Why is the L silent in "walk" but not in "bulk"?

TL;DR

Why is the letter L silent in walk, talk, calm, folk, half, chalk etc but not silent in bulk, hulk, milk, silk, bold, bald?

Explanation of the question and Research:

The letter L seems to be silent in many words like walk, talk, calm, folk, half, chalk. I checked their pronunciations in a few dictionaries and I have never heard anyone pronouncing it in these words. Maybe some people pronounce the L as suggested in this post on reddit but the standard pronunciation has silent L.

Another post on Quora suggests two points in which one is "l-vocalization":

The first one has got something to do with l-vocalisation. In many languages this consonant changes its characteristic over time. Initially, plain ’l’ gets velarized (or dark, in other words), after which the consonant gradually becomes a semi-vowel.
This is what happened to ‘l’ in a number of English words of Germanic origin such as ‘walk’. Being pronounced /walk/ a long time ago, they changed the pronunciation to /wawk/, and later on to /wɔ:k/ (monophthongisation of ‘au’ is also rather widespread). [answer by Igor Vegin].

One question (Where did the L in talk go?) on this site is similar but it does not explain why the L is pronounced in some words and silent in some words. In this question, the answer by Decapitated Soul says also "L-vocalization" as suggested in Quora's answer but does not explain my question.

Another answer I found is on Linguist List but does not answer my question, instead it says the the L was pronounced at some time but it is silent now.

My own observations:

The L is silent when it comes between a vowel and consonant. In words where the L is between vowels, the L is not silent like in filling, killing, color, pillar etc.

The L after /ɔː/ and /ɑː/ and before a consonant is silent in many words like calm, walk, talk, half, calm etc.

The L after /ɪ/ and before a consonant is not silent like milk, silk, film etc.

The L after /ʌ/ and before a consonant is not like hulk, bulk etc.

Question:

Is there any rule or this is just random discrepancy? Can anyone please explain?

Solution 1:

Based on the pairs chalk, bald and chalk, milk, the identity of both the consonant following the L and the vowel preceding the L could affect whether the L was lost. I'm not certain why, but my guess is that it's related to a more general pattern where only a restricted number of vowel sounds are permitted before syllable-final consonant clusters ending in labial consonants like /p b f v m/ or velar consonants like /k g/.

K is a non-coronal consonant, while D is a coronal consonant

It might seem odd to group all of the above consonants together, but they all can be categorized as "non-coronal consonants". Coronal consonants are the class of consonants made with the front part of the tongue: dental or alveolar consonants such as /t d n s z θ ð/, and also the postalveolar consonants /tʃ dʒ ʃ ʒ/. The most common consonant sounds in English fall in the category of coronal consonants, and many non-syllabic suffixes, such as /d/~/t/ and /z/~/s/, consist of a coronal consonant.

For whatever reason, many English vowel sounds can be found before a consonant cluster in syllables ending in a coronal consonant, but cannot be found before a consonant cluster in syllables ending in a non-coronal consonant.

For example, syllables ending in /st/ can have a variety of diphthongs or long/"tense" vowel sounds: Christ /aɪst/, toast /oʊst/, oust /aