EPISODE TRANSCRIPT
Phonological Tone (Moira Yip)
Automatically transcribed from the episode audio. Some words and names may be inaccurate.
Listen to the episodeDownload textDownload captions
Welcome back to the Deep Dive. This is where we take a whole stack of sources and, well, give you the shortcut to understanding the core ideas. That's right. Today, we're diving into something pretty complex, something fundamental to how many languages work. Tone. We're looking at the acoustic and cognitive architecture behind it. Yeah, and it's fascinating because if you speak a language like English or maybe Spanish... You definitely use pitch. We all do. But we use it for intonation, showing emotion, asking a question versus making a statement. Right. That feels universal. It pretty much is. But that's fundamentally different from how... what, maybe 70 % of the world's languages use pitch, which is tone, using pitch changes to actually distinguish words themselves.
And that's exactly our mission for you, the listener, today. We've got a deep dive into the source material lined up. We'll start right down at the physics of sound production and work our way up to some pretty high -level theories. The goal really is to give you a fast but thorough understanding of what tone actually is, how it functions in language, and maybe what it tells us about, you know, the complexity of human language itself. It sounds like a big undertaking. Moving from just sound waves to actual linguistic meaning. It is. And that's why we absolutely have to start by clarifying three key terms that honestly get mixed up all the time. We need to build that bridge from the acoustic signal to the linguistic meaning. Step by step. Okay, let's start at the absolute base level. The pure physics. The sound itself. Right. That's fundamental frequency.
We're usually written as five dollars. It's purely acoustic. You measure it in hertz. Hertz. Okay. And it's simply the rate your vocal folds vibrate. How many times they cycle open and closed per second. Faster vibration means higher few dollars neck. That's it. It's a physical measurement. Totally independent of whether anyone's listening or what language they speak. Then we bring in the human element. The listener. Exactly. That gets us to pitch. Pitch is perceptual. It's what you, the hearer, actually hear. That subjective feeling of a sound being high or low. So it tracks via the dollars, but it's not identical. Precisely.
Pitch generally follows 7 .Rs, but it's not a perfect one -to -one mapping. Our perception of pitch can be influenced by other things, like volume amplitude, or even the sounds that came just before it. It's subjective experience. Okay, physical fault of dollars, perceived pitch. And now the crucial one for today, the linguistic one. And that's tone. Tone is the properly linguistic concept. It's defined by its job in the language, its functional load. Does a difference in pitch change the meaning of a word or a grammatical form? Like your Cantonese example. Exactly. In Cantonese, saying yo with a high -level pitch, let's say yo -h means worry. But say the exact same syllable, yo, with a low -level pitch, yo, and it means again. Wow. Same sounds, different words, just based on pitch. Completely different words. The segments Y, A, U are identical. It's the tone, the contrast of use of pitch that distinguishes them as separate items in the lexicon. And that functional difference is what we're really digging into today. Got it. So tone is about meaning, not just emotion or questions. Once we have that clear, let's look under the hood. How do we actually produce these precise pitch differences? What's the physical machinery? Okay, yeah. It all happens in the larynx, the voice box. The key structures there are cartilages, the cricoid at the base, the thyroid, that's the Adam's apple, and then the two little arytenoid cartilages at the back.
And the vocal folds stretch between some of those. They stretch between the thyroid cartilage at the front and the arytenoid cartilages at the back. So changing the pitch means physically changing the state of those vocal folds. Right. How do you do that? Tension. Mass. Both, essentially. For pitch raising, the main player is the cricothyroid muscle. When it contracts, it pulls the thyroid cartilage forward and down, sort of tilting it. Okay. And that stretches the vocal folds, makes them longer and thinner. This stretching does two crucial things. It decreases their effective mass and, more importantly, drastically increases their stiffness or tension. And skiffer things vibrate faster. Exactly. Faster vibration, higher $5, higher perceived pitch. The cricothyroid is like the pitch accelerator. Studies clearly show its activity ramps up just before and during a pitch peak. Okay, so that's the accelerator.
What about the brakes? How do you lower the pitch? Is it just relaxing that muscle? Pitch lowering is actually a bit more complicated. Yes, it involves relaxing the cricothyroid, reducing that tension, but that's often not enough. And just relaxing might make the voice unstable or breathy. So another muscle usually gets involved. The thyroarytenoid muscle. When it contracts, it actually bunches up the vocal folds, making them shorter and thicker. Increasing their mouth. Precisely. More mass means slower vibration, lower viadollars. So it's this coordinated dance. Relax the stretcher muscle, contract the thickener muscle. That gives you controlled low pitch. This anatomical setup gives us the physical range, that continuous gradient viadollar curve we can produce. But now we hit that really tough question. You mentioned the core of phonology, maybe. How does that smooth, continuous physical signal get interpreted by the brain as discrete, meaningful categories in a language? Yes. This is the phonetic -phonological divide, and it's absolutely central. The actual acoustic output, the $50 curve you could measure with the machine, is always gradient. It slides up and down across the speaker's range. Right.
