Comparison of Keystrokes Required for Vietnamese Phrase Input: Traditional Methods(Telex or VNI) vs. v7 Method.
Phrases are categorized based on the number of words inside.
Note: The statistics provided are derived under optimal conditions, assuming the user inputs only consonants and tones, and the target phrase is both in the dictionary and appear in the top 9 predictions.
| Phrase length | Number of phrases | Keystrokes reduced (± 1 std) |
Speed improvement (± 1 std) |
|---|---|---|---|
| 1 | 6170 | 0.42 ± 1.12 | 3.1% ± 28.5%* |
| 2 | 51614 | 4.74 ± 1.64 | 44.3% ± 9.9% |
| 3 | 6236 | 7.93 ± 2.09 | 49.0% ± 7.4% |
| 4+ (not in dictionary) |
9589 | 8.79 ± 2.42 | 37.5% ± 5.1% |
The statistics show that v7 demonstrates significant improvements in performance when typing phrases with multiple words from the dictionary, greatly reducing keystrokes and boosting typing speed. This makes it effective since in Vietnamese communication phrases are often used rather than isolated words.
*: When typing single words, users are required to input the entire word and tone, sometimes lead to a slight increase in keystrokes compared to traditional methods(this is the only main drawback to traditional method since if user wants to type anh in v7, they must type anh0 anh type one more number to choose from the prediction list).
In the context of the Vietnamese language, analyzing involves the mapping of a Vietnamese word to a tuple of (consonant_family, rhyme_family, tone). This process is essential for various language processing tasks. It's important to note that the analysis is non-injective, meaning that multiple Vietnamese words may map to the same tuple. For instance, both (cuốc, quốc) or (mi, my) can be mapped to the same tuple.
For further details on consonant families, rhyme families, and tones, refer to the vietnamese.py file.
Use Vietnamese.analyze(word: str) for this process.
This function is a great tool for detecting if a word is Vietnamese or not!
from utils.vietnamese import Vietnamese
# Sample corpus for analysis
corpus = "kiểm soát phối hợp tra cứu trực tiếp trên máy tính để phát hiện phương tiện vi phạm qua hình ảnh mà đã có thông báo chưa đến nộp phạt thì tổ tra cứu sẽ thông báo"
# Analyze each word in the corpus
for word in corpus.split(' '):
print(Vietnamese.analyze(word))Output (consonant family, rhyme family, tone)
('k', 'iêm', 3)
('s', 'oan', 6)
('ph', 'ôi', 1)
('h', 'ơm', 7)
('tr', 'a', 0)
('k', 'ưu', 1)
('tr', 'ưng', 7)
('t', 'iêm', 6)
('tr', 'ên', 0)
('m', 'ay', 1)
('t', 'inh', 1)
('đ', 'ê', 3)
('ph', 'an', 6)
('h', 'iên', 5)
('ph', 'ương', 0)
('t', 'iên', 5)
('v', 'i', 0)
('ph', 'am', 5)
('k', 'oa', 0)
('h', 'inh', 2)
('0', 'anh', 3)
('m', 'a', 2)
('đ', 'a', 4)
('k', 'o', 1)
('th', 'ông', 0)
('b', 'ao', 1)
('ch', 'ưa', 0)
('đ', 'ên', 1)
('n', 'ôm', 7)
('ph', 'an', 7)
('th', 'i', 2)
('t', 'ô', 3)
('tr', 'a', 0)
('k', 'ưu', 1)
('s', 'e', 4)
('th', 'ông', 0)
('b', 'ao', 1)
Explanation
The Vietnamese language actually encompasses 8 tones Vietnamese Eight-Tone Analysis, contrary to the commonly known 6 tones. The number 6 refers to the number of diacritics used (which include none (a), acute (á), grave (à), hook (ả), tilde (ã), underdot (ạ)). Additionally, there are two extra tones in Vietnamese for syllables ending in /p/, /t/, /c/, and /ch/.
Example words for the seventh tone include: xuất, cấp, tất, chiếc, thích, mút... (with rhyme families being uân, âm, ân, iêng, inh, un respectively).
Example words for the eighth tone include: nhập, phục, đột, chục, mạch, kịp... (with rhyme families being âm, ung, ôn, ung, anh, im respectively).
