SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2606.25552 · 2026

SFL-MTSC: Leveraging Semantic Frame-Level Multi-Task Self-Consistency for Robust Multi-Intent Spoken Language Understanding

Berlin Chen, Po-Yen Chen

arXiv · PDF · Open in the Atlas

Code that ran

10 of 10 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; 3 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 7 failures of our instrument; 0 no recorded run), measured 2026-11-22; 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/2606.25552.json · badge: /api/ran/2606.25552.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
boyan1001/SFL-MTSC community
Papers with Code archive
10 ran · 0 no recorded run
FunctionStatusWhere it lives
SFL_MTSC Ran
ran; our contract draft was wrong, not the code
boyan1001/SFL-MTSC/self-consistency.py
code served (permissive licence) · get_code("1248b9954a5da626")
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) 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 math
from collections import defaultdict, deque
from typing import Any, Dict, List, Set, Tuple

# Create sample frame data with domain, intent, and slots
frame1_path0 = {
    'domain': 'restaurant',
    'intent': 'book',
    'slots': {'cuisine': 'italian', 'party_size': '4'}
}

frame2_path0 = {
    'domain': 'restaurant',
    'intent': 'book',
    'slots': {'cuisine': 'italian', 'party_size': '4'}
}

frame3_path0 = {
    'domain': 'restaurant',
    'intent': 'book',
    'slots': {'cuisine': 'french', 'party_size': '2'}
}

frame1_path1 = {
    'domain': 'restaurant',
    'intent': 'book',
    'slots': {'cuisine': 'italian', 'party_size': '4'}
}

frame2_path1 = {
    'domain': 'hotel',
    'intent': 'search',
    'slots': {'location': 'paris', 'stars': '5'}
}

frame3_path1 = {
    'domain': 'hotel',
    'intent': 'search',
    'slots': {'location': 'paris', 'stars': '4'}
}

# Create semantics list - list of lists of frames
semantics = [
    [frame1_path0, frame2_path0, frame3_path0],
    [frame1_path1, frame2_path1, frame3_path1]
]

ARGS = {
    'semantics': semantics,
    'alpha': 0.3,
    'tau': 0.55
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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
cluster_hybrid_jaccard Ran
ran; our contract draft was wrong, not the code
boyan1001/SFL-MTSC/self-consistency.py
code served (permissive licence) · get_code("0042b24f4b3203f4")
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
from collections import defaultdict, deque
from typing import Any, Dict, List, Set, Tuple

# Build minimal frame dictionaries that the function expects
# The function calls frame.get('slots', {}) and expects a dict of key-value pairs

frame1 = {
    'slots': {
        'name': 'Alice',
        'age': '30'
    }
}

frame2 = {
    'slots': {
        'name': 'Alice',
        'age': '30'
    }
}

frame3 = {
    'slots': {
        'name': 'Bob',
        'age': '25'
    }
}

frame4 = {
    'slots': {
        'city': 'New York',
        'state': 'NY'
    }
}

# Create a list of frames to cluster
frames = [frame1, frame2, frame3, frame4]

# Define arguments for cluster_hybrid_jaccard
ARGS = {
    'frames': frames,
    'alpha': 0.3,
    'tau': 0.55
}
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
compute_support Ran
ran, honoured the contract we drafted
boyan1001/SFL-MTSC/self-consistency.py
code served (permissive licence) · get_code("92c01831e7a2db56")
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 int, ndim 0, dtype family int, 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
import numpy as np

# The function expects a List[Dict] where each dict can contain a '_path_idx' key
# It iterates through the cluster, extracts '_path_idx' values, and counts unique indices

# Create a sample cluster with frames (dicts) containing '_path_idx' keys
cluster = [
    {'_path_idx': 0, 'data': 'frame1'},
    {'_path_idx': 1, 'data': 'frame2'},
    {'_path_idx': 0, 'data': 'frame3'},  # duplicate path_idx
    {'_path_idx': 2, 'data': 'frame4'},
    {'data': 'frame5'},  # no '_path_idx' key
]