But the mental representation, the phonology of the language... operates with categories, discrete units like high tone H or low tone L. It's like converting an analog signal to a digital one. So how do we as linguists or even as listeners tell the difference? When does a pitch change just? you know, a physical wobble? And when is it a signal that the language intends? Why should the listener care about this difference? You should care because your brain has to care. One type of variation is essentially noise predictable physical effects you need to filter out. The other type is the meaningful signal you need to decode. Okay, explain that. What makes something phonetic noise? Phonetic effects are typically variable and context dependent. They depend on things like how fast you're talking, the inertia of your vocal folds. It takes a tiny fraction of a second to actually hit the target pitch and co -articulation how neighboring sounds influence the pitch. Like it's harder to hit a really high tone right after a really low one. Exactly. There's a lag, a transition. And crucially, listeners are incredibly good at unconsciously compensating for these predictable phonetic variations. We sort of hear through the noise. to get to the intended underlying category. Whereas a phonological effect is different. It's not noise. Right. Phonological effects are categorical. Think of them like flipping a switch. It's either on or off, high or low.
If a rule applies consistently, regardless of speech rate, and results in a change that could potentially distinguish meaning, even if it doesn't in that specific word, it's likely phonological. The phonology provides the target category, and the phonetics just executes it as best it can within the physical constraints. This seems like a good place to bring in those three terms that describe pitch falling across an utterance. Declination, downdrift, and downstep. They show this gradient to categorical path, right? Perfectly. They illustrate the client from universal physics to language -specific grammar. Start with the universal one.
That's declination. It's a general gradual phonetic lowering of the overall pitch baseline across an utterance. It happens in pretty much all languages, tonal or not. Why does it happen? It's mostly tied to breath control. As you speak, your subglottal air pressure tends to decrease slightly, which naturally lowers the baseline pitch. It's just a physical consequence of running out of steam, essentially. Okay, so that's just physics. When does it become linguistic? It becomes linguistic when the language hijacks that tendency and makes it part of the grammar. That's downdrift. So it's phonologized declination. Exactly. Downdrift isn't universal. It's a specific rule in certain languages, and it's triggered by a particular phonological element, almost always a low L tone.
How does it work? In a language with downdrift, when you have a sequence like high -low -high, H -L -H, that second H tone is automatically realized phonetically lower than the first H tone. The L tone actively pulls the pitch register down for everything that follows it within a certain domain. Okay, so the L tone forces a step down. What about downstep, then? That sounds similar. It is similar, but... Even more abstract. This is where it gets tricky, especially in many African languages. Downstep, sometimes called non -automatic downstep, is when you get that same lowering effect on a high tone. But without an immediately preceding low tone. Exactly. The register steps down, but there's no overt L tone on the surface to trigger it. The cause is analyzed as a floating low tone. A floating tone, like a tone that's not attached to any vowel. Precisely.
It's phonologically there in the underlying representation. It's part of the word structure, but it doesn't get associated with any specific TBU tone -bearing unit, usually a vowel or mora. It's phonologically active but phonetically invisible itself. Okay, hold on. This is where I, and maybe the listener, might push back a bit. If the low tone isn't actually pronounced, how do we know it's there? Why not just say this language has a weird rule? Lower the pitch after this specific morpheme or something. That's a fair challenge. The reason we posit the floating L is twofold. First, the amount of lowering it causes is often identical to the lowering caused by a fully realized L tone in the same language, downdrift. It behaves like an L tone. Okay. Second, downstep often occurs precisely where we know, historically or through other alternations, that a vowel carrying an L tone used to be but got deleted. The segment disappeared, but its tone was left floating behind, still able to influence its neighbors. Ah, so the floating tone explains why the lowering happens there and not somewhere else. It provides the structural motivation. Exactly. Without the floating L, the downstep looks arbitrary, like magic.
With it, we see it as a regular phonological process to instability after segment deletion. This floating L can then cause iterative lowering. Each floating L encountered steps the whole pitch register down again. That makes sense. It's about finding the underlying system rather than just describing surface patterns. Okay, let's shift gears. We've got the mechanics and the basic phonetic phonology split. Now let's look at the big picture. Where is tone found, and how complex can these systems get? Right, the global distribution. As we said, maybe 70 % of languages. The hotspots are definitely sub -Saharan Africa, particularly lighter Congo languages. Also, large parts of the Americas, especially the Otomangian family in Mexico. And of course, East and Southeast Asia think Sino -Tibetan, Mon -Khmer, Thai -Kadai languages. And we mentioned Europe is mostly non -tonal, with those few exceptions like Swedish or Serbo -Croatian that are sometimes called pitch accent languages instead. Yeah, the key difference there is often inventory size and function. True tone languages usually have tones operating contrastively on almost every syllable or word. Pitch accent systems tend to have maybe one or two contrastive pitch patterns per word, often linked specifically to the stress syllable.
It's a bit less pervasive. Let's talk inventory size then. How many distinct tones can a language actually juggle? Well, the absolute maximum that's been reliably reported as phonologically contrastive is five distinct pitch levels. Languages like Dan, spoken in West Africa, are cited for this. But even four contrastive levels are pretty rare. So simpler systems are more common. Much more common. The most widespread type is the two -tone system, just high versus low, HVSL. Vast numbers of African languages work like this. But you mentioned something interesting earlier, that a simple H -L system can actually lead to three different pitches being heard on the surface.
Ah, yes, that's a key point about the phonetics -phonology interaction again. Take a language like Yoruba. The underlying phonological inventory seems to just be H and L. That's it. Okay. But the language also allows some syllables to be underlyingly toneless. They don't have an H or an L specified in the lexicon. When you actually speak, though, every syllable needs some pitch. So these toneless syllables often surface with a mid -pitch, M. Where does the mid -pitch come from? It's usually derived phonetically, often by interpolation between surrounding tones, or sometimes it's just a default neutral pitch realization for unspecified syllables. So you have two contrastive phonological tones, HL, but three distinct phonetic pitch levels heard, high, mid, low. Very cool.
