Ask Language Log: Easy but unused initial clusters?

« previous post | next post »

From Bob Moore:

I have recently become interested in an important Alaska native weaver named Jennie Thlunaut. The linguistic question is about the initial consonant cluster of her last name, "thl". My initial reaction on seeing the name was that this consonant cluster was not phonotactically possible in English, and that it would be hard for me to pronounce. I was surprised to find that it was very easy for me to pronounce, without the perception of a highly reduced vowel separating the initial consonants that I usually experience when trying to pronounce a foreign word containing a consonant cluster not found in English.

I confirmed that "thl" does not seem to be a possible word-initial consonant cluster in English by grepping for all case-insensitive instances of " thl" in the English Gigaword corpus. I found something between 100 and 200 of these, and I examined all of them, finding then all to be either (1) foreign words or names, (2) attempts to represent the pronunciation of foreign words or names, (3) representations of "lisping" in English, or (4) typos.

I am puzzled that there would be an easy-to-pronounce phonological sequence that is completely unused in a language. It seems like coding efficiency would favor using any sequence that is easy to pronounce. Is there a more general phonological principle in English that would block the use of "thl"? Are there other easy-to-pronounce consonant clusters that are not used in English?

There are two different questions here, I think:

  1. Why are some of the consonant clusters (or other sound sequences) that are not used in a given language easier for native speakers to pronounce, while others are harder?
  2. Why are some of the easy-to-produce sound sequences nevertheless not used in a given language?

The answer to the first question probably has a universal part and a language-specific part. (I say "probably" because I don't know of any research results, I'm just reasoning from general principles.)

The universal part is that syllable centers are generally sonority peaks, with sonority rising from the start of the syllable to the center, and then falling to the end. Whether this is exactly the right way to think about it or not, it's true that fricative+liquid onset sequences are more "natural" in some general sense than liquid+fricative onset sequences would be — and the opposite is for syllable-final sequences.

And the language-particular part is that English has lots of /fl/ and /sl/ onsets, with /ʃl/, /vl/, /zl/, etc. being more marginal but certainly possible. So adding /θl/ is not nearly as big a step as if English had no initial clusters at all, or no fricative+liquid clusters.

What about the second question, "Why are some of the easy-to-produce sound sequences nevertheless not used in a given language?"

Presumably this is because phonotactic redundancy — which is a fancy way of saying that some of the natural sound sequences are not used — creates a sort of error-correcting code at the phonetic level.


  1. Bob Ladd said,

    January 13, 2018 @ 3:12 am

    There is probably a historical explanation here as well. English /θ/ can often be traced back to Indo-European /t/ (e.g. three, Latin tres, etc.), so a cluster /θl/ might be expected to reflect an IE /tl/. But /tl/ and /dl/ onset clusters are cross-linguistically quite rare (presumably because they're hard to distinguish acoustically from /kl/ and /gl/). If Proto-IE didn't have any /tl/ initial clusters, there would be no straightforward historical source for English initial /θl/ clusters. They could have been borrowed, of course, but /θ/ itself is pretty rare cross-linguistically, so potential sources of loanwords with /θl/ would also be hard to find.

    But there's definitely a mismatch between ease of pronunciation and use in a given language, even for individual phonemes. English /ʒ/ doesn't normally occur word-initially, but nobody has any trouble pronouncing Zsazsa Gabor with initial /ʒ/. English /ŋ/ doesn't occur word-initially either, but many English speakers have a hard time pronouncing Ngaio Marsh with initial /ŋ/.

  2. Michael Watts said,

    January 13, 2018 @ 3:13 am

    I tend to agree with the last paragraph at a very fundamental level — something like "the structure of language allows for situations like this because it is advantageous" — but I think that in any specific case it may be easy to find more direct, and less satisfying, answers.

    LL has mentioned before that many syllables (e.g. /ty/, /si/) are perfectly compatible with Chinese phonology, but just don't exist. Even among the syllables defined by the hanyu pinyin standard, which /ty/ and /si/ are not, many (sěng; nǘ) don't exist. See also the discussion of duang.

    An explanation that I recall from that earlier discussion was essentially that words in a modern language by and large descend by regular processes from words in an ancestor language. Word-initial "thl" may pose no problems for modern English speakers, but perhaps any structure in an older form of English that would have become modern word-initial "thl" was impossible for those earlier speakers.

  3. Michael Watts said,

    January 13, 2018 @ 3:14 am

    Ah, I see Bob Ladd anticipated my comment.

  4. Michael Watts said,

    January 13, 2018 @ 3:18 am

    On a different note, I feel like that fact that "slow" and "throw" are both normal English words probably has something to do with the easiness of "thlow".

  5. Andy said,

    January 13, 2018 @ 3:24 am

    As a non-specialist, I'm wondering if is there any reason (apart from IPA conventions) why /thl/ is considered to be a cluster rather than a single sound. In Welsh for example, the common sound of the double 'll' (as in Llewelyn, or Llanelli), although pronounced 'chl' or 'khl', is apparently written simply as [ɬ] in IPA (according to a wikipedia search just now), which is just a single character/letter.

    A related note: most English speakers have no problem with the French 'j' (as in 'je', 'jolie') and would not pronounce is as 'dj' apart from in names such as Angelina Jolie. But I don't think there are any examples of English words in which this French 'j' sound is used.

  6. Andy said,

    January 13, 2018 @ 3:26 am

    Ah … like Michael, I now see that Bob Ladd had already anticipated my second comment!

