At the EPSJ conference Takahiro Ioroi presented some statistics about the relative frequency of lexical stress patterns in English words. The pedagogical point was to investigate how far L2 English learners are “exposed to attested patterns in the inputs available”.
Ioroi did this by combining data on stress placement from the Carnegie Mellon University Pronouncing Dictionary with word frequency data from the British National Corpus and a word list from a collection of EFL textbooks for Japanese schools.
In this way he demonstrated that the most frequent exemplar of an initial-stressed disyllable in the school textbooks was people (at 3606 per million), followed by very, other, many and our (sic), while in the BNC it was other (1336 per million) followed by only, also, people and any.
The methodology was irreproachable. But some of Ioroi’s findings demonstrate the truth of the old adage “garbage in, garbage out”.
Let’s not quibble about our (which NSs usually pronounce as a monosyllable).
What about disyllables with final stress? The most frequent one in the textbook corpus was about, which is fair enough. But the most frequent one in the BNC, and second most frequent in the textbooks, comes out as into.
Into? But into is stressed on the first syllable, ˈɪntu, ˈɪntə. It does not have final stress. CMUPD says it does: IH0 N T UW1, which is how they represent ɪnˈtuː. CMUPD is wrong, wrong, wrong.
(In running speech, which is not under consideration here, into may of course lose all stress.)
A useful generalization about English words is that all polysyllables have a primary or secondary lexical stress on either the first or the second syllable. So revolution, for example, has the main stress on the penultimate but on the initial syllable a secondary stress: ˌrevəˈluːʃən. Having supplied stress patterns for several complete dictionary headword lists, I can say that the only exceptions I am aware of are (for some speakers) the two unusual words peradventure and forasmuch. Although they are written as single words, some speakers pronounce them pərədˈventʃə, fərəzˈmʌtʃ, as if they were prepositional phrases, like for a change fərəˈtʃeɪndʒ. (Alternatively, they can be ˌpɜːrədˈventʃə, ˌfɔːrəzˈmʌtʃ, which fit the rule.)
What do Ioroi’s stats tell us about such polysyllables? The most frequent BNC words with lexical stress on neither of the first two syllables are purportedly insufficient and valuation. Again, I am afraid, CMUPD is to blame for supplying wrong information, having forgotten to show secondary stress on the initial syllable of each. (But CMUPD does get the stress pattern of revolution correct.)
For the textbook corpus the results are even odder, since the most frequent such words come out as various proper names, mostly Japanese: Sugihara, Nakamura, Yamagata, Morimoto, Antonelli, which CMUPD shows as having stress only on the penultimate. The fact is that these, too, have initial secondary stress. The incontrovertible evidence for this is the ‘stress shift’ effect when followed by another accented word: ˌSugihara’s ˈwidow (found in this passage).
Again, CMUPD is wrong. GIGO.