What about languages that actually contrast three levels phonologically? Three -tone systems, HML, are also quite common, found across different language families globally. Noop in Nigeria is one example. Punjabi, an Indo -Aryan language, is another good case. Punjabi? How does that work? On stressed syllables, Punjabi clearly contrasts three levels. You get words like kara with a low tone, meaning horse, kara with a mid -tone, meaning whip, and kara with a high tone, meaning leper. The segments are identical, only the pitch level distinguishes them. So managing three levels is definitely possible. Four is rare. Five is the max. Yeah. Got it. Now, how do linguists actually write these tones down?
You mentioned different systems. Yeah. There are two main traditions, largely split by geography. The accent system is dominant in descriptions of African and American languages. It uses diacritic marks directly on the vowel. Like accents in French or Spanish, but for tone? Sort of, but with dedicated meanings. Typically, an acute accent means high tone, like ah. A grave accent means... Low tone, A. A macron often means mid -tone. What about tones that change pitch on one syllable, like a fall or rise? Those are contour tones, and the accent system uses combinations.
A circumflex might be used for a falling tone, HL, like R, and a caron, or HH, for a rising tone, LH, like A. It's pretty intuitive for systems with just a few level tones and maybe simple contours. But it gets messy if you have lots of levels or complex contours. It can, yeah. Which is partly why the other system developed, mainly for Asian languages, especially Chinese dialects with their complex tonal systems. That's the numerical system. Based on numbers. Yes, usually associated with the linguist Y .R. Chow, sometimes called Chow tone letters. It imagines the speaker's pitch range divided into five levels, where one is the lowest and five is the highest.
A tone is then described by one or more numbers indicating its pitch path. A high -level tone would be 55. A low -level tone, 11. A high -falling tone might be 51. A dipping tone, like the Mandarin third tone, could be 214. That seems very precise. You can capture quite complex shapes. It offers great phonetic detail, absolutely. But it has a significant drawback, especially for comparing across speakers or dialects. What's that? The numbers 1 to 5 are relative to that specific speaker's pitch range at that moment. My 5 -a -lo might be acoustically lower than her 4 -fun.
So a tone transcribed as 24, rising. in one dialect might actually occupy the same relative pitch space as a tone transcribed as 13 also rising in another dialect spoken by someone with a different overall range ah so the numbers look different but the tones might be functionally the same relative rise Exactly. The numerical system captures phonetic shape well, but it can obscure the underlying phonological categories, especially things like whether a tone belongs to a higher or lower register. That's a key theoretical distinction the numbers alone don't always reveal reliably. Okay, that makes sense. Let's tackle those contour tones more directly now. The rises, falls, dips within a single syllable. The big question seems to be, are they single, complex units, or are they just sequences of simpler level tones packed together? That's a major theoretical debate.
First, some observations. Falling tones, like HL51, are way more common across the world's languages than rising tones, LHU15. Interesting. Why might that be? We'll get to that. It relates to production difficulty. Also, complex contours, ones that change direction twice, like a fall -rise, HLH515, or a rise -fall, LHL151, are extremely rare, especially in African systems. So the dominant idea is the sequence hypothesis. The contours are just level tones stringed together. That's generally the starting assumption, yes. A falling tone is underlyingly H followed by L. A rising tone is L followed by H. What's the strongest evidence for breaking them down like that? It often comes down to syllable structure, specifically the moraic constraint. The idea is that each underlying tone, H or L, needs something to dock onto, a tone -bearing unit, TBU. In many languages, the TBU isn't the whole syllable, but the mora.
Remind us what a mora is. It's a unit of syllable weight. Roughly, a short vowel has one mora. A long vowel, or a vowel followed by certain consonants, like nasals or liquids sometimes, has two moras. A syllable with two moras is called hefi. Okay, so how does that relate to contours? If a contour tone is really a sequence of two level tones, like H plus L, then according to this hypothesis, it should generally require a two mora, heavy syllable, to host both tone units. A short, single mora, light syllable should only be able to host one tone unit, meaning only level tones. And do we see languages behaving like that? We do. There are many examples. A really compelling one comes from Wang Sian Chinese.
Syllables ending in a stop consonant, like P -T -K, are considered short, monomaraic. And guess what? They can only bear level tones. But there's a specific process, a type of reduplication, where one of these stop -final syllables needs to bear a contour tone phonologically. The language solves this dilemma in a fascinating way. The final stop consonant changes into a nasal consonant, like M -N -OR. Nasals can be moraic, so changing the consonant makes the syllable heady, bimoraic, and now it can successfully carry the required contour tone. Wow. The sound change is purely driven by the need to make room for the second part of the contour tone. Exactly. It's powerful evidence that the contour is being treated as a sequence of two things needing two docking spots. Okay. That's strong support for the sequence idea. But you mentioned there's still a unit hypothesis. Some people argue contours are single units sometimes. Yes. The debate isn't completely settled. In some languages, particularly certain Asian languages with very stable contour contrasts, there isn't always clear evidence for decomposition.
And sometimes contours behave phonologically like single, albeit complex, entities. What kind of behavior suggests they're units? One argument comes back to phonetic difficulty. Producing a rising tone, L plus H, is articulatorily harder and takes slightly longer than producing a falling tone, H plus L. Why is that? It involves more complex muscle coordination to initiate the rise from a low point. Some languages seem to restrict rising tones, banning them even on heavy syllables where a falling tone would be perfectly fine.
If the rise was just L followed by H, you might expect it anywhere a sequence is allowed. Treating it as a single, marked, complex unit helps explain why it might face special restrictions. So the jury might still be out. Or maybe both analyses are needed, depending on the language. That's often the case in linguistics. The sequence hypothesis is usually the default, but we have to be open to evidence that points towards unit status in specific cases. Okay, this complexity clearly requires some sophisticated theoretical tools. Let's move into Section 3 and talk about the architecture of tonal theory itself. How do linguists model these systems formally, starting with tonal features? If languages max out maybe four or five levels, how can we represent those contrasts efficiently using binary features like phonologists love to do? Right. The goal is always explanatory economy. Since a four -level contrast seems to be the practical upper limit for robust systems, a system with just two binary features is generally sufficient. What are the features? The most common proposal uses register and pitch. Register is usually seen as the more dominant feature, dividing the overall pitch space into two main bands, plus upper register and S upper register. Okay, a high band and a low band. Exactly. And then within each register band, the second feature, pitch plus high, distinguishes two levels. So that gives us four possible combinations. Precisely.
You get one plus upper plus high, the highest tone, let's call it H or H plus. two plus upper high a lower tone within the high band often called mid or m plus i or lowered high three upper plus high a higher tone within the low band often called raised low or m or sometimes also mid four upper high The lowest tone, L or L, that neatly generates up to four distinct levels. It also seems to capture why mid -tones sometimes pattern with highs and sometimes with lows. It can, yes. The two high tones, plus upper high and upper plus high, would naturally fall somewhere in the middle of the total pitch range, potentially behaving differently depending on whether the primary register feature aligns them with the upper or lower set. Now, the source material mentioned Duanmu's idea connecting these abstract features back to the actual larynx muscles we talked about earlier. Yes, that's a really appealing aspect of this feature system. Duanmu proposed a potential link. Register plus upper might correlate with the tension of the vocalist muscle, part of the thyroarytenoid. Plus upper could mean stiff vocal folds, potentially leading to a clear or even creaky voice quality, while upper could mean slack folds, perhaps leading to a breathy voice. And pitch. The pitch feature plus high would then correlate more directly with the activity of the cricothyroid muscle, the primary stretcher, essentially controlling how thin or thick the folds are vibrating. What's the advantage of linking features to specific muscle groups like that? Beyond grounding the abstract features in phonetics, it offers an elegant explanation for something quite counterintuitive.
Overlapping contours. Overlapping contours. What does that mean? It means you could have, say, a falling tone that starts high but ends quite low, maybe represented phonetically as 52. And in the same language, you could have a rising tone that starts low but ends mid -range, maybe 24. Notice that the endpoint of the high register fall, 2, is actually lower on the 1 to 5 scale than the endpoint of the low register rise, 4. If pitch were just one linear scale, this would be weird. Right, the ranges seem to cross over. But if register and pitch are controlled by potentially independent muscle systems, it makes sense. The 52 fall happens entirely within the plus upper register. Maybe plus upper plus high going to plus upper high. The 24 rise happens entirely within the upper register. Maybe upper high going to upper plus high. They operate in different compartments of the voice, allowing their phonetic ranges to overlap. That's a really neat way to think about it. Okay, so we have features. Now, how do we structure them? Why did phonology need a big revolution, like autosegmental phonology, specifically for tone? What was wrong with the old way? The old way, standard generative phonology in the 60s and 70s treated all features, including tone, as if they were just properties inherent to individual segments, like consonants and vowels.
A high tone feature was part of the feature bundle for a vowel, just like plus back or plus round. And it didn't work for tone. It failed spectacularly for tone because tone does things that segmental features just don't do. It seems to float around, spread across multiple vowels, stay behind when vowels get deleted. It behaves independently. So autosegmental phonology, ASP, developed by John Goldsmith and others, proposed what? It proposed that features like tone exist on a separate parallel tier, an auto -segmental tier, distinct from the tier where the consonants and vowels live. The segmental tier. Think of it like a musical melody line floating above the lyrics.
And the melody only connects to the lyrics at certain points. Exactly. The tonal tier connects to the segmental tier via association lines, which link a tone, like H or L, to a specific tone -bearing unit, TBU, on the segmental tier, usually a vowel or mora. And this separation explains tone's weird behaviors. The source material listed five key ones. Yes. ASP provides elegant explanations for all of them. One, mobility. If a tone's association line can be detached and relinked to a different TBU, that explains why tones can seem to shift or dock onto neighboring syllables, like in chizigula. Features like plus nasal can't just jump ship like that. Two, stability.
If the TBU, say a vowel, gets deleted, the tone that was linked to it doesn't have to disappear. it can remain on its tier and relink to a neighboring TVU that survives. This explains why tone often survives vowel deletion in languages like Haya or Pirahon. Right. The tone persists even when its original anchor is gone. Three. One -to -many association. A single tone on the tonal tier, say... one H, can have association lines linking it to multiple TBUs on the segmental tier. That's a tonal spreading or plateau where a whole sequence of syllables shares the same high pitch, like the unbounded H spread in Cholungo. Makes sense.
Four, many -to -one association. This is the flip side. You can have multiple tones on the tonal tier, like an L followed by an H, both linking to a single TBU on the segmental tier. That's the autosegmental analysis of a contour tone. Multiple tones sharing one docking spot. Which links back to our contour discussion. Exactly. And five, toneless segments. ASP allows for TBUs on the segmental tier that simply don't have an association line linking them to any tone on the tonal tier, at least underlyingly. Their surface pitch is then filled in later by phonetic rules, like interpolation. Japanese pitch accent has syllables like this. That framework seems incredibly powerful for capturing how tone works.
But ASP, at least initially, often assumed its rules and principles were universal and inviolable. That's where Actimality Theory, OT, came in, right? To explain the variation between languages. Precisely. OT, developed in the early 90s, took the insides of ASP, like tiers and association, but provided a new way to think about cross -linguistic variation. It said basically all languages share the same set of universal constraints, but they rank those constraints differently in terms of importance. Constraints. What kind of constraints are we talking about for tone? They fall into two broad categories. Facefulness constraints demand that the output form, what you actually say, be faithful to the input form the underlying representation for tone key ones are max st don't delete tones from the input and dpt don't insert new tones that weren't in the input okay so faithfulness says keep things the same what's the opposing force that's markedness constraints these demand that the output form be simple natural or unmarked for tone common markedness constraints include things like h High turns are articulatorily costly.
Avoid them if possible. Or no contour tour. Avoid contour tones on a single TBU. They're complex. So language variation arises from the conflict between wanting to be faithful and wanting to be simple. Exactly. Every language grammar is essentially a specific ranking of these universal but viable constraints. The optimal output for given input is the one that violates the lowest ranked constraints possible. Let's take a concrete example. The source material heavily features the obligatory contour principle, OCP. What is that, and how does OT explain its different effects?
The OCP is a powerful markedness constraint that basically says, adjacent identical elements are forbidden. In tonology, it typically applies to adjacent identical tones on the tonal tier, especially high tones, so HH is bad. So the OCP wants to get rid of sequences like HH. How can languages resolve that conflict differently based on constraint ranking? let's consider the conflict between ocp avoid hh and mxt keep all input tones we can see three different outcomes depending on how they're ranked one ocp maxt deletion If avoiding the OCP violation is more important than preserving every input tone, the language will simply delete one of the adjacent H tones and input HH surfaces as just H. This is exactly what happens in Mucin's rule. Famous in many Bantu languages like Shona, the OCP forces deletion.
Okay, OCP wins, the tone gets deleted. What's another option? Two, OCP no fusion. Fusion. Maybe deleting a tone is bad, max T is high, but merging two adjacent tones into one is acceptable. No fusion is ranked lower than OCP. In this case, an input H and H might surface as a single H tone linked to both original TBUs. The OCP is satisfied because there's only one H node on the tier, but arguably both input tones are preserved in the linkage. So the HH sequence is avoided by merging them. What if OCP is ranked low? Three, MAXT, OCP. Violation tolerated. If preserving input tones is paramount and the language doesn't mind OCP violations much, then an input HH will simply surface as HH. The OCP constraint is violated, but that's deemed acceptable to satisfy the higher -ranked Max -T. Languages like Kishamba, another Bantu language, actually allow sequences of H -tones, often realized phonetically with down -step between them. The ranking dictates whether you delete, fuse, or tolerate the OCP violation.
That OT framework really clarifies how languages can use the same basic ingredients but end up with very different surface patterns. Okay, moving into Section 4, let's see how tone interacts with the rest of the grammar. It's not just marking words, is it? Not at all. Tone is deeply integrated. It functions morphologically as a meaningful effix in its own right. Like adding a suffix, but it's just a tone. Exactly. We mentioned Cantonese using a specific high tone, often a changed tone derived from a high level 55, to mark familiar names or diminutives. You add the tone and the meaning changes. Or look at Aieo, a language from New Guinea. Verbal aspect, whether an action is ongoing, completed, etc., can be signaled simply by imposing a different fixed tonal melody onto the same monosyllabic verb root. The tone is the aspect morpheme. That's incredible. What about more complex morphological processes like reduplication? Does tone just get copied along with the segments? That's where it gets really interesting because the answer is... It depends.
Reduplication acts like a diagnostic. In Mandarin Chinese, for example, how tone behaves under reduplication depends entirely on the grammatical category of the word being reduplicated. If you reduplicate certain kinship nouns, the tone on the reduplicated part is deleted, leaving it toneless. If you reduplicate classifiers, measure words, the original tone is copied faithfully onto the reduplicated part, tone tone. But if you reduplicate adjectives to form adverbs, the reduplicated part gets a fixed high tone, 55, regardless of the adjective's original tone, tone 55. Wow. So the grammar dictates whether the tone is deleted, copied, or replaced based on the construction. Precisely. It shows that tonal rules aren't just blind phonological processes.
They're sensitive to morphological and syntactic context. This gets even more intricate in the complex verb structures of many Bantu languages. Ah, the morphological domains. What's going on there? Bantu verbs are famous for having multiple suffixes added to a root, creating very long words. Crucially, this morphology defines specific nested domains within the verb. There's the basic verb stem, then a larger verbal macro stem, including extensions, then an even larger inflectional stem, including tense suspect markers. And tonal rules care about these domains. Absolutely. Rules like the OCP -driven processes, like Meason's rule, deleting the second H and HH, often apply cyclically within these domains. A rule might apply first within the stem, then its output becomes the input for the rule applying across the macro stem and so on. The phonology has to precisely track the internal morphological boundaries.
