Return to Answer

Fixed a few typos.

Source Link

edited May 6, 2021 at 21:38

masterpiga

As already mentioned by others, there are many biases at play here.

There is no such thing as "the sound of European languages", and most European languages sound nothing like English.
Similarly, CJK languages sound very different from each other.
The program that you wrote is not actually reading words aloud, it's just spelling.them. How they sound is a function of how the readers will map the sequence of characters onto phonemes. English pronunciation is by and large not deterministic given a sequence of characters in isolation, and pronunciation varies depending on a word's etymology, among other factors. İf you show English readers a sequence of characters that they cannot clearly map onto a similar word, İ would expect that their agreement concerning how it should be pronounced would be relatively low. Note that this wouldn't be the case for other languages, such as Turkish or Italian, in which the mapping from sequences.of characters to sequences of phonemes is pretty much deterministic.
From the way that you describe the algorithm, it does not seem to be designed to produce English-like.words words. This will make the words appear exotic to the readers, and so they will be tempted to say that they sound (read) like some exotic language. Far eastern languages are probably considered more exotic, by virtue of the fact that western European and Anglo-Saxon cultures have been blending at many levels for many centuries.
İf you want to generate words that sound like English, you should probably consider an algorithm that works (at the very least) at the syllable level, and that applies the appropriate morphological transformations at the boundaries between syllables.
Have you tried generating, say, 1000 words, then analyzing the distribution of syllables (or.bigrams character bigrams/trigrams, to keep things simpler) and comparing it with the distributions of the most frequent 1000 words of a bunch of languages?
This is completely speculative. İ know asa handful of Japanese words, and the sample is probably not very representative, but it seems to me that, when transliterated, it is.common common to see very regular words, with alternating vowels and consonants and frequent bowelvowel endings. Also, İ don't think that Japanese has any strong vowel.harmony rule harmony, so randomish vowel patterns may fly. İt may be possible, then, that your algorithm is producing words that are somewhat more similar to transliterated Japanese than they are to English words.

As already mentioned by others, there are many biases at play here.

There is no such thing as "the sound of European languages", and most European languages sound nothing like English.
Similarly, CJK languages sound very different from each other.
The program that you wrote is not actually reading words aloud, it's just spelling.them. How they sound is a function of how the readers will map the sequence of characters onto phonemes. English pronunciation is by and large not deterministic given a sequence of characters in isolation, and pronunciation varies depending on a word's etymology, among other factors. İf you show English readers a sequence of characters that they cannot clearly map onto a similar word, İ would expect that their agreement concerning how it should be pronounced would be relatively low. Note that this wouldn't be the case for other languages, such as Turkish or Italian, in which the mapping from sequences.of characters to sequences of phonemes is pretty much deterministic.
From the way that you describe the algorithm, it does not seem to be designed to produce English-like.words. This will make the words appear exotic to the readers, and so they will be tempted to say that they sound like some exotic language. Far eastern languages are probably considered more exotic, by virtue of the fact that western European and Anglo-Saxon cultures have been blending at many levels for many centuries.
İf you want to generate words that sound like English, you should probably consider an algorithm that works (at the very least) at the syllable level, and that applies the appropriate morphological transformations at the boundaries between syllables.
Have you tried generating, say, 1000 words, then analyzing the distribution of syllables (or.bigrams/trigrams, to keep things simpler) and comparing it with the distributions of the most frequent 1000 words of a bunch of languages?
This is completely speculative. İ know as handful of Japanese words, and the sample is probably not very representative, but it seems to me that, when transliterated, it is.common to see very regular words, with alternating vowels and consonants and frequent bowel endings. Also, İ don't think that Japanese has any strong vowel.harmony rule, so randomish vowel patterns may fly. İt may be possible, then, that your algorithm is producing words that are somewhat more similar to transliterated Japanese than English words.

As already mentioned by others, there are many biases at play here.

There is no such thing as "the sound of European languages", and most European languages sound nothing like English.
Similarly, CJK languages sound very different from each other.
The program that you wrote is not actually reading words aloud, it's just spelling.them. How they sound is a function of how the readers will map the sequence of characters onto phonemes. English pronunciation is by and large not deterministic given a sequence of characters in isolation, and pronunciation varies depending on a word's etymology, among other factors. İf you show English readers a sequence of characters that they cannot clearly map onto a similar word, İ would expect that their agreement concerning how it should be pronounced would be relatively low. Note that this wouldn't be the case for other languages, such as Turkish or Italian, in which the mapping from sequences.of characters to sequences of phonemes is pretty much deterministic.
From the way that you describe the algorithm, it does not seem to be designed to produce English-like words. This will make the words appear exotic to the readers, and so they will be tempted to say that they sound (read) like some exotic language. Far eastern languages are probably considered more exotic, by virtue of the fact that western European and Anglo-Saxon cultures have been blending at many levels for many centuries.
İf you want to generate words that sound like English, you should probably consider an algorithm that works (at the very least) at the syllable level, and that applies the appropriate morphological transformations at the boundaries between syllables.
Have you tried generating, say, 1000 words, then analyzing the distribution of syllables (or character bigrams/trigrams, to keep things simpler) and comparing it with the distributions of the most frequent 1000 words of a bunch of languages?
This is completely speculative. İ know a handful of Japanese words, and the sample is probably not very representative, but it seems to me that, when transliterated, it is common to see very regular words, with alternating vowels and consonants and frequent vowel endings. Also, İ don't think that Japanese has any strong vowel harmony, so randomish vowel patterns may fly. İt may be possible, then, that your algorithm is producing words that are somewhat more similar to transliterated Japanese than they are to English words.

Source Link

created May 6, 2021 at 21:32

masterpiga

As already mentioned by others, there are many biases at play here.

There is no such thing as "the sound of European languages", and most European languages sound nothing like English.
Similarly, CJK languages sound very different from each other.
The program that you wrote is not actually reading words aloud, it's just spelling.them. How they sound is a function of how the readers will map the sequence of characters onto phonemes. English pronunciation is by and large not deterministic given a sequence of characters in isolation, and pronunciation varies depending on a word's etymology, among other factors. İf you show English readers a sequence of characters that they cannot clearly map onto a similar word, İ would expect that their agreement concerning how it should be pronounced would be relatively low. Note that this wouldn't be the case for other languages, such as Turkish or Italian, in which the mapping from sequences.of characters to sequences of phonemes is pretty much deterministic.
From the way that you describe the algorithm, it does not seem to be designed to produce English-like.words. This will make the words appear exotic to the readers, and so they will be tempted to say that they sound like some exotic language. Far eastern languages are probably considered more exotic, by virtue of the fact that western European and Anglo-Saxon cultures have been blending at many levels for many centuries.
İf you want to generate words that sound like English, you should probably consider an algorithm that works (at the very least) at the syllable level, and that applies the appropriate morphological transformations at the boundaries between syllables.
Have you tried generating, say, 1000 words, then analyzing the distribution of syllables (or.bigrams/trigrams, to keep things simpler) and comparing it with the distributions of the most frequent 1000 words of a bunch of languages?
This is completely speculative. İ know as handful of Japanese words, and the sample is probably not very representative, but it seems to me that, when transliterated, it is.common to see very regular words, with alternating vowels and consonants and frequent bowel endings. Also, İ don't think that Japanese has any strong vowel.harmony rule, so randomish vowel patterns may fly. İt may be possible, then, that your algorithm is producing words that are somewhat more similar to transliterated Japanese than English words.