SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2104.00783 · NAACL · 2021

Action-Based Conversations Dataset: A Corpus for Building More In-Depth Task-Oriented Dialogue Systems

Yi Yang, Zhou Yu, Derek Chen, Howard Chen, Alexander Lin

arXiv · PDF · Open in the Atlas

Code that ran

9 of 11 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 0 have no contract verdict; 9 failures of our instrument; 2 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/2104.00783.json · badge: /api/ran/2104.00783.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
asappresearch/abcd canonical (classifier label)
classifier basis: census
9 ran · 2 no recorded run
FunctionStatusWhere it lives
ast_report Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
asappresearch/abcd/utils/evaluate.py
code served (permissive licence) · get_code("617ad3e5efd78dcd")
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 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 numpy as np

# Create predictions: (bslot_preds, value_preds)
# bslot_preds should be shape (batch_size, num_bslot_classes)
# value_preds should be shape (batch_size, num_value_classes)
batch_size = 2
num_bslot_classes = 4
num_value_classes = 3

bslot_preds = np.random.randn(batch_size, num_bslot_classes).astype(np.float32)
value_preds = np.random.randn(batch_size, num_value_classes).astype(np.float32)
predictions = (bslot_preds, value_preds)

# Create labels: (bslot_labels, value_labels)
# bslot_labels should be shape (batch_size,) with class indices
# value_labels should be shape (batch_size,) with class indices
bslot_labels = np.array([0, 2], dtype=np.int64)
value_labels = np.array([1, 0], dtype=np.int64)
labels = (bslot_labels, value_labels)

ARGS = {
    'predictions': predictions,
    'labels': labels
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 1, 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
check_cache Ran
ran; our contract draft was wrong, not the code
asappresearch/abcd/utils/help.py
code served (permissive licence) · get_code("5b554aaa4782e58b")
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 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 os
import tempfile

class Args:
    def __init__(self):
        self.model_type = "bert"
        self.task = "classification"
        self.cascade = False
        self.use_intent = False

# Create a temporary directory for testing
cache_dir = tempfile.mkdtemp()

# Construct the args object
args = Args()

# Define ARGS dict
ARGS = {
    'args': args,
    'cache_dir': cache_dir
}
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, ndim 1, 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
check_directories Ran
ran; our contract draft was wrong, not the code
asappresearch/abcd/utils/help.py
code served (permissive licence) · get_code("749a1a25d79c4775")
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.

import os
import tempfile

class Args:
    def __init__(self):
        self.model_type = "bert"
        self.task = "intent_detection"
        self.cascade = False
        self.use_intent = True
        self.input_dir = tempfile.mkdtemp()
        self.output_dir = tempfile.mkdtemp()
        self.prefix = "exp"
        self.filename = "run1"
        self.suffix = "v1"

args_instance = Args()

ARGS = {
    "args": args_instance,
}
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
load_candidates Ran
ran; our contract draft was wrong, not the code
asappresearch/abcd/utils/load.py
code served (permissive licence) · get_code("a7ab719397915122")
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.

import json
import torch
import tempfile
import os

torch.manual_seed(0)

# Create a temporary directory for input files
temp_dir = tempfile.mkdtemp()

# Create utterances.json file
utterances_data = [
    "hello world",
    "how are you"
]
with open(f'{temp_dir}/utterances.json', 'w') as f:
    json.dump(utterances_data, f)

# Create utt_vectors.pt file
utt_vectors = torch.randn(2, 4)
torch.save(utt_vectors, f'{temp_dir}/utt_vectors.pt')

# Create a simple args object
class Args:
    def __init__(self, input_dir):
        self.input_dir = input_dir

ARGS = {
    'args': Args(temp_dir)
}
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 — 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
load_data Ran
ran; our contract draft was wrong, not the code
asappresearch/abcd/utils/load.py
code served (permissive licence) · get_code("2efa83d98dfd8f8a")
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 dict. 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 sys
import json
import tempfile

# Create a temporary directory and a mock JSON file for testing
temp_dir = tempfile.mkdtemp()
mock_data = {"samples": [{"id": 1, "text": "example"}]}
mock_file_path = os.path.join(temp_dir, 'abcd_v1.json')
with open(mock_file_path, 'w') as f:
    json.dump(mock_data, f)

# Create a minimal args object with required attributes
class Args:
    def __init__(self, input_dir, version):
        self.input_dir = input_dir
        self.version = version

args = Args(input_dir=temp_dir, version='1')
already_cached = False

ARGS = {
    'args': args,
    'already_cached': already_cached
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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
prepare_action_labels Ran
ran; our contract draft was wrong, not the code
asappresearch/abcd/utils/process.py
code served (permissive licence) · get_code("40ff57a189e78563")
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 dict. 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

def prepare_action_labels(ontology):
    action_list = []
    for section, buttons in ontology['actions'].items():
        actions = buttons.keys()
        action_list.extend(actions)
    return {action: idx for idx, action in enumerate(action_list)}

ontology = {
    'actions': {
        'section1': {
            'action_a': {'label': 'Action A'},
            'action_b': {'label': 'Action B'},
            'action_c': {'label': 'Action C'}
        },
        'section2': {
            'action_d': {'label': 'Action D'},
            'action_e': {'label': 'Action E'}
        }
    }
}