Okay, shifting from word internal stuff to the sentence level. How do tone languages handle things like asking questions, showing surprise, or indicating topic -focused things that non -tonal languages often do with overall sentence intonation? That's a great question. While some tone languages do also use overall pitch range modifications for pragmatic effect, like raising the whole pitch register for a question in dang may, many rely heavily on sentence final particles. The little words tacked on to the end of the sentence. Exactly. These particles often carry little segmental meaning themselves, but they bear specific tones that convey the pragmatic or illocutionary force. Mandarin Chinese has several particles like ma for questions, ba for suggestions, ne for continuation. And Cantonese is even more extreme. Cantonese is famous for it, reportedly having somewhere between 35 and 40 distinct sentence final particles. They often use a segmentally minimal particle, like just a vowel A, which then serves as a host for different tones to signal very subtle pragmatic nuances, doubt, confirmation, impatience, shared knowledge, and so on. The particle is almost just a placeholder for the pragmatic tone. So the pragmatics gets segmentalized onto these particles. How does the larger syntactic structure, the sentence tree, influence tone syntax plays a crucial role in defining the phonological phrase phvh the phvh is the typical domain within which many tonal rules especially sandy rules where tones change based on context apply and how is the phvh determined it's often determined by mapping rules that look at the syntactic structure a common mechanism is edge alignment Languages differ on which edge matters. For instance, in Xiamen Chinese, a min dialect, the right edge of most syntactic phrases, like verb phrases or noun phrases excluding adjuncts, must align with the boundary of a tone group, triggering specific Sandy rules within that group. Other languages might align rules to the left edge of phrases. So the syntax carves up the sentence into domains for tone rules. What about Mandarin's famous third tone, Sandy? Where L plus L becomes LH plus L. Does that respect syntactic boundaries?
That's where it gets even more interesting. Mandarin tone zandy is famously complex. While sensitive to syntax to some extent, it also shows a strong preference for creating binary prosodic constituents, roughly. Two -syllable rhythmic units, or feet, sometimes called metrical rhythmic units, MRUs. So the rhythm matters more than the syntax sometimes. It seems so. The third -tone Sandy rule applies cyclically, often creating these binary feet. It might apply within a two -syllable foot, even if that foot crosses a syntactic boundary, or it might fail to apply between two third -tones if they belong to different feet, even if they're adjacent syntactically. The prosodic structure, the rhythmic organization, seems to exert its own influence, sometimes overwriting strict syntactic domains. That adds another layer of complexity.
Okay, let's put these theoretical tools into action now, looking at Section 5 regional patterns. We'll start with African systems, which you mentioned are characterized by mobility and downstep. Yes, if you had to pick one defining feature of many African tone systems, especially Bantu, it's tonal mobility. Unlike some Asian languages where a tone feels permanently welded to its lexical root, African tones often seem much more independent. How does that mobility manifest? Their surface alignment, where the tone actually docks, seems less determined by strict lexical pre -linking and more by general phonological principles operating across phrases or words. Things like align a high tone to the stressed syllable or align a tone to the left ray edge of the word or shift a tone to avoid an OCP violation. The tones seem to float more readily and dock according to grammatical context. The example of jishang. from Cameroon was mentioned regarding the tone -bearing unit.
Right. Zhishang provides really clear evidence that the mora is the TBU in that language. Crucially, contour tones, which need two timing units, are generally only allowed on syllables that are demonstrably bimaraic, either having a long vowel or a final consonant that contributes moraic weight. And it also showed something about which sounds can actually carry the pitch. Yes. Zhisheng Phonetics reveals that only sonorant segments, vowels, nasals, liquids can be the physical bearers of the philodollar contour. Obstruents, stops fricatives, can contribute to syllable weight, be moraic, and thus provide a potential docking site for a tone phonologically, but they cannot themselves carry the audible pitch. The tone has to be realized on an adjacent sonorant. Fascinating phonetic constraint. Now the downstep controversy.
Is it a real L -tone? causing the lowering downdrift or a ghost L floating tone? You said Africa is key here. It's a long standing debate. The unification approach tries to argue that all downstep ultimately derives from a low tone, whether it's overtly present or just historically deleted, but still active, essentially making all downstep a form of downdrift. Josham challenges that. Zhisheng seems to be a counterexample. According to the sources, it clearly exhibits non -automatic downstep high tones lowering without an overt preceding L, but it lacks automatic downdrift, where an overt L always triggers lowering. So you can have the ghost L effect without the real L effect.
Apparently so, in Zhang. This suggests that the abstract floating L analysis, non -automatic downstep, might be necessary and distinct from simple downdrift, challenging theories that try to reduce everything to just one mechanism. The nature of downstep remains a complex research area. Okay, let's hop continents to Asian systems. The sources highlight their tonal richness and complex sandy chains. Yeah, if African systems often showcase mobility, many East and Southeast Asian systems showcase sheer inventory richness and intricate tone sandy. Contextual tone changes.