  7. John Walden said,

    January 13, 2018 @ 3:32 am

    "In Welsh for example, … double 'll' (as in Llewelyn, or Llanelli),… pronounced 'chl' or 'khl'"

    Although common, strictly speaking this is a mispronunciation. The sound is really an unvoiced L. You put your tongue in the L position, perhaps slightly flattened and blow, without a /k/ in front.

  8. Gregory Kusnick said,

    January 13, 2018 @ 3:37 am

    Genre seems pretty thoroughly assimilated into English.

  9. Rubrick said,

    January 13, 2018 @ 3:49 am

    For fans of fictional British rabbits, the Lapine language features the initial "thl" cluster, as in Thlayli (Bigwig).

  10. Chris said,

    January 13, 2018 @ 4:14 am


  11. Tom Dawkes said,

    January 13, 2018 @ 4:36 am

    This is almost certainly the voiced lateral fricative [ɬ]. English language spellings of Tlingit — noted in Wikipedia as Lingít [ɬìnkít] — have often used this trigraph to cope with what is to English-speaking ears an outlandish sound (though Tlingit has a range of lateral fricatives and affricates, including [tɬ’] and [ɬ’]).
    A similar case appears in the representation of Zulu laterals. Zulu has both the voiceless [ɬ] AND the voiced [ɮ], and Zulu orthographies used 'hl' for the former and both 'dl' and 'dhl' for the latter, as in the place name Isand(h)lwana. A common Zulu family name is 'Dhlamini'.

  12. Michael Watts said,

    January 13, 2018 @ 5:11 am

    A related note: most English speakers have no problem with the French 'j' (as in 'je', 'jolie') and would not pronounce is as 'dj' apart from in names such as Angelina Jolie. But I don't think there are any examples of English words in which this French 'j' sound is used.

    I have encountered this opinion [that the sound [ʒ] is foreign to English, even if it's easy to pronounce] from many people. I'm always confused, because /ʒ/ is very much a part of normal English. It is required in order to pronounce "vision" / "collision" / "precision" / "television" / etc etc etc etc. Those aren't rare. More generally, it is the natural reduction of /zj/. My best theory is that people confuse the fact that there is no single letter representing this sound with the idea that the sound itself does not occur.

  13. Robert T McQuaid said,

    January 13, 2018 @ 5:56 am

    English phonics includes the ts sound, as in the final consonant cluster of cats. Still, most professional announcers pronounce tsunami omitting the initial t.

    One possible problem: Pronouncing "the tsunami" at the speed used to teach a kindergartner is easy, but at the normal speech rate it leaves me tongue tied. Other initial clusters may offer the same difficulty.

  14. Joshua Gibson said,

    January 13, 2018 @ 6:09 am

    I think Tom was accidentally overstating his case, Mr. Watts. As Bob Ladd noted [ʒ] is either non-existent or very rare as a word-initial sound in English (The French pronunciation of initial-j having morphed into /dʒ/ ). Even genre, mentioned above, is frequently pronounced with [dʒ], though [ʒ] is more common. All of your examples are of words ending in "sion", with the [dʒ] embedded in the word. Likewise, /θl/ clearly occurs in English, for instance "Kathleen", it just is not employed at the beginning of words.

    Which is itself an interesting question: why sounds that are relatively common within words are unused to begin them, even when historical and related languages still use them. There are many examples of this – Some initial consonant clusters that have fallen out of English still appear orthographically and the lost of part of the initial cluster does not represent inability of English speakers to pronounced the sound. Pronouncing the initial /k/ in "knife" seems difficult but is not – "nicknack" is easy to pronounced. Likewise the initial "p" in Ptolemy seems hard but occurs easily in "ineptitude".

  15. David Marjanović said,

    January 13, 2018 @ 6:24 am

    IIRC, Gothic still had a few words beginning with [θl], descended from the similarly few Proto-Indo-European words that began with [tl]; but Northwest Germanic merged this cluster into the much more common [fl]. So, English lacks [θl] because it was so rare that it was lost.

    Another example of this phenomenon comes to mind from (High) German, where not only Proto-Germanic [tw], but also [dw] have become [tsv] in initial position. [dw] was rare to begin with – still is in English; there are less than 10 words beginning with it –, and while [tw] and [dw] were at first shifted to [tsw] and [tw] as expected, this new [tw] was so rare that most of these words joined the [tsv] set, while the remaining one or two joined the [kv] set. (…I'm not sure when the [w] > [ʋ] shift happened, let alone [ʋ] > [v] which hasn't in fact happened in most accents of German to this day.)


    The Welsh ll is not a sequence [xl], it's a single sound; it's a fricative like [x] and lateral like [l], but it's both at the same time – it doesn't switch from one to the other.


    Further occurrences of [ʒ] in English: Asia; for many Americans a slightly fronted version of it is also the outcome of any |z-j| sequence even across word boundaries, like in is your or was your. But yes, all the word-initial occurrences seem to be obvious loanwords, mostly from French; that's because word-initial [z] is likewise restricted to loanwords in English outside of Somerset.

  16. RP said,

    January 13, 2018 @ 7:01 am

    As far as "genre" and "Asia" are concerned, both have variant pronunciations – with /dʒ/ and /ʃ/ respectively. Similarly "garage" – though few dictionaries give anything other than /ʒ/ for "mirage". So Michael Watts' examples of "vision" etc are more compelling (at least for non-initial examples of /ʒ/). While this can be analysed as a simplified /zj/ in origin, it is difficult to believe that many speakers, if any, would use /zj/ in a word such as "vision" (whereas "presume", for exampe, varies between /ʒ/ and /zj/, with the latter considered more correct in prescriptivist quarters).