ARGS = {
    'ontology': ontology
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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
prepare_intent_labels Ran
ran; our contract draft was wrong, not the code
asappresearch/abcd/utils/process.py
code served (permissive licence) · get_code("2d1639331600ea76")
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 dict. 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

# Construct the ontology dict that the function expects
# The function accesses ontology['intents']['subflows']
# and iterates through its items, extending intent_list with subflows

ontology = {
    'intents': {
        'subflows': {
            'flow_1': ['intent_a', 'intent_b', 'intent_c'],
            'flow_2': ['intent_d', 'intent_e'],
            'flow_3': ['intent_f']
        }
    }
}

ARGS = {
    'ontology': ontology
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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
setup_dataloader Ran
ran; our contract draft was wrong, not the code
asappresearch/abcd/utils/process.py
code served (permissive licence) · get_code("c560474429a6b107")
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 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 torch
import torch.nn as nn
from torch.utils.data import Dataset

torch.manual_seed(0)

class DummyDataset(Dataset):
    def __init__(self, size=10):
        self.size = size
        self.data = [
            (torch.randn(4), torch.tensor(i % 2))
            for i in range(size)
        ]
    
    def __len__(self):
        return self.size
    
    def __getitem__(self, idx):
        return self.data[idx]
    
    def collate_func(self, batch):
        x_list = [item[0] for item in batch]
        y_list = [item[1] for item in batch]
        x = torch.stack(x_list)
        y = torch.stack(y_list)
        return (x, y)

train_dataset = DummyDataset(size=20)
val_dataset = DummyDataset(size=10)

datasets = {
    'train': train_dataset,
    'val': val_dataset
}

ARGS = {
    'datasets': datasets,
    'batch_size': 4,
    'split': 'train'
}
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 1, 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
setup_gpus Ran
ran; our contract draft was wrong, not the code
asappresearch/abcd/utils/help.py
code served (permissive licence) · get_code("38bb85bb95096b63")
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 Args. 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 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

class Args:
    def __init__(self):
        self.debug = False
        self.epochs = 10

ARGS = {
    "args": Args()
}
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 Args — a Args output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Args 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
load_tokenizer Not yet run
unverified (harvested, no recorded run)
asappresearch/abcd/utils/load.py
code served (permissive licence) · get_code("10442a3755e649e8")
How it ran

no invocation recorded.

ranking_report Not yet run
unverified (harvested, no recorded run)
asappresearch/abcd/utils/evaluate.py
code served (permissive licence) · get_code("cb0b6764e6ec1900")
How it ran

no invocation recorded.

Also present in other papers' code

Of 11 samples lifted from this paper's linked repositories, 11 carry a function identity and 0 do not. Of the 11 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 11 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.

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

Existing goal-oriented dialogue datasets focus mainly on identifying slots and values. However, customer support interactions in reality often involve agents following multi-step procedures derived from explicitly-defined company policies as well. To study customer service dialogue systems in more realistic settings, we introduce the Action-Based Conversations Dataset (ABCD), a fully-labeled dataset with over 10K human-to-human dialogues containing 55 distinct user intents requiring unique sequences of actions constrained by policies to achieve task success. We propose two additional dialog tasks, Action State Tracking and Cascading Dialogue Success, and establish a series of baselines involving large-scale, pre-trained language models on this dataset. Empirical results demonstrate that while more sophisticated networks outperform simpler models, a considerable gap (50.8% absolute accuracy) still exists to reach human-level performance on ABCD. 1 Subflows recover-username, 1 recover-password, 1 reset-2fa, 1 status-service-added, 2 status-service-removed, 2 statusshipping-question, 2 status-credit-missing, 2 manage-change-address, 2 manage-change-name, 2 manage-changephone, 2 manage-payment-method, 2 status-mystery-fee, 3 status-delivery-time, 3 status-payment-method, 3 statusquantity, 3 manage-upgrade, 3 manage-downgrade, 3 manage-create, 3 manage-cancel, 3 refund-initiate, 4 refundupdate, 4 refund-status, 4 return-stain, 4 return-color, 4 return-size, 4 bad-price-competitor, 5 bad-price-yesterday, 5 out-of-stock-general, 5 out-of-stock-one-item, 5 promo-code-invalid, 5 promo-code-out-of-date, 5 mistimed-billingalready-returned, 5 mistimed-billing-never-bought, 5 status, 6 manage, 6 missing, 6 cost, 6 boots, 7 shirt, 7 jeans, 7 jacket, 7 pricing, 8 membership, 8 timing, 8 policy, 8 status-active, 9 status-due-amount, 9 status-due-date, 9 manage-pay-bill, 9 manage-extension, 9 manage-dispute-bill, 9 credit-card, 10 shopping-cart, 10 search-results, 10 slow-speed 10

For agents

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

get_harvested_code_for_paper("2104.00783")
get_code_for_paper("2104.00783")
have("2104.00783")

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

curl https://syntology.ai/api/ran/2104.00783.json

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

[![Syntology run record](https://syntology.ai/api/ran/2104.00783.svg)](https://syntology.ai/paper/2104.00783)

Connect an agent — have() is free.