Performance of Human Languages, Benchmarked: Speed, Symbols, Error Correction, Latency, Install Time

This is the languages episode of “A Programmer's View”* Which human language is the fastest? Japanese is spoken at about eight syllables a second, Mandarin Chinese at under six, yet both carry about 40 bits of information a second. This film benchmarks 12 widely spoken languages the way programmers benchmark machines: English, Mandarin, Cantonese, Hindi, Spanish, Arabic, French, Russian, German, Japanese, Korean and Thai. There are 5 tests. Throughput: speech rate against information rate, in bits per second. Symbol set: how many distinct sounds (phonemes) a language builds with, from about 20 in Japanese to more than 50 in Hindi, plus tone in Mandarin, Cantonese and Thai. Error correction: what keeps "dog bites man" from turning into "man bites dog" when part of a message is lost, by word order in English and Chinese, by case endings in Russian, and by the redundancy Claude Shannon measured in written English. Latency: where the verb comes in a sentence, and how long a listener waits for it. Install time: how many weeks of class it takes to learn each language, according to the U.S. Foreign Service Institute, and why that says as much about the learner as about the language. Some cells in the report are blank, because nobody has measured them. And every test turns out the same way: what a language saves in one place, it pays for in another. Along the way Speech rate and information rate: syllables per second against bits per second, from a 2019 study of 17 languages (about 39 bits per second on average) Why a fast language is not a faster one: a Japanese syllable carries about 5 bits, a Mandarin syllable about 7 Thai as the slowest speech rate in the study, French as the highest information rate How many sounds a language has: phoneme counts for Japanese, Spanish, English, French, German, Russian and Hindi, and why the counts differ between descriptions Tone languages: Mandarin, Cantonese and Thai How many syllables a language has: a few hundred kinds in Japanese, almost 7,000 in English Why a computer needs eight bits to store one letter Word order against grammatical case: English and Mandarin go by position, Russian by case endings; German, Japanese and Korean mark case too Positional and named arguments in programming, as a way to see the same difference Grammatical agreement as redundancy: Spanish marks one plural three times Claude Shannon, 1948: about half of written English is redundancy; information theory, noise and compression A 2023 experiment: with the grammar stripped away, English and Russian speakers still picked the subject almost nine times out of ten Word order across the world's languages: SVO, SOV and VSO; verb-final Hindi, Japanese and Korean; verb-first Modern Standard Arabic; 1,376 languages in WALS How long it takes an English speaker to learn a language: the FSI estimates of 30 weeks for Spanish and French, 36 for German, 44 for Hindi, Russian and Thai, and 88 for Arabic, Mandarin, Cantonese, Japanese and Korean More than 7,000 living languages, of which this report covers twelve Chapters 0:00 Japanese and Mandarin: which carries more? 0:50 A benchmark report: twelve languages, five tests 1:17 Test 1: throughput 3:15 Test 2: the symbol set 4:22 Test 3: error correction 5:42 Saying it three times 6:07 Shannon and the spare half of English 7:22 Test 4: latency 8:23 Test 5: install time 9:29 The report Sources C. Coupé, Y. M. Oh, D. Dediu, F. Pellegrino, "Different languages, similar encoding efficiency" (Science Advances, 2019), data and code: https://github.com/keruiduo/SupplMatI... CNRS press release on the same study: https://www.cnrs.fr/en/press/similar-... PHOIBLE 2.0, sound inventories: https://phoible.org The World Atlas of Language Structures Online (word order, case, tone): https://wals.info C. E. Shannon, "A Mathematical Theory of Communication" (1948): https://people.math.harvard.edu/~ctm/... K. Mahowald et al., "Grammatical cues to subjecthood are redundant in a majority of simple clauses across languages" (Cognition, 2023): https://arxiv.org/pdf/2201.12911v3 E. Gibson et al., "How Efficiency Shapes Human Language" (Trends in Cognitive Sciences, 2019): https://tedlab.mit.edu/tedlab_website... Real Academia Española, Diccionario panhispánico de dudas, "plural": https://www.rae.es/dpd/plural U.S. Department of State, Foreign Service Institute, "Foreign Language Training": https://www.state.gov/foreign-languag... Ethnologue, 27th edition: https://ethnologue.com/ethnoblog/welc...