Richness in what sense? The number of lexical tone contrasts can be very high. Cantonese famously has six or seven distinct tones, depending on the analysis. This tonal richness allows languages with relatively simple syllable structures, like Mandarin, which only has about 400 unique combinations of consonants and vowels, to create a vast lexicon. Right, you mentioned Mandarin goes from 406 segmental syllables to over 1 ,200 distinct syllables once you add the four main tones. Exactly. Tone dramatically expands the meaning -carrying capacity of a simple syllable inventory. Cantonese also has that change tone. Use morphologically. Yes, the change tone is interesting. It's often a high -rising or high -level tone, like 35 or 55, that replaces the original lexical tone, typically used for familiar names, diminutives, or certain types of common nouns. What's telling is that it often simplifies the original tone. If the original tone was a complex contour, like low -falling 21, the change tone replaces it with a simpler high pitch. It suggests a morphological process overriding phonological faithfulness to the base tone. bending example of Sandy seems to be the min -chain shift in dialects like Shenanmen and Chaoyang. What on earth is going on there? It's truly remarkable and frankly sounds incredibly difficult to learn. It's an extreme form of tone Sandy where basically every tone in the language's inventory changes into a different tone from the inventory when it occurs in a non -final position within a phonological phrase. Every single one. It's not just one tone changing, but all of them shift. Pretty much. It creates a chain effect. For example, in Schemen, the citation tone 53, high falling, shifts to 31, low falling, in sandy context. The citation tone 31, low falling, shifts to 55, high level. The citation 55 shifts elsewhere, and so on. No tone stays the same, and no two citation tones shift to the same sandy tone. That sounds less like a natural phonetic process, like assimilation, and more like just memorizing a giant substitution table. That's largely the consensus that these change shifts are significantly lexicalized.
Learners likely have to memorize the specific pairing for each tone. Tone X becomes tone Y in this context. But is it completely random or is there some underlying logic? It's not entirely random. The pattern is constrained by principles of markedness and inventory reduction. Non -final positions in a phrase are typically less prominent, weaker positions. Languages often restrict the types of contrasts allowed in weak positions, for example. Complex contours or very high tones. be disallowed there ah so the target inventory for the sandy tone is smaller or simpler exactly the input tone has to shift but it can only shift to one of the tones that are permitted in that non -final weaker context it usually shifts to the closest available permissible tone because multiple input tones might ideally want to shift to the same simple output tone like mid -level but can't To maintain contrast, they're forced to shift elsewhere, creating the chain reaction.
So it's memorized pairings. But the pairings themselves are shaped by phonological constraints on permissible tones in context. Wow. Okay, third continent, American systems. The sources point to stress tone interactions and floating tones, especially in mixtec. Yes, the Otomangian languages of Mexico, like Mixtec and Trach, are famous for their complex tonal systems, often involving intricate interactions with stress and morphology. Mixtec provides classic examples of floating tones behaving in very specific ways. Like the floating high tone in San Miguel Grande Mixtec.
What does it do? This floating H tone is underlyingly present but not linked. It wants to dock and it moves rightward across the word looking for a place. Crucially, it tends to skip over vowels that are specified as toneless. which surfaces mid why skip them the analysis suggests it's governed by a constraint called tonal prominence faith basically the language wants to preserve the inherent tonal specification of each vowel landing the floating h on a toneless vowel would change its status from neutral mid to high violating this faithfulness so the h keeps moving until it finds a vowel that already has a lexical tone h or l where docking is less disruptive to the overall profile That's incredibly specific behavior.
Does anything stop this floating H? Yes. Glottal stops are reported to act as barriers. The floating H cannot cross or dock onto a vowel immediately following a glottal stop. It shows that segmental features can interact with and block purely tonal processes. Mixtech also flips the usual stress -tone relationship, right? Usually stress attracts tone, but here... Here, tone seems to determine stress. In some mixed tech varieties, stress placement isn't fixed lexically, but is assigned based on the tonal sequence. Specifically, stress tends to fall on a syllable that is immediately followed by a syllable bearing a lower pitch.
So like H followed by M or L, or M followed by L. Exactly. The pitch drop seems to create the prominence. Stress falls on the peak right before the descent. It's tone -dependent stress, which is typologically quite rare. And finally for the Americas, TRIQ shows tone acting as pure morphology. Completely. TRIQ has grammatical morphemes, particularly person and number markers, on verbs that consist solely of a floating tone. For instance, adding a floating four tone on a one to five scale might mark the first person plural exclusive. This floating tone docks onto the verb root, combining with the root's own tone to create a specific contour tone that signals that grammatical meaning. No segments at all, just the tone carries the grammatical function.
Amazing. Okay, final section, section six, perception and acquisition. How do our brains actually handle these complex systems? Starting with adult perception. Our ears are pretty sensitive, aren't they? Incredibly sensitive. Psychoacoustic tests show we can detect just noticeable differences, JNDs, in frequency as small as 0 .3 hertz in ideal conditions. But languages don't use contrast that small, do they? Not even close. Even languages with relatively crowded tonal spaces, like Mambila with four level tones, keep those tones spaced much further apart, maybe 10 hertz or more between adjacent levels. Why the big gap? It ensures robustness.
Languages build in a large perceptual buffer zone so that minor production variations or noise don't cause confusion between phonologically distinct categories. We can hear tiny differences, but languages only make use of much larger, reliably perceivable differences for contrast. And there seems to be a cognitive limit on how many levels we can reliably track. The five -tone maximum. Yes, experimental work by Pollack back in the 50s asking listeners to identify absolute pitch levels presented in isolation found that performance drops off sharply beyond about five categories.