  17. rosie said,

    January 13, 2018 @ 7:01 am

    More examples of initial clusters which are easy to say but don't occur initially in English:
    mr- (e.g. [Yevgeny] Mravinsky)
    mw- (e.g. Fr. "moi"; [Emil] Młynarski)
    nw- (e.g. Fr. "noix"; [Chi-chi] Nwanoku)
    Indeed several clusters ending in [w] (which might get devoiced by assimilation if it follows an unvoiced consonant) which don't occur in English except in exotic borrowings (e.g. bwana) and are about as easy to say as the corresponding clusters ending in [j] which do occur in English, thanks to the realisation of long u as [ju:].

  18. Jarek Weckwerth said,

    January 13, 2018 @ 7:05 am

    Isn't it the case that the absence of a sequence that is otherwise, in principle, compatible with a language's phonotactics rather arbitrary? That's what the rather standard term "accidental gap" suggests, after all. In other words, unless you are willing to dig quite deep, the best first answer to the question of "why isn't this sequence there" is, erm, "because". It is disappointing I know, but we are grown up, right?

    In terms of perceived ease: This is at least partially a question of token, not type, frequency. For example, in English, the most frequent nitial cluster is /fr-/, and that is of course due to the high frequency of a single word, from. In language-independent, phonetic articulatory theorizing that can be summarized using a sonority hierarchy or similar machinery, it is less optimal than many other clusters that are less frequent. (The second is /st-/, also sub-optimal and therefore widely discussed).

    As another example, in Polish, the most frequent cluster is /pʃ-/. This is again due to just three words, przy 'at', przez 'through' and przed 'before', and the associated morphemes that just happen to be used in a gazillion of words. No-one would seriously claim that /pʃ-/ is phonetically good. The argument needs to go the other way: the cluster is there because it happened to "serve" in a couple of frequent words.

    Katarzyna Dziubalska-Kołaczyk and her team have done quite a lot of work on this, including this book. (Disclosure: I'm at the same faculty. If this is too much of a shameless plug, then please delete.)

  19. Keith said,

    January 13, 2018 @ 7:36 am

    As an initial cluster, I don't think "spw", "shp" or "tsr" exist natively in English.

    There's nothing difficult about, for example, "spwout" or "shpout", but then since initial "sp" is very common, and /ʃp/ is not much different.

    I travelled around Yugoslavia (as it then was) with somebody who found "Cres"/ t͡srɛ̂ːs/ impossible, though I found it easy.


    A related note: most English speakers have no problem with the French 'j' (as in 'je', 'jolie') and would not pronounce is as 'dj' apart from in names such as Angelina Jolie. But I don't think there are any examples of English words in which this French 'j' sound is used.

    "Leisure" is usually pronounced /ˈlɛʒə/ in British English. I've also heard "seizure" pronounced most often /siːʒə/. Both these entered English from French.

  20. John Swindle said,

    January 13, 2018 @ 8:18 am

    For English speakers the /θl/ cluster in "Thlunaut" is unfamiliar but not difficult, for reasons discussed, whereas the voiceless lateral fricative would be both unfamiliar and difficult even if it were closer to the original.9

    As for tsunami, I give you Sugiyama.

  21. Doug said,

    January 13, 2018 @ 8:18 am

    Even odder (to me at least) than the absence of "thl" words is that /skl/ is almost unknown as an initial cluster in English. There's "sclerosis" & related words, and a few proper names, but /skl/ is oddly rare given that /sk/, /kl/ are both common, as is /skr/.

    The (near) lack of /skl/ words seems unmotivated.

  22. Zeppelin said,

    January 13, 2018 @ 9:54 am


    There aren't all that many Proto-Indo-European roots with three initial consonants to begin with. And a lot of roots beginning with s can occur with or without the s in the daughter languages for unclear reasons ("s mobile"), so that further reduces the odds of skl-.

    You get more CCR- clusters in the zero grade, with the resonant being syllabic, but that ends up as CCuR- in Germanic, so a PIE zero-grade *skl̩- just gets you PGerm *skul-.
    I actually can't think of any PIE root with *skl-. The scl- in "sclerosis" seems to be the result of metathesis (*skle-ro < *skelh₁-ro) for example.

  23. Rose Eneri said,

    January 13, 2018 @ 10:02 am

    I know this topic has been discussed before, but I think it is on point here. I always thought that the consonant cluster (CC) "shtr" was not native to English, yet it is becoming quite common, as in saying "adminishtration". I can understand reducing a
    CC, such as, wr to r, or kn to n. But, injecting an entirely new CC with no contrastive value is a mystery to me.

  24. ~flow said,

    January 13, 2018 @ 10:37 am

    I'd hypothesize that just as languages would appear to have a somewhat uncanny and impractical tendency to 'hoard usage frequencies' as it were, they also tend to amass 'negative hoards', 'empty quarters' or 'fallow fields'.

    Like take Chinese for an easy example with its +400 basic syllables (not counting tones); depending on how many words you allow into your corpus, you'll find many dozens of bound and free morphemes of the forms (Pinyin here) yi and xi; on the other hand, some syllables are rare, like rua, chuo; then there are frequent re, ru, ruo, but *ra, *rai, *ruai are absent (to the best of my knowledge and the table at

    Maybe there's a universal statistical tendency like Zipf's law / the Pareto principle at work here, tendencies which work on completely unrelated aggregates of 'things' like words in a language or lengths of rivers or mountain heights on Earth.

