gabrielchua/singlish-to-english-synthetic

---
license: cc-by-nc-sa-4.0
task_categories:
- translation
language:
- en
pretty_name: Singlish to English 🇸🇬
size_categories:
- n<1K
---

# Singlish to English 🇸🇬

> Singapore is known for its efficiency and Singlish is no different - it's colourful and snappy. - [Tessa Wong, BBC News, 2015](https://www.bbc.com/news/magazine-33809914)

This is a synthetic dataset generated by GPT-4.

Each json pair contains one Singlish sentence about an everyday activity (e.g. cooking) and its English translation.

# Sample entry
```json
singlish: "Eh, chop the garlic - you can a not?",
english: Hey, do you know how to chop the garlic?"
```

# Data Generation Code

```python
import json
import pandas as pd
from openai import OpenAI

client = OpenAI()

NUM_SAMPLE = 10
ACTIVITIES = ['cooking',
'studying',
'sleeping',
'eating',
'working',
'exercising',
'reading',
'cleaning',
'shopping',
'driving',
'walking',
'bathing',
'going to work',
'listening to music',
'watching TV',
'playing video games',
'using a computer',
'texting',
'socializing',
'meditating',
'commuting',
'doing laundry',
'ironing clothes',
'dusting',
'vacuuming',
'painting',
'drawing',
'grocery shopping',
'sewing',
'taking a nap',
'jogging',
'biking',
'swimming',
'playing sports',
'checking emails',
'playing with children',
'watching movies',
'playing board games',
'attending school or classes',
'going to the gym',
'playing a musical instrument',
'singing',
'dancing',
'writing',
'photography',
'traveling',
'visiting friends',
'attending events',
'volunteering',
'attending meetings']

dataset = {}

for index, activity in enumerate(ACTIVITIES):
print(index, activity)
response = client.chat.completions.create(
model="gpt-4-1106-preview",
messages=[{"role": "system",
"content": "You are an expert in translating Singlish to English"},
{"role": "user",
"content": f"Create {NUM_SAMPLE} random Singlish (s) to English (e) translation pairs in json. Write full sentences about {activity}."\
f"Don't exaggerate the use of Singlish, and be natural, as how a real Singaporean would speak."\
f"Start the keys from {(index*NUM_SAMPLE)+1}. For example,"\
"{'X':{'s': 'aiyo, why like that', 'e': 'oh my, how did this happen'}"\
"..., 'X+5': {'s': 'don't play play', 'e': 'don't fool around'} }"}],
temperature=0.01,
response_format={"type":"json_object"}
)
output = response.choices[0].message.content
output_json = json.loads(output)
dataset.update(output_json)

# Save the current state of the combined dictionary
with open('singlish_to_english_v0.1.json', 'w') as f:
json.dump(dataset, f, indent=None)

# Convert to tabular csv
df = pd.read_json("singlish_to_english_v0.1.json")
df = df.T
df = df.reset_index()
df.columns = ["index", "singlish", "english"]
df.to_csv("singlish_to_english_v0.1.csv", index=False)
```

原始信息汇总

Singlish to English 🇸🇬

数据集概述

• 许可证: cc-by-nc-sa-4.0
• 任务类别: translation
• 语言: en
• 数据集大小: n<1K
• 数据集名称: Singlish to English 🇸🇬

数据集描述

该数据集是由GPT-4生成的合成数据集。每个JSON对包含一个关于日常活动的Singlish句子及其英语翻译。

样本条目

json { "singlish": "Eh, chop the garlic - you can a not?", "english": "Hey, do you know how to chop the garlic?" }

数据生成代码

数据集的生成代码如下：

python import json import pandas as pd from openai import OpenAI

client = OpenAI()

NUM_SAMPLE = 10 ACTIVITIES = [cooking, studying, sleeping, eating, working, exercising, reading, cleaning, shopping, driving, walking, bathing, going to work, listening to music, watching TV, playing video games, using a computer, texting, socializing, meditating, commuting, doing laundry, ironing clothes, dusting, vacuuming, painting, drawing, grocery shopping, sewing, taking a nap, jogging, biking, swimming, playing sports, checking emails, playing with children, watching movies, playing board games, attending school or classes, going to the gym, playing a musical instrument, singing, dancing, writing, photography, traveling, visiting friends, attending events, volunteering, attending meetings]

dataset = {}

for index, activity in enumerate(ACTIVITIES): print(index, activity) response = client.chat.completions.create( model="gpt-4-1106-preview", messages=[{"role": "system", "content": "You are an expert in translating Singlish to English"}, {"role": "user", "content": f"Create {NUM_SAMPLE} random Singlish (s) to English (e) translation pairs in json. Write full sentences about {activity}." f"Dont exaggerate the use of Singlish, and be natural, as how a real Singaporean would speak." f"Start the keys from {(index*NUM_SAMPLE)+1}. For example," "{X:{s: aiyo, why like that, e: oh my, how did this happen}" "..., X+5: {s: dont play play, e: dont fool around} }"}], temperature=0.01, response_format={"type":"json_object"} ) output = response.choices[0].message.content output_json = json.loads(output) dataset.update(output_json)

# Save the current state of the combined dictionary
with open(singlish_to_english_v0.1.json, w) as f:
    json.dump(dataset, f, indent=None)

Convert to tabular csv

df = pd.read_json("singlish_to_english_v0.1.json") df = df.T df = df.reset_index() df.columns = ["index", "singlish", "english"] df.to_csv("singlish_to_english_v0.1.csv", index=False)