SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2603.05690 · 2026

FreeTxt-Vi: A Benchmarked Vietnamese-English Toolkit for Segmentation, Sentiment, and Summarisation

Paul Rayson, Dawn Knight, Mo El-Haj, Hung Huy

arXiv · PDF · Open in the Atlas

Code that ran

4 of 4 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 3 have no contract verdict; 1 failure 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/2603.05690.json · badge: /api/ran/2603.05690.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
VinNLP/Free-txt-vi canonical (classifier label)
classifier basis: census
4 ran · 0 no recorded run
FunctionStatusWhere it lives
clean_word Ran
ran; our contract draft was wrong, not the code
VinNLP/Free-txt-vi/api/internal/services/word_cloud.py
code served (permissive licence) · get_code("0af5ea09c17e7992")
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 coroutine. 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
import asyncio

async def clean_word(word: str) -> str:
    """
    Clean a word by removing apostrophes, quotes, underscores, and other unwanted characters.
    """
    word = re.sub("[\\'′`´]", '', word)
    word = re.sub('[""″]', '', word)
    word = re.sub('[‛‟]', '', word)
    word = re.sub('[‹›]', '', word)
    word = re.sub('[«»]', '', word)
    word = re.sub('[„"]', '', word)
    word = re.sub('[\\u2018\\u2019]', '', word)
    word = word.replace('_', ' ')
    word = re.sub('\\s+', ' ', word).strip()
    return word

ARGS = {
    "word": "hello'world_test"
}
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 coroutine — a coroutine output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a coroutine 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
process_text_with_vncorenlp Ran
ran on a synthesized input
VinNLP/Free-txt-vi/api/internal/services/vncorenlp_singleton.py
code served (permissive licence) · get_code("262e3ede6eee0660")
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 coroutine. 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
import gc
import asyncio

# Stand-in for vncorenlp_model that has the word_segment method
class MockVnCoreNLPModel:
    def word_segment(self, text: str) -> list[str]:
        """Mock word segmentation - just splits by spaces"""
        return text.split()

# Create the mock model instance
vncorenlp_model = MockVnCoreNLPModel()

# Define arguments for process_text_with_vncorenlp
ARGS = {
    "text": "This is a test sentence. Here is another one! And a third sentence with some more words to make it longer.",
    "max_words": 200
}
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 coroutine — a coroutine output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a coroutine 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
process_text_with_vncorenlp_safe Ran
ran on a synthesized input
VinNLP/Free-txt-vi/api/internal/services/vncorenlp_singleton.py
code served (permissive licence) · get_code("55cf5fe952b057b7")
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 coroutine. 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
import gc
import asyncio

# Mock VnCoreNLP model
class MockVnCoreNLPModel:
    def word_segment(self, text):
        """Mock word_segment that returns list of words"""
        if not text or not text.strip():
            return []
        return text.split()

# Create a global mock model instance
vncorenlp_model = MockVnCoreNLPModel()

# Test text samples
short_text = "Hello world. This is a test."
medium_text = "This is a medium length text. " * 20
long_text = "This is a word. " * 500
very_long_text = "This is a test sentence with multiple words. " * 300

ARGS = {
    "text": "The quick brown fox jumps over the lazy dog. This is a test sentence for VnCoreNLP processing. It contains multiple sentences separated by periods. Each sentence will be processed and segmented into words. This is designed to test the chunking mechanism for large texts.",
    "max_words": 200,
}
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 coroutine — a coroutine output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a coroutine 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
split_text_for_vncorenlp Ran
ran on a synthesized input
VinNLP/Free-txt-vi/api/internal/services/vncorenlp_singleton.py
code served (permissive licence) · get_code("af5854bb005e1f2d")
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 coroutine. 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

async def split_text_for_vncorenlp(text: str, max_words: int=200) -> list[str]:
    """
    Split text into chunks that can be processed by VnCoreNLP.
    Each chunk will contain approximately max_words words.
    Reduced from 500 to 200 to avoid Java heap space issues.

    Args:
        text: Input text to split
        max_words: Maximum number of words per chunk (default 200 for VnCoreNLP)

    Returns:
        List of text chunks
    """
    sentences = re.split('[.!?]+', text)
    chunks = []
    current_chunk = []
    current_word_count = 0
    for sentence in sentences:
        sentence = sentence.strip()
        if not sentence:
            continue
        sentence_words = len(sentence.split())
        if sentence_words > max_words:
            sub_sentences = re.split('[,;:]', sentence)
            for sub_sentence in sub_sentences:
                sub_sentence = sub_sentence.strip()
                if not sub_sentence:
                    continue
                sub_words = len(sub_sentence.split())
                if sub_words > max_words:
                    words = sub_sentence.split()
                    for i in range(0, len(words), max_words):
                        chunk_words = words[i:i + max_words]
                        chunks.append(' '.join(chunk_words))
                elif current_word_count + sub_words > max_words and current_chunk:
                    chunks.append(' '.join(current_chunk))
                    current_chunk = [sub_sentence]
                    current_word_count = sub_words
                else:
                    current_chunk.append(sub_sentence)
                    current_word_count += sub_words
        elif current_word_count + sentence_words > max_words and current_chunk:
            chunks.append(' '.join(current_chunk))
            current_chunk = [sentence]
            current_word_count = sentence_words
        else:
            current_chunk.append(sentence)
            current_word_count += sentence_words
    if current_chunk:
        chunks.append(' '.join(current_chunk))
    return chunks

# Setup code
sample_text = "This is the first sentence. This is the second sentence! And here is a third one? Let me add some more text to make it longer. " * 5

ARGS = {
    "text": sample_text,
    "max_words": 200,
}
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 coroutine — a coroutine output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a coroutine 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 4 samples lifted from this paper's linked repositories, 4 carry a function identity and 0 do not. Of the 4 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 4 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

FreeTxt-Vi is a free and open-source web-based toolkit for creating and analysing bilingual Vietnamese-English text collections. Positioned at the intersection of corpus linguistics and natural language processing (NLP), it enables users to build, explore, and interpret free-text data without requiring programming expertise. The system combines established corpus analysis features such as concordancing, keyword analysis, word relation exploration, and interactive visualisation with modern transformer-based NLP components for sentiment analysis and summarisation. A key contribution of this work is the design of a unified bilingual NLP pipeline that integrates a hybrid VnCoreNLP + Byte Pair Encoding (BPE) segmentation strategy, a fine-tuned TabularisAI sentiment classifier, and a fine-tuned Qwen2.5 model for abstractive summarisation. Unlike existing text analysis platforms, FreeTxt-Vi is evaluated as a set of language processing components. We conduct a three-part evaluation covering segmentation, sentiment analysis, and summarisation, and demonstrate that our approach achieves competitive or superior performance compared to widely used baselines in both Vietnamese and English. By reducing technical barriers to multilingual text analysis, FreeTxt-Vi supports reproducible research and promotes the development of language resources for Vietnamese, a widely spoken but underrepresented language in NLP. The toolkit is applicable to a wide range of domains, including education, digital humanities, cultural heritage, and the social sciences, where qualitative text data are common but often difficult to process at scale.

For agents

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

get_harvested_code_for_paper("2603.05690")
get_code_for_paper("2603.05690")
have("2603.05690")

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

curl https://syntology.ai/api/ran/2603.05690.json

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

[![Syntology run record](https://syntology.ai/api/ran/2603.05690.svg)](https://syntology.ai/paper/2603.05690)

Connect an agent — have() is free.