Language Log

Like others, I've often noted  analogies between prosody (as modulations of pitch, voice quality, timing, and so on) and text rendering, whether in calligraphy or typography — e.g. "Intonational focus", 4/29/2011; "Prosodic lettering", 5/8/2011; and many other posts about the communicative use of color, font choice, spatial placement, punctuation, and so on. Some aspects of textual prosody are perceptually natural, like size and spatial separation, while others are conventional, like the use of font choices in dictionary entries. And the conventions change over time and space, like capitalization in English.

Attempts by style guides to lock this variation down are roughly as effective as other efforts to limit individual and cultural creativity, and the growth of social media opens up new horizons for orthographic sociolinguistics.

Marc Andreessen's recent appearance on Joe Rogan's podcast presented a striking example of AI promotion (or AI hype, as you please). We can discuss his extraordinary claims and predictions another time. My topic this morning is something Andreessen does that AI still can't do, namely talk like a human being. I'm referring to the way that humans talk in spontaneous conversation, not in fluent reading or in well-rehearsed presentations, which AI text-to-speech can imitate increasingly well.

Or maybe I should say, "AI prosody"?

In a series of posts over the past year, I've suggested that evaluation of reading performance ought to go beyond the question of whether individual words are correctly decoded and pronounced.

In "Analysis of prosodic timing in reading" (4/5/2026), I suggested that inter-word timings in fluent reading can give a surprisingly clear picture of prosodic phrasing, despite the many other effects on word durations in speech.

That post looked at data from the Speech Accent Archive, which involves reading a short and somewhat weird passage. Since then, I've explored readings of a variety of other texts, so far only in English. The results continue to look promising.

This post documents one small step in a larger plan for improved evaluation of prosody in reading. It compares word-level timing in a large number of recordings, from the Speech Accent Archive at GMU, of 3038 people reading the 69-word "Please call Stella" passage. 661 of these people are native speakers of English, with accents from all over the anglophone world, while the remaining 2377 readers have native languages from Afrikaans to Zulu. The reading and speaking level of those non-native readers varies a lot, and many of them have problems in decoding or pronunciation that affect their timing.

Automated analysis of such problems should be useful in foreign-language teaching. And similar analyses might help in early reading instruction for students in anglophone classrooms, whatever their native language.

Let's start with a quick comparison of word-level timing in the 661 native English speakers; the 85 native French speakers; the 99 native Korean speakers; and the 82 native russian speakers.

I calculated word-level time points for those 927 speakers, using a forced-alignment system originally developed many years ago with Jiahong Yuan — a summary of the technology and a few of its application can be found here (open-access version). Here's the output for speaker english1 — note that the segment ID sp means "silent pause".

The key conclusions:

1. Time between word onsets gives a good picture of phrase structure, despite the many other effects on timing;
2. Individual non-native readers, aside from being overall a bit slower, usually show lengthened inter-word intervals in unexpected places, due to decoding or pronunciation problems.

Over the years, I've posted several times about the problematic word (and concept) disfluency — there's a too-long list at the end of "Spontaneous (dis)fluency" (8/27/2025). Among other ideas, I've suggested using the term interpolations (see this 2019 post for example). But as far as I can tell, this suggestion has had no impact on other people's usage.

So here's another try: How about making a list of the ways in which fluent spontaneous speech is not like fluent reading, and calling them all spontaneities?

In "Reading Instruction in the mid 19th century" (8/15/2025), I noted a suggestion, due to Ran Liu of Amira Learning, that a computational analysis of prosodic features could be an effective way to evaluate how well grade-school students understand what they're reading. Beyond that, Maryellen MacDonald has suggested that phrasal prosody can be seen as the phase-level analog of phonemic blending (i.e. putting the sounds of 'c' 'a' 't' together into "cat") — which might help to explain the benefits of McGuffey-style elocution lessons.

Both ideas raise the question of how to evaluate the prosody of a given student's reading. And there's a simple and obvious way to do this, described and exemplified below.

