <= u'\uffef':
        return True
    elif uchar >= u'⺀' and uchar <= u'\u2eff':
        return True
    elif uchar >= u'\u3000' and uchar <= u'〿':
        return True
    elif uchar >= u'㇀' and uchar <= u'㇯':
        return True
    elif uchar >= u'⼀' and uchar <= u'\u2fdf':
        return True
    elif uchar >= u'⿰' and uchar <= u'⿿':
        return True
    elif uchar >= u'\u3100' and uchar <= u'ㄯ':
        return True
    elif uchar >= u'ㆠ' and uchar <= u'ㆿ':
        return True
    elif uchar >= u'︐' and uchar <= u'\ufe1f':
        return True
    elif uchar >= u'︰' and uchar <= u'﹏':
        return True
    elif uchar >= u'☀' and uchar <= u'⛿':
        return True
    elif uchar >= u'✀' and uchar <= u'➿':
        return True
    elif uchar >= u'㈀' and uchar <= u'㋿':
        return True
    elif uchar >= u'㌀' and uchar <= u'㏿':
        return True
    else:
        return False

def tokenizeString(sentence, lc=False):
    """
    :param sentence: input sentence
    :param lc: flag of lowercase. default=True
    :return: tokenized sentence, with 

    """
    sentence = sentence.strip()
    sentence_in_chars = ''
    for c in sentence:
        if isChineseChar(c):
            sentence_in_chars += ' '
            sentence_in_chars += c
            sentence_in_chars += ' '
        else:
            sentence_in_chars += c
    sentence = sentence_in_chars
    if lc:
        sentence = sentence.lower()
    sentence = re.sub('([\\{-\\~\\[-\\` -\\&\\(-\\+\\:-\\@\\/])', ' \\1 ', sentence)
    sentence = re.sub('([^0-9])([\\.,])', '\\1 \\2 ', sentence)
    sentence = re.sub('([\\.,])([^0-9])', ' \\1 \\2', sentence)
    sentence = re.sub('([0-9])(-)', '\\1 \\2 ', sentence)
    sentence = re.sub('\\s+', ' ', sentence)
    sentence = re.sub('^\\s+', '', sentence)
    sentence = re.sub('\\s+$', '', sentence)
    return sentence

ARGS = {
    "sentence": "Hello world! This is a test sentence with some numbers 123 and punctuation.",
    "lc": 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 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
cal_meteor Not yet run
unverified (harvested, no recorded run)
shadowkiller33/parascore/src/baseline.py
code served (permissive licence) · get_code("b49bbd4994987830")
How it ran

no invocation recorded.

cal_sen_bleu Not yet run
unverified (harvested, no recorded run)
shadowkiller33/parascore/src/baseline.py
code served (permissive licence) · get_code("12aff521567b9608")
How it ran

no invocation recorded.

get_model Not yet run
unverified (harvested, no recorded run)
shadowkiller33/parascore/bert_score/utils.py
code served (permissive licence) · get_code("9caae53148142d2b")
How it ran

no invocation recorded.

sent_encode Not yet run
unverified (harvested, no recorded run)
shadowkiller33/parascore/bert_score/utils.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from cyr19/reproducibility, is pointer only (licence: NONE) · get_code("7aa792f0b1258927")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from cyr19/reproducibility, 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. No invocation recorded.

Also present in other papers' code

Of 14 samples lifted from this paper's linked repositories, 14 carry a function identity and 0 do not. Of the 14 distinct functions with an identity, 4 are also present in other papers whose code we hold (2 with at least one paper dated earlier, 0 the same month, 3 later) and 10 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
sent_encode 3 other papers
1 dated earlier, 2 later · 1 only as an assembly brick
1904.09675 (2019-04)
BERTScore: Evaluating Text Generation with BERT
2207.05987 (2022-07)
DocPrompting: GENERATING CODE BY RETRIEVING THE DOCS
2209.13654 (2022-09)
Embarrassingly Easy Document-Level MT Metrics: How to Convert Any Pretrained Metric Into a Document-Level Metric
get_model 2 other papers
2 dated earlier
1904.09675 (2019-04)
BERTScore: Evaluating Text Generation with BERT
2106.02208 (2021-06)
BERTTune: Fine-Tuning Neural Machine Translation with BERTScore
read_file_to_list 2 other papers
2 later
2024.naacl-long.187 (2024 (year only))
Fair Abstractive Summarization of Diverse Perspectives
2404.14215 (2024-04)
Text-Tuple-Table: Towards Information Integration in Text-to-Table Generation via Global Tuple Extraction
read_pickle 2 other papers
2 later
2024.naacl-long.187 (2024 (year only))
Fair Abstractive Summarization of Diverse Perspectives
2404.14215 (2024-04)
Text-Tuple-Table: Towards Information Integration in Text-to-Table Generation via Global Tuple Extraction

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

In this paper we revisit automatic metrics for paraphrase evaluation and obtain two findings that disobey conventional wisdom: (1) Reference-free metrics achieve better performance than their reference-based counterparts. (2) Most commonly used metrics do not align well with human annotation. Underlying reasons behind the above findings are explored through additional experiments and in-depth analyses. Based on the experiments and analyses, we propose ParaScore, a new evaluation metric for paraphrase generation. It possesses the merits of referencebased and reference-free metrics and explicitly models lexical divergence. Based on our analysis and improvements, our proposed reference-based outperforms than referencefree metrics. Experimental results demonstrate that ParaScore significantly outperforms existing metrics. Our codes and toolkit are released in https://github.com/ shadowkiller33/ParaScore.

For agents

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

get_harvested_code_for_paper("2202.08479")
get_code_for_paper("2202.08479")
have("2202.08479")

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

curl https://syntology.ai/api/ran/2202.08479.json

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

[![Syntology run record](https://syntology.ai/api/ran/2202.08479.svg)](https://syntology.ai/paper/2202.08479)

Connect an agent — have() is free.