People just can't reliably assign sounds to more than five absolute pitch bins. And this might explain why no language seems to have more than five contrastive level tones. It's considered a likely perceptual constraint. While we can hear relative pitch changes very well, Our ability to identify absolute pitch levels is limited, and this seems to cap the maximum inventory size for level tones globally. We also have evidence that our brains treat tone linguistically, not just as raw sound, right? Yeah. The idea of categorical perception. Absolutely. Classic studies compared native Mandarin speakers and English speakers listening to synthesized sounds that gradually changed from a level tone to a rising tone. What was the difference? The English speakers perceived a smooth, continuous change. As the rise got steeper, they just heard it as more rising. The Mandarin speakers, however, perceived a sharp, sudden switch. Sounds on one side of a specific acoustic boundary around a 12 hertz rise were consistently labeled level tone, and sounds just slightly different on the other side were consistently labeled rising tone. So their language experience created a category boundary where physically there was none. Exactly. They perceive the sounds through the filter of their phonological system.
Pitch perception becomes categorical for linguistically relevant contrasts, just like how English speakers perceive a sharp boundary between B and P along the voice onset time continuum. What helps us perceive tone besides the raw pitch, Fever Dollars? Well, Fever Dollars is definitely the primary cue. Other things help. Amplitude, loudness, often mirrors the Fevedalar contour higher pitch, often correlates with slightly higher amplitude, and this can provide a secondary cue. Also, as we mentioned, listeners perform incredible feats of contextual compensation, normalizing for the speaker's overall pitch range and accounting for effects like declination and downdrift to identify the intended underlying tone correctly.
Okay, let's turn to kids. First language acquisition, L1. How early do infants tune into the tones of their language? amazingly early. Studies with infants exposed to tonal languages like Yoruba show that even by six to eight months old, they are already better at discriminating a specific tonal contrast relevant to their language, like high versus non -high, compared to infants learning non -tonal languages like English. So they're already specializing perceptually before they can even speak. Yes. And conversely, English -learning infants, who might initially be sensitive to some tonal contrasts, gradually lose that sensitivity because it's not relevant or contrastive in their linguistic input. The environment shapes the perceptual system very early on. What about when kids start producing tones? Is there a universal order, like certain tones are always easier?
Not entirely universal. It's heavily influenced by the specific target language system. While there's a general tendency for rising tones, L plus H, to be acquired later across languages, likely due to their articulatory complexity. Right, the production difficulty we talked about. The specific order for level tones varies. In Mandarin, kids typically master the high -level 55 and high -falling 53 tones first. The dipping 214 and rising 35 tones come much later and are more error -prone. In Cantonese, it's the extreme -level tones the highest 55 and the lowest 11 that are acquired earliest. In Thai, interestingly, the mid -level and low -level tones tend to be acquired first, with high and falling tones coming later. So kids prioritize the most distinct or perhaps most frequent tones in their specific input language. It seems that way. They latch onto the most stable and contrastive points in their particular system first. And finally, what about acquiring the complex rules involving tone?
like Sandy or the rules linked to syntax. Those tonal alternations generally come relatively late in acquisition, often well after children have mastered basic sentence structure and segmental phonology. Mandarin third tone Sandy, for instance, is notoriously difficult, with kids still making frequent errors even after age three. And Deborah DeMuth's work on Sosotho showed a difference between purely phonological rules and syntax phonology interface rules. Yes, that's a crucial finding. Sisotho children seem to master purely phonological tonal rules, like high tone doubling, spreading an H tone, relatively early, maybe around age three. But rules whose application depended on identifying a specific syntactic domain, like lowering tones only at the end of a verb phrase, VP, were mastered significantly later. So understanding the syntactic structure needed to correctly apply the tone rule is the harder part. It appears so. The interface between syntax and phonology, especially for complex phenomena like tone, seems to be one of the later developing aspects of grammar acquisition. Okay, we've covered a huge amount of ground, from the larynx muscles all the way to infant perception and OT constraints. As we wrap up, what are the big takeaways here? I think what's really striking, synthesizing all this, is seeing tone not just as pitch variation, but as this incredibly intricate structured system. It's grounded in physical anatomy, yes, but it operates through abstract phonological tiers. It's governed by universal competing pressures like the OCP and faithfulness, modeled beautifully by things like ASP and OT.
And we've seen how different language families have exploited this system in really different ways, haven't we? Africa focusing on mobility, using things like floating tones for downstep. Asia, building these massive lexical inventories and developing incredibly complex, almost Rube Goldberg -esque Sandy chains. Exactly. The regional strategies highlight the flexibility within the universal toolkit. Which leads us to that final provocative thought, maybe. There's this constant tension, isn't there? There really is. On one hand, we see clear evidence for universal constraints rooted in perception and articulation, things like the probable five -level limit or the inherent difficulty of producing rising tones. These pressures should, in theory, push languages towards simpler, more natural systems. But then we see languages like Shaiman Chinese with its chain shift. Exactly.
Or Mixtec with its tone -dependent stress and quirky floating tone rules. Languages devise these wildly complex... sometimes seemingly arbitrary systems that fly in the face of phonetic simplicity and seem to demand sheer brute force memorization from learners. So for you listening, the question to ponder is about this balance. How much does the universal toolkit, the constraints favoring simplicity and naturalness actually shape language? And how much is language shaped by, well, historical accidents? The specific way constraints got ranked in this particular language is history, leading to the preservation or even elaboration of a really complex marked pattern like that min -chain shift. The fact that these constraints are viable, that rankings differ, means the world's languages showcase this incredible diversity born from the collision of physics, perception, grammar, and just time. It really makes you wonder what other complex, maybe slightly illogical patterns exist in language, maybe even in your own language, that are only there because, generations ago, one specific constraint ranking one out over another. It's a fascinating puzzle.