ARGS = {
    'cluster': cluster
}
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 int, ndim 0, dtype family int, 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
filter_slot_clusters Ran
ran; our contract draft was wrong, not the code
boyan1001/SFL-MTSC/self-consistency.py
code served (permissive licence) · get_code("cf78332e15a4a1e4")
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 defaultdict. 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 math
from collections import defaultdict, deque
from typing import Any, Dict, List, Set, Tuple

def compute_support(cluster: List[Dict]) -> int:
    path_indices = set()
    for frame in cluster:
        p = frame.get('_path_idx')
        if p is not None:
            path_indices.add(p)
    return len(path_indices)

def filter_slot_clusters(slot_clusters, k):
    sup_threshold = math.ceil(k / 2)
    retained = defaultdict(list)
    for di_key, clusters in slot_clusters.items():
        for cluster in clusters:
            support = compute_support(cluster)
            if support >= sup_threshold:
                retained[di_key].append(cluster)
    return retained

# Construct test data
slot_clusters = {
    "slot_1": [
        [
            {"_path_idx": 0, "data": "frame0"},
            {"_path_idx": 1, "data": "frame1"},
            {"_path_idx": 0, "data": "frame2"},
        ],
        [
            {"_path_idx": 2, "data": "frame3"},
        ],
    ],
    "slot_2": [
        [
            {"_path_idx": 0, "data": "frame4"},
            {"_path_idx": 1, "data": "frame5"},
            {"_path_idx": 2, "data": "frame6"},
        ],
        [
            {"_path_idx": 3, "data": "frame7"},
            {"_path_idx": 4, "data": "frame8"},
        ],
    ],
}

k = 4

ARGS = {
    "slot_clusters": slot_clusters,
    "k": k,
}
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 defaultdict — a defaultdict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a defaultdict 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
hybrid_jaccard Ran
ran, honoured the contract we drafted
boyan1001/SFL-MTSC/self-consistency.py
code served (permissive licence) · get_code("3dcb645e4d8116d4")
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
import numpy as np
from typing import Any, Dict, List, Set, Tuple

# Construct frame objects (dictionaries with 'slots' key)
frame_a = {
    'slots': {
        'restaurant_type': 'italian',
        'location': 'downtown',
        'price_range': 'moderate'
    }
}

frame_b = {
    'slots': {
        'restaurant_type': 'italian',
        'location': 'uptown',
        'cuisine': 'pasta'
    }
}

# Define ARGS with all parameters for hybrid_jaccard
ARGS = {
    'fa': frame_a,
    'fb': frame_b,
    'alpha': 0.3
}
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 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
jaccard Ran
ran, honoured the contract we drafted
boyan1001/SFL-MTSC/self-consistency.py
code served (permissive licence) · get_code("ffd9acbb3ca646ac")
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
import numpy as np
from typing import Any, Dict, List, Set, Tuple

# Test case 1: Two non-empty sets with some intersection
set_a = {1, 2, 3, 4, 5}
set_b = {3, 4, 5, 6, 7}

ARGS = {
    "a": set_a,
    "b": set_b,
}
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 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
reintegrate_cluster Ran
ran; our contract draft was wrong, not the code
boyan1001/SFL-MTSC/self-consistency.py
code served (permissive licence) · get_code("daf246cda5e400ef")
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 math
from collections import defaultdict, deque
from typing import Any, Dict, List, Set, Tuple

# Construct valid arguments for reintegrate_cluster

# di_key: Tuple of (domain, intent)
di_key = ("restaurant", "book_table")

# cluster: List[Dict] - list of frame dictionaries with 'slots' key
cluster = [
    {
        'slots': {
            'restaurant_type': 'italian',
            'party_size': '4',
            'time': '19:00'
        }
    },
    {
        'slots': {
            'restaurant_type': 'italian',
            'party_size': '4',
            'location': 'downtown'
        }
    },
    {
        'slots': {
            'restaurant_type': 'italian',
            'party_size': '5',
            'time': '19:00'
        }
    }
]