    When you look at the universe at large, you see lots of stars and galaxies and black holes, but at the same time, there's not only the empty space between those, there's also vast expanses that are apparently completely devoid of any kind of matter. Also, scientists believe that as much as they can today chart mass distributions, galaxies seem to cluster into strands of matter. That could well be different. But then, matter attracts matter and more matter attracts even more matter; on the flip side, more matter in a place also means more kinetic energy, so two or three bodies may work *both* to attract a fourth one *or* to forcefully expel (eject) that third body (which is one factor to explain why not more matter has already condensed into a few huge and dense bodies). So attractive and repellent forces contribute to create a dynamical and deeply structured system that is neither 'all eggs in a single basket', nor are those eggs all evenly spread out, given the positions of that egg carton that we may liken the phonology of a given language to.

    Likewise, analogy and familiarity may work in language to allow more and more similar things, like 'sl-', 'shr-', 'fl-', 'fr-' and so on; but at some point, that particular place (initial fricative-liquid clusters) becomes crowded, so 'thl-' and 'sr-' have a hard time to enter, as their admission would put an ever higher load on both listeners and speakers. So that's a lot to do with expectations and experience, too: English speakers have little expositions to words with 'thl-', a little more to words with 'sr-' (as in Sri Lanka). If a Ms Thlunaut was to become POTUS and a much spoken-about person, then certainly 'thl-' would be one step closer to attaining citizenship in English.

    BTW 'thlipsis' is interesting b/c English has the tendency to regard all of Latin and Greek as legitimate spoil. Roy Andrew Miller had a catchy term for when one language (Japanese in his case) views the total of another language's lexicon (formerly Chinese and today English in his case) as being admissible to or part of that language (can't find that term right now tho). Thlipsis is hardly a common English word, but because it is (presumably) Greek, you sort-of-can adduce it as evidence for "there are English words with thl-".

  25. Zeppelin said,

    January 13, 2018 @ 10:41 am

    Rose Eneri:

    Str- > shtr- is presumably conditioned by the English postalveolar r. Shtr- uses the same place of articulation for the whole cluster, while str- requires you to slide back your tongue for the r. Shtr- is easier to articulate, in other words. This also explains why the shift seems to be happening mainly in American English, where the r is typically further back.

  26. Robert Coren said,

    January 13, 2018 @ 11:16 am

    I'm slightly puzzled by "with /ʃl/, /vl/, /zl/, etc. being more marginal but certainly possible" — which reads to me as if these combinations do occur initially in English, if rarely. I cannot offhand think of any English words that begin with /vl/ or /zl/. The few words with /ʃl/ that I can think of (shlock, shlub) are imported from Yiddish.

    [(myl) I guess it depends on what we mean by "occurs in English". There are slavic names like "Vladimir"; there's the neologism "vlog"; there's the mineral "vladimirite"; there's the ethnonym "Vlach" (defined by the OED as "A member of the Latin-speaking race occupying portions of south-eastern Europe; a Walachian or Romanian", and cognate with "Welch"; there's "zloty" ("A gold or silver coin of monarchic Poland"); there are the 17th century "minced oaths" "Zlid" (="God's lid") and "Z'life" (="God's life"). ]

  27. John said,

    January 13, 2018 @ 11:33 am

    This question seems to be based on the belief that this initial cluster is pronounced /θl/, however I think this is a misreading. The person asking the question mentions that the person in question is an Alaskan native, and no native Alaskan language features initial consonant clusters as far as I know. There are however several Athabaskan languages in the state, and these all have a series of lateral affricates /t͡ɬ t͡ɬʰ t͡ɬ’/, which in many of the languages are usually written , but in the absence of a means of typing on a keyboard could be substituted, which could give as alternative written forms. That's my theory as to this spelling.

  28. Andreas Johansson said,

    January 13, 2018 @ 11:46 am

    @Robert Coren:

    The one English word I can think of with initial /vl/ is "vlog", which of course is a recent neologism (it's a contraction of "video blog"). Still, that it's got some currency must show that many speakers find it unobjectionable.

    I can't think of any English word begining in /zl/.

    On the subject of gaps, English has initial pr-, pl-, tr-, kr-, and kl-, but no tl-. Anyone know if there's a diachronic reason for this gap?

    (If there is, I imagine it also explains why, where skl- is rare, stl- doesn't exist.)

  29. Jarek Weckwerth said,

    January 13, 2018 @ 11:52 am

    For /zl/, you'd have to go to Slavic borrowings/quotes such as Zlatan (Ibrahimović) or zloty (the currency of Poland). I don't think they turn too many heads phonetically.

    (A correction to my previous post [written from memory…]: /fr/ is the THIRD most frequent initial, not THE most frequent one; that honor goes to /st/, of course.)

  30. Tim Morris said,

    January 13, 2018 @ 11:59 am

    @Andreas Johansson:

    I can't think of any English word begining in /zl/.

    Zloty is pronounced that way, though in Polish it's spelled "złoty" and pronounced with initial /zw/. Again, not a common word, somewhat analogous to the /ʃl/ words from Yiddish.

  31. Tim Morris said,

    January 13, 2018 @ 12:02 pm

    Coke to Jarek Weckwerth :) I was also going to mention Italian initial clusters like /sb/ and /sd/ (sbirro, sdraiato), though in practice English speakers unvoice the stop consonants in them, and many Italians may do so too, for all I know.