Several people have asked me about this paper — Maya Inbar, Eitan Grossman, and Ayelet Landau, "A universal of speech timing: Intonation units form low-frequency rhythms", PNAS 8/19/2025:

Intonation units (IUs) are a hypothesized universal building block of human speech [W. Chafe, Discourse, Consciousness and Time: The Flow and Displacement of Conscious Experience in Speaking and Writing (1994); N. P. Himmelmann et al., Phonology 35, 207–245 (2018)). Linguistic research suggests they are found across languages and that they fulfill important communicative functions such as the pacing of ideas in discourse and swift turn-taking. We study the rate of IUs in 48 languages from every continent and from 27 distinct language families. Using an analytic method to annotate natural speech recordings, we identify a low-frequency rate of IUs across the sample, with a peak at 0.6 Hz, and little variation between sexes or across the life span. We find that IU rate is only weakly related to speech rate quantified at the syllable level, and crucially, that cross-linguistic variation in IU rate does not stem from cross-linguistic variation in syllable rate.

Note that 0.6 Hz is 1/0.6 ≈ 1.7 seconds in the time domain.

In "Reading Instruction in the mid 19th century" (8/16/2025), I underlined the old-fashioned focus on "elocution", in which readers were trained "to convey to the hearer, fully and clearly, the ideas and feelings of the writer". Much of today's reading instruction turns that into measures of "oral reading fluency", measured as words correct per minute ("wcpm"). This can result in high-scoring readers like those described in this passage from the Introductory Remarks in 1844 edition of McGuffey's Rhetorical Guide, which warns against the consequences of failure to teach "elocution" from the very start:

In "AI win of the week" I explored the inter-personal dimensions of Rousseau's 1754 contention that "there is neither rhythm nor melody in French music, because the language is not capable of them". In the comments, AntC objected that "But, but. Rousseau wrote an opera, in French, to his own Libretto. audio + full score available on Youtube".

For now, I have only two comments on this. First, trolls are often happy to abandon consistency in the service of pwning their audience. And second, the 1754 edition of Rousseau's screed, published two years after the debut of his opera, goes into considerable detail about how he painfully transferred the musicality of Italian prosody to the composition and performance of a work with French lyrics.

But rather than diving further into Rousseau's argument about the relative musicality of different languages' prosody, the point of today's post is to note its resonance with another mid-18th century prosodic dispute, namely Joshua Steele's refutation of James Burnett's claim that English prosody gives its syllables "nothing better than the music of a drum, in which we perceive no difference except that of louder or softer, according as the instrument is more or less forcibly struck".

"Shakespearean focus" featured a 2016 skit in which different actors promoted different focus choices in Hamlet's famous line:

PAAPA ESSIEDU: «To be or not to be that is the question»…
TIM MINCHIN: Right sorry, sorry. I mean – y-yes… I’m – yes – but if you don’t mind a note: O-OR: «To be OR not to be.»
BENEDICT CUMBERBATCH: Try this: «To be or NOT to be, that is the question.»
DAVID TENNANT: Calm down, right. It’s simple. Don’t lose focus: «To be or not to be, THAT is the question.»
RORY KINNEAR: No, no, no, no, no. Idiots! «To be or not to be, that IS the question.»
SIR IAN MCKELLEN: Lend me your ears: «To be or not to be, that is THE question.»
DAME JUDI DENCH: «To be or not TO be.»
PRINCE CHARLES: «To be or not to be, that is the QUESTION.»

David Z. commented:

This reminds one (i.e., me) of the scene in WC Fields's film The Old Fashioned Way, about a troupe of actors, where benefactor Cleopatra Pepperday is promised a role, reciting the line "Here comes the prince." She practices the line many times, emphasizing different words, although she never actually gets to deliver it on stage.

Something for everyone

In comments on "Affected brogue", 12/19/2024, Benjamin Orsatti and others put Bernard Mayes forward as a quintessential RP speaker, including this advice:

[I]f you'd like to listen to, say, 150 consecutive hours of Bernard Mayes (the man narrates my dreams now), you can do what I'm doing and borrow "The Decline and Fall of the Roman Empire" audiobook, Blackstone edition from your local library.

This seemed like a good way to use one of the Audible credits that I've somehow accumulated, so I downloaded Mayes' narration of volume 3 of that work, all 39:03:05.09 of it. Listening to a few minutes of it, I was reminded of a joke that I (believe I) heard from Michael Studdert-Kennedy:

The archetypal Englishman, being forbidden by custom to wave his hands, waves his larynx instead.