ARGS = {
    'di_key': di_key,
    'cluster': cluster
}
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
slot_pair Ran
ran; our contract draft was wrong, not the code
boyan1001/SFL-MTSC/self-consistency.py
code served (permissive licence) · get_code("559ffc35906cec9b")
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 set. 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:
# - Takes 'frame' parameter
# - Calls frame.get('slots', {}) - so frame must be dict-like
# - Accesses slots as a dict with .items()
# - Returns a set of tuples of strings

# Create a simple frame dict with slots
frame = {
    'slots': {
        'slot_name_1': 'slot_value_1',
        'slot_name_2': 'slot_value_2',
        'slot_name_3': 42,  # Test non-string value (will be converted to str)
    }
}

ARGS = {
    'frame': frame
}
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 set — a set output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a set 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
slot_values Ran
ran; our contract draft was wrong, not the code
boyan1001/SFL-MTSC/self-consistency.py
code served (permissive licence) · get_code("2c009fecb1b66ecb")
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 set. 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
from typing import Any, Dict, List, Set, Tuple

# Test case 1: frame with slots as a dict
frame_with_slots = {
    'slots': {
        'name': 'John',
        'age': 25,
        'city': 'New York'
    }
}

# Test case 2: frame with empty slots dict
frame_empty_slots = {
    'slots': {}
}

# Test case 3: frame with slots as None
frame_none_slots = {
    'slots': None
}

# Test case 4: frame with slots as non-dict (invalid)
frame_invalid_slots = {
    'slots': 'not a dict'
}

# Test case 5: frame without slots key
frame_no_slots_key = {
    'other_key': 'value'
}

# Use the first test case as the main argument
ARGS = {
    'frame': frame_with_slots
}
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 set — a set output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a set 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
voting_key Ran
ran; our contract draft was wrong, not the code
boyan1001/SFL-MTSC/self-consistency.py
code served (permissive licence) · get_code("b951e0aff025c2ea")
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 math
from collections import defaultdict, deque
from typing import Any, Dict, List, Set, Tuple

# Construct valid arguments for voting_key function
ARGS = {
    "cluster": [
        {
            "slots": {
                "entity_type": "person",
                "name": "Alice",
                "confidence": "0.95"
            }
        },
        {
            "slots": {
                "entity_type": "person",
                "name": "Alice",
                "confidence": "0.92"
            }
        },
        {
            "slots": {
                "entity_type": "organization",
                "name": "Bob",
                "confidence": "0.85"
            }
        }
    ]
}
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

Also present in other papers' code

Of 10 samples lifted from this paper's linked repositories, 10 carry a function identity and 0 do not. Of the 10 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 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.

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

Prompt-based spoken language understanding (SLU) with large language models (LLMs) often suffers from inconsistent intentslot structures due to decoding stochasticity, particularly in multi-intent scenarios. In view of this, we propose Semantic Frame-Level Multi-Task Self-Consistency (SFL-MTSC), a novel structured aggregation framework operating at the semantic frame level. Instead of output-level majority voting, SFL-MTSC decomposes predictions into intent-specific frames, applies domain-intent grouping and slot-level clustering, and evaluates cluster reliability using path support scoring. Reliable frames are retained and re-integrated to form the final prediction. Zero-shot experiments on the MAC-SLU benchmark dataset show improved slot F1 and overall accuracy over single-path inference, while intent accuracy remains largely stable across most settings.

For agents

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

get_harvested_code_for_paper("2606.25552")
get_code_for_paper("2606.25552")
have("2606.25552")

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

curl https://syntology.ai/api/ran/2606.25552.json

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

[![Syntology run record](https://syntology.ai/api/ran/2606.25552.svg)](https://syntology.ai/paper/2606.25552)

Connect an agent — have() is free.