  32. Robert Coren said,

    January 13, 2018 @ 12:08 pm

    @Tim Morris: I was deliberately ignoring borrowings like "zloty" and "Vladimir".

  33. jih said,

    January 13, 2018 @ 12:09 pm

    Historical inheritance explains the scarcity of certain onset clusters, but not their complete absence, since there are other sources of words in the lexicon of any language. Iberian Spanish, just like other Romance languages and English, has no words starting with /tl/, but Mexican Spanish has quite a few of them (from Nahuatl). A couple of toponyms with this initial sequence are Tlatelolco and Tlaxcala. There are also common nouns with this sequence. At least one of them, tlacoyo, is found in the menu of some Mexican restaurants in the US; so I guess it could become an English word as well.
    Are there other languages with a /θl/ thl- sequence from which English could borrow words?
    New words can also arise more-or-less out of thin air (googol > google, quark, gizmo). why hasn't anybody given a name like "thlenk" to a new gizmo or "Thlamoco" to a corporation?

  34. Jarek Weckwerth said,

    January 13, 2018 @ 12:14 pm

    @ Tim Morris: Thank you. Coke as in processed coal, of course, is it? ;)

    Anyway, in the UK you'll find that zloty is not that uncommon with all the Polish people, but Zlatan in particular is not unknown at all as one of the top footballers of the last decade. So, as ~flow was saying above, /θl/ may just need someone famous.

    And, now that I think of the UK, the treatment of Welsh ll names as /θl/ is quite usual in fact. Wells's pronunciation dictionary has 20 of those initially (albeit never as the first option).

  35. Joyce Melton said,

    January 13, 2018 @ 12:34 pm

    RP said: (whereas "presume", for exampe, varies between /ʒ/ and /zj/, with the latter considered more correct in prescriptivist quarters).

    Say what? I've never noticed either of those pronunciations, though I can imagine a British speaker using the second so I may have heard it and just not noticed. Most, if not all, Americans would use a simple /z/ in that spot.

  36. Zeppelin said,

    January 13, 2018 @ 12:39 pm


    Another point regarding (lack of) *skl- in English: *sk- was palatalised to sh- in English, which means all words beginning in sk- are loans from other languages (compare native English "ship" with borrowed "skiff" and "skipper", native "shirt" but the Norse loan "skirt", and so on).

    So we'd need a foreign word beginning with *skl- (rare or possibly absent to begin with in PIE, so the *skl- would have to be secondary) to be borrowed into English, but only after the sk>sh shift was completed. The odds just aren't very good.
    And in the one case I can think of where such a word was borrowed, the cluster was just simplified to *sl- — that's what happened with "slave".

  37. Jarek Weckwerth said,

    January 13, 2018 @ 12:41 pm

    @Joyce Melton: John Wells: — Preference poll, British English: -ˈzjuːm 76%, -ˈzuːm 16%, -ˈʒuːm 8%. (This is what people reported to him in writing, so it has to be taken with a massive lump of salt.)

  38. Zeppelin said,

    January 13, 2018 @ 12:42 pm

    Oh, and the same thing happens word-internally of course — see "muscle" (which I just learned is actually the same word as "mussel", etymologically, so we even have an ortographic doublet based on this sound change!)

  39. Jimbino said,

    January 13, 2018 @ 1:42 pm

    I hear few people pronounce "amphitheater" correctly. I wonder how Bob Moore does with "phthalate"?

  40. jih said,

    January 13, 2018 @ 2:01 pm

    Words of Greek origin provide for a very interesting comparison, I think. I suppose English speakers who are familiar with the word thlipsis (cited by ˜flow) pronounce it with [θl-]?
    Compare with psycology, xerox, etc., where other word-initial clusters suggested by the spelling are systematically simplified.

  41. Sally Thomason said,

    January 13, 2018 @ 2:59 pm

    Tom Dawkes is surely right in his statement that the "thl" spelling is not a consonant cluster at all but rather a single phoneme/speech sound — either a voiceless lateral fricative (as in Welsh Lloyd) or a lateral affricate, glottalized or not. Lateral fricatives & affricates are common throughout the Pacific Northwest, though Tlingit is a more likely source for an Alaskan name than are any of the Wakashan, Salishan, and Chemakuan languages spoken in southern BC and Washington state. Not that this is relevant to the discussion of possible and actual English consonant clusters, of course; it's only relevant to the issue of how to pronounce that name.

  42. Bob Ladd said,

    January 13, 2018 @ 4:09 pm

    @Tim Morris: actually, in Italian, those initial clusters sb- and sd- (and sg- as well) are pronounced all voiced, i.e. /zb, zd, zg/.

  43. Bill said,

    January 13, 2018 @ 5:57 pm

    "Sklodowskite" is a uranium mineral named after Maria Salomea Skłodowska (Marie Curie).

  44. Jarek Weckwerth said,

    January 13, 2018 @ 6:48 pm

    @Bill Skłodowska is of course /skwɔˈdɔfska/ in Polish ;)

  45. Jarek Weckwerth said,

    January 13, 2018 @ 6:50 pm

    But then there's sclaff.

  46. Jarek Weckwerth said,

    January 13, 2018 @ 6:55 pm

    Oh my. Collins also lists
    scliff, and
    Sheer fun.

