Language Log

It's clear that text-to-speech programs have gotten better and better over the past 60 years, technical details aside. The best current systems rarely make phrasing or letter-to-sound mistakes, and generally produce speech that sounds pretty natural on a phrase-by-phrase basis. (Though there's a lot of variation in quality, with some shockingly bad systems in common use.)

But even the best current systems still act like they don't get George Carlin's point about "Rhetoric as music". Their problem is not that they can't produce verbal "music", but that they don't (even try to) understand the rhetorical structure of the text.  The biggest pain point is thus what linguists these days call "information structure", related also to what the Prague School linguistics called "communicative dynamism".

From Jon Stewart's 1997 interview with George Carlin (starting at about 1:17.6):

well- well uh to- to go backward with the question,
don't forget, what we do is oratory.
It's rhetoric.
It's not just comedy, it's a form of rhetoric
and- and with rhetoric, you- you look and you listen for rhythms,
you- you look for ways
to sing at the same time you're talking, and to go
[skat-like phrases, based on rhythmic patterns of /d/-initial syllables…]

The following is a guest post by Martin Ho Kwan Ip,  who is now a postdoc at Penn. See "Anne Cutler 1945-2022", 6/8/2022, for some background and links.

I am one of Anne's most recent students (her 44th student from the MARCS Institute in Australia). I met Anne for the first time in 2014 when she was invited to give a talk at the University of Queensland (we had been corresponding by email but had never met until then). Although I was fascinated with languages, I was still an undergraduate student in psychology and foreign languages; I knew next to nothing about speech and was totally unfamiliar with many of the concepts and jargon in linguistics. But her talk was like a story and it was so memorable – she showed us some of the different mental challenges associated with listening (like when she used speech waveforms to show us how gaps between words are not as clear as we think), why different languages are needed to better understand how the mind works when we listen, how infants’ early segmentation abilities influence later vocabulary growth – this was the first language-related talk I had attended and I was just so, so intrigued. 

I received the following letter and observations from the editor of a science journal:

I am an editor of a scientific journal (Geophysical Journal International) and
am writing to ask for a suggestion on communicating with some Chinese
authors on what I think is a linguistics issue. (I am also a long-term reader
of Language Log, and have enjoyed and learned from it).

We will be rejecting the paper because it is outside the range of topics the
journal handles. But it also has a writing style that I'd like to warn the
authors to avoid. Here is a sample (from the usual "review of previous work"):

This guest post by Mark Dow is an excerpt from an interview conducted via email in May/June 2021. The complete interview appears in PN Review #263 (Jan./Feb. 2022).

 Nigel Fabb is Professor of Literary Linguistics at the University of Strathclyde in Glasgow. His books include What is Poetry: Language and Memory in the Poems of the World (Cambridge 2015); with Morris Halle, Meter in Poetry: a New Theory (Cambridge 2008); Language and literary structure: the linguistic analysis of form in verse and narrative (Cambridge 2002); and Linguistics and Literature (Blackwell 1997). In 2022 he will have two new books, Thrills, epiphany, sublime: how literature surprises us. (Anthem) and with Venla Sykäri (eds.) Rhyme and Rhyming in Verbal Art, Song and Language (Studia Fennica Folkloristica).

Mark Dow is author of Plain Talk Rising (poems) and American Gulag: Inside U.S. Immigration Prisons (California).

As a published translator myself, I certainly strive to make my translations worthy of being considered as art.  But it isn't always an easy task.  Witness "The Tricky Art of Translation and Maria Dahvana Headley’s Modern Beowulf", CD Covington, Tor.com (Mon Feb 7, 2022):

It’s not very often that a thousand-year-old poem has a new translation that gets people hyped up, at least in the Anglophone world, but Maria Dahvana Headley’s recent Hugo Award-winning translation of Beowulf stirred up a lot of interest—there’s even a video series of writers and entertainers reading it out loud. (Alan Cumming’s section is excellent—he really knows his way around alliterative verse.)

…

We're following up on "Dinosaur Intonation" (8/28/2021) and "Hummed 'I don't know'" (8/29/2021). And today's installment starts with a distinction. There are two largely separate issues:

1. Intonational choices (for performances of "I don't know" or any other phrase).
2. Various types of articulatory reduction or replacement,
   from crisp hyper-articulated performances,
   through progressively slurred versions,
   to hums, grunts, or even whistled or instrumental imitations.

Homer Simpson's version, from this YouTube clip, lenites the consonants pretty much to extinction, and reduces the second ("don't") vowel as well, going beyond Michael Watt's comment about a friend who spelled it "iono":

[The numbers in black are f0 estimates in Hertz (cycles per second)]

But today we're going to focus on the intonational choices, rather than the words-to-hum continuum. And the method will be socratic — we'll give examples, and ask questions. The answers will emerge in later installments, or perhaps in the comments.

This Chinese article is about a man who has made a living by painting slogans and ads on village walls for thirty years. Some of the slogans are rather bizarre, as may be seen by looking at the many photographs in the article.

The article says it is such a well-paying job that the man was able to buy 6 apartments in his hometown with his earnings. Painting on walls is one of the major ways to advertise or propagate goods and ideas in the countryside.

There are many examples of such signs in the article, but I couldn't understand all of them upon first glance, so I wondered if the country folk would be able to read the signs. I asked a number of my graduate students from China, and they all said, yes, the country folk not only would be able to read them, but would enjoy them and would be motivated to buy the products and services promoted by the signs.

Dave Itzkoff, "Elon Musk Hosts a Mother’s Day Episode of ‘Saturday Night Live’", NYT 5/9/2012 ("The much-discussed Tesla and SpaceX executive took a self-deprecating approach, telling viewers, 'I’m pretty good at running human in emulation mode.'"):

Musk, the billionaire chief executive of Tesla and founder of SpaceX, appeared in several “S.N.L.” sketches this weekend, playing characters that included a doctor at a hospital that caters to Generation Z patients, the producer of an Icelandic TV talk show and the video game villain Wario.

He used his opening monologue to share some personal details about himself, introducing viewers to his mother and discussing his diagnosis of Asperger’s syndrome —  […]

Musk took a self-deprecating tone in his “S.N.L.” monologue, telling the audience: “Sometimes, after I say something, I have to say, ‘I mean that,’ so people really know that I mean it. That’s because I don’t always have a lot of intonational variation in how I speak. Which I’m told makes for great comedy.”

The question of intonational variation in the speech of people on the autism spectrum is an interesting one. In the literature and in clinical presentations, I've seen phrases like "As is well known, autistic individuals have monotone intonation", and also "As is well known, autistic individuals have singsong intonation".

This apparently reflects the fact that most observers of intonation only notice differences between what they expect and how people talk. So depending on  their relationship to the speakers and the contents and contexts of interaction, they might perceive the same speakers' intonation as inappropriately monotone or inappropriately varied. There may also be relevant subgroups within the large and extremely varied space of people "on the spectrum" — autism is one of the many DSM-defined behavioral categories that are "phenotypically diverse", which a clinician friend explains is the Greek translation of "We have no f-ing clue"…

From Doonesbury 5/2/2021:

Linguists have paid a lot of attention over the years to  wanna-contraction, starting with George Lakoff's 1970 paper "Global rules" — see these lecture notes for a discussion, if you're interested. But gotta-contraction has gotten a lot less attention — 7 Google scholar hits vs. 658.

The reason for this difference is simple: "want to" is occurs in different structures that have different contraction frequencies, thus entangling syntax, morphology and phonology in a pattern that people have been trying since 1970 to figure out how to untangle. "Got to" seems to occur in the same structures, but these turn out to involve quite different senses of get, which maybe even should be considered different words.

Following up on yesterday's "'Guess that'", this morning I looked at whether "speakers use the unstressed optional complementiser that to maximise rhythmic alternation of weak and strong syllables" in the case of complements following the verb believe. I again used data from Shuang Li's INTERVIEW: NPR Media Dialog Transcripts dataset.

One of the benefits of checking linguistic hypotheses in real-world data is that you sometimes stumble on unexpected and potentially interesting patterns. This morning's Breakfast Experiment™ provides an example.

Yesterday, as I prepared for a seminar on prosody and syntax, the following passage caught my eye (in Gerrit Kentner and Isabelle Franz, "No evidence for prosodic effects on the syntactic encoding of complement clauses in German", Glossa 2019):

A language production experiment by Lee & Gibbons (2007) suggests that speakers use the unstressed optional complementiser that to maximise rhythmic alternation of weak and strong syllables, as it is more often produced when the top of the complement clause starts in a stressed (Lucy) as opposed to unstressed (Louise) syllable (1).

(1) Ian guessed (that) {Louise, Lucy} signed the contract

Since Kentner and Franz found a contrary result in their experiment, I thought I'd see whether the effect that Lee & Gibbons found was replicated in a more natural dataset. So I turned to Shuang Li's INTERVIEW: NPR Media Dialog Transcripts dataset, which contains 3,199,859 transcribed turns.

Looking through the Penn Parsed Corpus of Modern British English (PPCMBE2), I saw that one of its sources is Chapter 10 of Volume 2 of Jane Austen's Emma. I've been using seven or eight different audiobook versions of that novel as a source of examples and exercises in ling521 over the past few years, so I thought I'd take a look at the relationship between syntactic structure and performance prosody in that chapter.

Listening to the second sentence raises some interesting questions:

Busy as he was, however, the young man was yet able to shew a most happy countenance on seeing Emma again. [source]

Details aside, it seems clear that in this sentence

• "however" is a kind of prosodic tag;
• "however" is prosodically bound to the phrase that precedes it.

Thereby, however, hangs a tale or two.