Note: This is the reason why the project name is v7: Việt with the 8th tone (count from index 0 is 7).
In contrast to analyzing, synthesizing is the process of mapping a tuple of (consonant_family, rhyme_family, tone) to a list of words(due to the non-injective property mentioned above), it's worth noting that in most cases, the resulting list typically contains just one word.
Use Vietnamese.synthesize(consonant: str, rhyme: str, tone: int) or you can use Dictionary.db[consonant][rhyme][tone] for this process. Both methods yield the same result, but the second method is generally faster as it directly retrieves the keys.
from utils.vietnamese import Vietnamese
from utils.dictionary import Dictionary
print(Vietnamese.synthesize('k', 'u', 4))
print(Vietnamese.synthesize('k', 'uông', 6))
print(Vietnamese.synthesize('0', 'iên', 1))
print(Vietnamese.synthesize('0', 'i', 1))
print(Vietnamese.synthesize('z', 'i', 2))
print(Vietnamese.synthesize('ng', 'iêm', 7))
print(Vietnamese.synthesize('ng', 'iêm', 5))
print(Dictionary.db['z']['iên'][6])
print(Dictionary.db['d']['iên'][7])
print(Dictionary.db['k']['oeo'][5])
print(Dictionary.db['k']['oang'][0])
print(Dictionary.db['g']['i'][2])
print(Dictionary.db['z']['ăng'][7])
print(Dictionary.db['g']['i'][2])
print(Dictionary.db['kh']['ung'][3])
print(Dictionary.db['ng']['ênh'][2])
print(Dictionary.db['h']['oang'][3])
print(Dictionary.db['x']['uân'][6])
print(Dictionary.db['kh']['uênh'][2])
print(Dictionary.db['kh']['uênh'][7])
print(Dictionary.db['kh']['uênh'][6])
print(Dictionary.db['k']['oai'][1])
print(Dictionary.db['k']['ưu'][3])
# for c in Vietnamese.consonant_families:
# print(Dictionary.db[c]['i'][0])(Uncomment the last 2 lines to see the differences between different consonants go with i)
Output
['cuốc', 'quốc']
['yến']
['í', 'ý']
['gì']
['nghiệp']
['nghiệm']
['giết']
['diệt']
['quẹo']
['quang']
['ghì']
['giặc']
['ghì']
['khủng']
['nghềnh']
['hoảng']
['xuất']
['khuềnh']
['khuệch']
['khuếch']
['quái']
['cửu']
This is the process showcased in the demo GIF above. If you prefer testing via a Python script, you can use the following code.
from imethod.v7 import InputMethod # Dictionary mode
# from imethod.v7ai import AIInputMethod # AI mode
inputAgent = InputMethod()
print(inputAgent.predict('xi0chao2mo5ng2'))
print(inputAgent.predict('xi0chao2mo')) # Not completed
print(inputAgent.predict('ximg0ch2')) # No match (`ximg`)
print(inputAgent.predict('xi0')) # Exact match when predict just one word and that word is provided with rhyme
print(inputAgent.predict('b7')) # Wildcard rhyme -> All possibilities in descending order of frequency (based on a large corpus)
print(inputAgent.predict('b7t2'))
print(inputAgent.predict('ba7ti2'))
print(inputAgent.predict('bac7ti2'))Output
['xin chào mộ người', 'xin chào mộ ngày', 'xin chào mọi người', 'xin chào mọi ngày', 'xinh chào mộ người', 'xinh chào mộ ngày', 'xinh chào mọi người', 'xinh chào mọi ngày']
None
None
['xi']
['biệt', 'bạc', 'bật', 'buộc', 'bạch', 'bậc', 'bột', 'bẹp', 'bọc', 'bịt', 'bọt', 'bạt', 'bực', 'bộc', 'bịp', 'bục', 'bịch', 'bệt', 'bụt', 'bặt', 'bập', 'bụp', 'bệch', 'buột', 'bẹc']
['biệt tài', 'bạc tình', 'bật tường', 'bạch tiền']
['bạc tình', 'bạch tiền']
['bạc tình']
...and many more in sandbox.
Vietnamese Eight-Tone Analysis
Date Created: 10:05 AM, Tue 27 Feb 2024
Data sources:
- Dictionary and Frequency
- Training v7gpt: