Word-initial consonants are systematically lengthened across diverse languages | Hacker News Reader

Word-initial consonants are systematically lengthened across diverse languages

phys.org

28 pointswglb7 comments

Neat. I always assumed that you just draw the boundaries based on recognizing words you already know.

Full paper: https://www.nature.com/articles/s41562-024-01988-4

Another fun tidbit:

> Regarding utterance-initial positions, no language in our sample shows evidence in favour of lengthening. However, 15 languages show evidence for utterance-initial shortening.

As a Finnish speaker I've often wondered about this. In Finnish (and Hungarian) the stress is almost always on the first syllable, so it is easy for speakers to distinguish word boundaries. How do speakers of other languages do it?

Now I'm wondering if the first consonant is lengthened in Finnish too or if the syllabic stress makes it unnecessary.

About half of English words are one syllable long and most of the rest are two. If it's stressed, it's probably a word.

Stress placement on disyllabic words works at a grammatical level, not just a phonetic one. "To record" and "a record" (reCORD, REcord) have different stress placement because of their grammatical roles (noun/verb).

Japanese: there's no stress, all syllables are the same length, except there are explicitly lengthened vowels which are semantically different and can occur almost in any position in a word. A syllable ends with a vowel or with "n", and there are no consonant clusters at the beginning of a syllable. There can be "long" consonants but they are always on the split between syllables, always inside a word, and actually make recognizing the boundaries easier. Usually it's easy enough to just listen to the sequence of syllables and cut it into probable words. Regular grammatical markers (particles that represent grammatical roles, endings of verbs and adjectives, which are also basically verbs) usually give enough structure to split ambiguous combinations correctly. (Not a native speaker.)

Spanish: most words are reasonably short, grammatical features like the articles and the endings of nouns / adjectives / verbs intersperse the speech all the time and give you clues where a word may begin, and what kind of word. The stress is not very regular but the very irregularities have some internal patterns. You only need a pretty short syllable buffer before a word can be recognized unambiguously, and the rest of it mostly predicted, except maybe for the ending, but you know which kind of ending it may be. Even if you don't know a word, it's easy to see its boundaries and its grammatical role. If the words were as long and grammatically varied as Finnish, this could become much harder! (Not a native speaker.)

English has a lower contrast between vowels than Spanish, it's easier to mishear a word, and harder to recognize it in slightly distorted forms, like singing. Again, the articles and, to some extent, the verb endings, help recognize some boundaries, but less efficiently. Many of the most frequent words only have one or two syllables though (like "blue" or "open"). Most really long words are recognizable as a single word long before they end (like "discombobulated"), they rarely sound like shorter unrelated words glued together. The sheer amount of daily training as a speaker, and predictability of most words in a sentence, still allow to speak and understand English at a very high speed. Intonation sometimes helps, too.

The thing that immediately jumps to mind is that the first syllable is more likely to be stressed, and stressed syllables are usually pronounced longer.

From the paper:

> Our corpus data are not annotated for suprasegmental features such as stress or tone. However, on the basis of available phonological descriptions, only 4 of the 51 languages can with some certainty be considered to have fixed initial word stress, while most other languages are either tone languages or stress languages with non-initial stress

It sounds like they account for languages that always have initial stress, but not for languages that simply have mostly initial stress or at least some bias towards it, e.g. German. There's also no mention of secondary stress, which plays a big role in English for example. It also sounds like they're assuming that if a language is tonal, then there must be no stress-like suprasegmental features which can affect syllable duration, which is untrue.

I'm inclined to believe the results anyway and it's a really interesting finding, but they would need to redo the study fully controlling for stress to get meaningful results in my opinion.

Vocalists, especially choral singers, will learn all sorts of techniques to put syllables to a particular rhythm. Some languages (Hebrew, Latin, Sanskrit) lend themselves to rhythmic chanting and singing, with generously interspersed vowels. You can also hear bad singing, with extended sssibilantsss... or multiple pops, as each singer finishes a "t" or "p" out of sync.

It's interesting to think about lengthening of different types of consonants, such as glottal stops! But mentally, if you put two t's together, it makes sense to hold them longer?

The linguistic principle of sandhi means that when we verbalize words, the adjoining sounds can coalesce, so sometimes if you're between two words, it would make sense that the sound is held for longer as partly a remnant of the first word, running into the second.

How do you lengthen a consonant? My imagination fails me.

Reply on news.ycombinator.com