  47. Lazar said,

    January 13, 2018 @ 9:01 pm

    @Robert T McQuaid:

    English phonics includes the ts sound, as in the final consonant cluster of cats.

    Strictly speaking, it doesn't. /ts/ as found in Japanese or German is an affricate, which differs both phonologically and phonetically from the biphonemic sequence /ts/ in English "cats". English phonotactics allows this sequence at the end of a syllable, but not at the beginning of one.

  48. Jonathan Smith said,

    January 13, 2018 @ 9:05 pm

    Onset /skl/ is given in descriptions of Irish and Scottish Gaelic, which could account for some of the above "English" words… but don't know about its origins. Some of these words (sclate, sclave) seem to reflect word-medial configurations in Old French for instance…

  49. Bloix said,

    January 13, 2018 @ 11:39 pm

    Not sure why the focus on thl-, when tl-, dl-, shl-, vl- are also unused.
    I mention these (not jl-, hl- , ml-, etc) because these consonants can be followed by R. You'd think if an initial consonant can take R it would take L as well.

  50. John Swindle said,

    January 14, 2018 @ 6:51 am

    Just guessing here, but it might be that in Tlingit weaving, too, not every possible sequence is instantiated.

  51. Kimball Kramer said,

    January 14, 2018 @ 8:59 am

    I'd like to point out to rosie: The initial "mw" sound is often used in English. Listen for the word "mwa" said during a kiss of the cheek to indicate the lack of any sexual content.

  52. Adrian said,

    January 14, 2018 @ 10:41 am

    Interesting discussion. I noticed recently a placename "Vandlebury" on an old map and surmised it was a mistake. I was right (the correct name is Dandlebury) and realised my suspicion was aroused by the fact that initial v in English placenames is rare, except in certain parts of Wales (where it comes from Welsh f) and the South West (where f is often voiced). Of course, words in v- are relatively rare in English generally, for some reason.

  53. Robert Coren said,

    January 14, 2018 @ 11:03 am

    @Bloix: Really, vr-? The only English word I can think of with initial /vr/ is the onomatopoeic vroom.

  54. Bill said,

    January 14, 2018 @ 12:00 pm

    vrddhi, vriesea, vrouw, vrykolakas, vraic, vraicqueur, vraisemblance to name a few…

  55. Bloix said,

    January 14, 2018 @ 1:02 pm

    Yes, I was thinking of vroom. I think one example that's not a loan-word is enough to show that the cluster is possible. But if you want to delete vr- from the list of clusters we do have, that works for me. And you can delete shl- from the list of clusters we don't have, if you want, given all the Yiddish or Yinglish borrowings. It's still odd that we have tr- and dr- but not tl- and dl-.
    And why do we have fl- but not vl- when the only difference is that V is voiced? There can't be any genuine difficulty in saying one that doesn't apply to the other, can there?

  56. Lazar said,

    January 14, 2018 @ 4:07 pm

    It's still odd that we have tr- and dr- but not tl- and dl-.

    That seems like a common constraint in Standard Average European; the lateral release required for /tl/ and /dl/ really sets them apart from other stop-liquid sequences. Mexican Spanish is something of an exception to this, having taken on [tl] as an adaptation of Náhuatl [tɬ] in words like tlapalería.

    And why do we have fl- but not vl- when the only difference is that V is voiced? There can't be any genuine difficulty in saying one that doesn't apply to the other, can there?

    Not intrinsically, I don't think, but English has an aversion to voiced fricatives at the starts of clusters. More so in some cases than others: no one has trouble saying Vladimir, but on the other hand the [zb] in sbarro isn't just difficult but even incomprehensible to an English-speaking layman.

  57. Lazar said,

    January 14, 2018 @ 4:13 pm

    One interesting related case is zhlub or s(c)hlub. I learned it with [ʒl], but I more often hear [ʃl]; I think the former lies just on the outer bounds of "usable but unused" in English phonotactics.

  58. Christian Weisgerber said,

    January 14, 2018 @ 6:30 pm

    Old English did not distinguish voicing in its fricatives /f/, /s/, /θ/. They were realized as voiceless [f], [s], [θ] in initial and final position, but as voiced [v], [z], [ð] between vowels. Only in Middle English did the fricatives split into separate voiceless and voiced phonemes.

    This original distribution is still visible in English today:

    Initial /z/ is rare and limited to loans and onomatopoeia.

    Initial /v/ is not rare, but almost all instances are French~Latin loanwords. Only a handful of native English words begin with /v/ (vane, vat, vixen) and these are due to influence from southwestern dialects where initial fricatives were voiced (cf. Dutch).

    Initial /ð/ somehow appeared in function words, but in content words, initial th is still /θ/.

    The Old English fricative distribution also underlies the knife ~ knives alternation and explains various observations made in earlier comments.

  59. stedak said,

    January 14, 2018 @ 11:40 pm

    About the name Thlunaut, Tom Dawkes and Sally Thomason are right: it's a Tlingit name and the initial sound is the voiceless lateral fricative [ɬ], which sounds a lot like the cluster [θl] to Anglophones. The weaver Jennie Thlunaut took that spelling for her surname in the early 20th century, when there was no phonetically accurate orthography for Tlingit yet. In the current Alaskan orthography, the name would apparently be spelled Lunaat', using L to spell [ɬ]; since they only have the fricative [ɬ] and not the approximant [l] as in English, they're free to use the letter L to spell their sound.

    Another orthography note: many web pages have a problem with Tlingit words since the orthography uses underscores on g, k, and x to represent uvular consonants, and many fonts can't handle that. According to Wikipedia, Tlingit e-mail users have devised a workaround using digraphs: gh, kh, and xh.

    And regarding the same lateral fricative in Welsh, there's a long history of English-speakers using clusters "kl" or "fl" to spell it, e.g. Lloyd turned into Floyd. There's a passage in Alan Garner's The Owl Service where an English girl reading the story of Lleu Llaw Gyffes pronounces it "Clue Claw". Sociolinguistics also plays a major role in this novel: the Welsh boy knows he won't be able to escape his lower-class background unless he can get the Welsh accent out of his speech. (There's a topic I'd love to see on Language Log: Greatest novels where sociolinguistics plays a major role.)

  60. Marisa Brook said,

    January 15, 2018 @ 2:09 am

    Not to overlook the whole /ɬ/ thing, but…

    I'd say we're seeing roles here for two synchronic phonotactic constraints in PDE. One is sonority-contour, as Mark mentioned in the post: there needs to be a dramatic leap in sonority from C1 to C2 in English (disregarding sibilants because those are phonotactially strange in English as in a lot of other languages). The other is referred to as homorganicity or 'homorganicity of place': roughly speaking, English doesn't like it when the C1 and C2 are pronounced too close together (Clements and Keyser 1983:43-44; Rice 1992). Admittedly, for this to work you have to propose that /r/ is unspecified for place (Rice 1992:76) or something along those lines. But it potentially accounts for the weirdness of */pw/, */bw/, */tl/, */dl/, and even */θl/ (interdental + alveolar) in spite of their entirely acceptable sonority-contours. The cluster /pw/ is found only in a few loanwords in PDE as far as I know ('pueblo', 'puissant'). Same for /bw/ ('bueno'). /tl/ and /dl/ and /θl/, if encountered in words/names from other languages, tend to be modified in both perception and production by English speakers (e.g. insertion of an epenthetic schwa) (see Dupoux et al. 2001, Berent et al. 2007, etc.).

    Clements and Keyser (1983:41) say this about the 'marginal' cases in English two-consonant onset clusters (/vl/, /vr/, /sv/, /zw/, etc.): "We believe that the traditional confusion regarding whether clusters such as these should be accounted for in a general account of the English syllable reflects their intermediate status between fully acceptable and fully deviant clusters."

    Disclaimer: I am a sociolinguist. But I do a little bit of laboratory phonology in my spare time, and when that occurs, it's 'grey area' phonotactics (in any language) that interests me the most, so this is up my alley. I enjoy the occasional opportunity to make a non-linguist with English as their L1 notice how normal-but-unusual it is to have /dw/, /gw/, and even /θw/ clusters at the beginnings of words ('dweeb', 'dwarf', 'Dwayne', 'Dwight', 'Gwendolyn', 'thwart', 'thwack'). If anyone has any thoughts about diachronic reasons underlying those particular gaps, I'd be very curious!

  61. R. Fenwick said,

    January 15, 2018 @ 3:19 am


    As an initial cluster, I don't think "spw", "shp" or "tsr" exist natively in English.

    On the first instance, native English vocabulary doesn't generally allow initial clusters of two labials of any kind: *pf-, *bw-, *vw– (with some loaned exceptions like pfennig, bwana, voilà). I think it's an even more general linguistic aversion, since even among those languages with contrastive consonantal labialisation, labialised labials are vanishingly rare (the only counterexamples I can think of are West Circassian and Arrernte). And on the third, native English vocabulary also doesn't allow initial [stop+fricative] clusters either: *tf-, *ks-, and so on (though plenty of other languages do: ancient Greek, not least, with the dedicated letter Ξ/ξ doing service for that cluster).

    Now, the second one is more interesting: shp– [ʃp-]. This one is mostly illegal because the fricative [ʃ] only arose in English from palatalisation of other consonants in contact with a front segment. English ship, for instance, came from the same root as the Norse loan skiff, but the initial sk– was palatalised to [ʃ] before the vowel –i-. In sp-, though, the p– blocked this palatalisation from happening, leaving s– intact. The result is that in native English words, there are no initial clusters of [ʃ]+consonant of any sort: we have spstslsmsn-, but no *shp– *sht– *shl– *shm– *shn-.

    The thing I find really interesting about this situation, though, is that since [ʃ] has become a phoneme in English, it's allowed us to take on a number of loans from German and Yiddish (where s– became ʃ– before all consonants) that now furnish English with virtually the full set of [ʃ]-initial clusters: spiel, shtupp, schlep, schmuck, schnozz.

  62. Andreas Johansson said,

    January 15, 2018 @ 4:04 am

    @R. Fenwick

    Shr- as in "shrill", "shrew", etc. is native, no?

  63. Christian Weisgerber said,

    January 15, 2018 @ 7:33 am

    @R. Fenwic

    Old English shifted [sk] to [ʃ] in all positions (as did German, but Dutch differs). The few sk words of Anglo-Saxon origin (ask, dusk, tusk) are the result of metathesis.

  64. Robert Coren said,

    January 15, 2018 @ 10:37 am

    @Bill: I would not consider any of those vr- words to be English.

    @Christian Weisgerber: I would not agree that zoo (and all the other zoology derivatives) and zero are loans, and they're certainly not onomatopoeia.

  65. Lazar said,

    January 15, 2018 @ 4:39 pm

    @Robert Coren: Yeah, from "vrddhi" in a linguistic paper perhaps, I've never seen any of those words in an English text.

    @Andreas Johansson: The distinction between /.ʃɹ/ and /.sɹ/ is basically neutralized in English, I think. I say "Sri Lanka" with [ʃ], so there's really nothing in my lexicon that has [.sɹ]; conversely, there are some varieties in the American South (or at least so I've read) that use [sɹ] in the English "shr-" words.

  66. Lazar said,

    January 15, 2018 @ 4:43 pm

    @Christian Weisgerber: I'm a little confused. I thought /sk/ was the older configuration in "ask", going back to PG "*aiskōną".

  67. ajay said,

    January 15, 2018 @ 5:00 pm

    there's a long history of English-speakers using clusters "kl" or "fl" to spell it, e.g. Lloyd turned into Floyd.
    Or (presumably) Llewelyn into Fluellen in Henry V.

  68. Lazar said,

    January 15, 2018 @ 5:11 pm

    @ajay: Not to mention the adaption of "tlhingan" as "Klingon" by 22nd-century human spacefarers.

  69. Christian Weisgerber said,

    January 15, 2018 @ 6:01 pm

    There were competing forms ascian and acsian already in Old English. You would expect ascian to evolve into ash, which is apparently an attested dialectal form. Meanwhile acsian produced ax, which was an accepted (some say the dominant) literary variant until Early Modern English when ask took over again. It looks like this word suffered metathesis several times in the history of English.

  70. JPL said,

    January 15, 2018 @ 6:56 pm

    Just want to toss in the example that occurred to me immediately (although not word- initial, it could be considered syllable- initial), and after 69 comments (69 on this topic?) seems to be still not tossed in: 'athlete' (a- thlete). (The same observation applies to the name 'Kathleen' given by Joshua Gibson above.) (And although Tom Dawkes et al. have detached the original example, as a voiceless lateral (fricative), from the question under subsequent discussion)

  71. Lazar said,

    January 15, 2018 @ 7:43 pm

    @Christian Weisgerber: Oh, cool. I knew that both forms coexisted in OE, but I didn't really know about the details that you metnion (e.g. "ash").

    @JPL: Eh, I wouldn't consider either of those to be syllable-initial. Cf. "Atlantic", where the /t/ definitely seems to undergo the lenition you'd expect at the end of a syllable.

  72. Bob Moore said,

    January 15, 2018 @ 8:10 pm

    Thanks to @stedak for confirming the origin of the name "Thlunaut". I would defend my English pronunciation in terms of the consonant cluster "thl" by the fact that the Tlingit people I have talked with about Jennie Thlunaut, seem to my ear the be using "thl" to pronounce her name when they are speaking English, whatever they might say in Tlingit. Similarly, when mentioning the name of their own people, many, if not most, say "klinkit" in English, just as most of us non-natives do, although they surely do not pronounce it that way when speaking their own language.

    In any case, as Sally Thomason mentioned, the name "Thlunaut" was simply what sparked my musing about consonant clusters in English, which has nothing to do with how that name should be pronounced in either Tlingit or English.

    I would like to close with an explanation of why I am interested in Jennie Thlunaut. She was at one point the only surviving practitioner of the art of Chilkat weaving, which came to be called that because it was practiced by the Tlingit of the Chilkat Valley north of Haines, Alaska, after it had died out everywhere else on the Northwest Coast, including northern British Columbia, where it originated among the Tsimshian people.

    Fortunately, Jennie realized that this art was in danger of dying with her, so in her later life she made it her mission to teach what she knew, so that now there are enough Chilkat weavers to keep it alive, all of whom are her artistic descendants. There is a small, but wonderful exhibit at the Portland Art Museum right now of weaving by four generations of Tlingit weavers, including Jennie and her teacher Cora Benson. See and also the documentary film "Lineage: Tlingit Art across Generations" ( featuring several of the weavers whose work is shown in the exhibit. I should also mention for the U Penn crew that the Penn Museum has an outstanding collection of Chilkat robes, although I am not sure whether any of them are currently on display.

  73. John Swindle said,

    January 15, 2018 @ 11:57 pm

    "Atlantic" doesn't get a schwa or a vocalic "l" inserted between the first and second syllables. "Athlete" often does. Neither "ath-leet" nor "ath-ul-leet" has a "thl" cluster, but couldn't the latter be a way of breaking up what's perceived as a difficult or impossible consonant cluster?

  74. John Walden said,

    January 16, 2018 @ 3:15 am

    There's some subjectivity with 'easy-to-pronounce' . Spanish speakers have difficulty with what are to English speakers unremarkable initial clusters like 'Spain' and 'star' and have great difficulty with 'breakfast', which is mispronounced with an unfortunate result.

  75. Rodger C said,

    January 16, 2018 @ 8:26 am


    My mother in fact always said "srimp" and "srunk."

    Another Appalachianism/ Southernism is "tushes" for "tusks."

    "Tlhingan" as the original of "Klingon" is, I presume, a post facto explanation, but its a danged clever one.

  76. BZ said,

    January 16, 2018 @ 1:04 pm

    For all the people saying there shouldn't be a difference between tr and tl, I am a Russian immigrant of 27 years. I have almost no accent, but tr trips me up. When I try to say it, it either comes out sounding like "ch" or I roll the "r" as in Russian. This despite tl being a common Russian cluster that sounds exactly the same as in English.

  77. AnonMoos said,

    January 20, 2018 @ 5:16 am

    "stw" is kind of a gap among word-initial clusters in English.

RSS feed for comments on this post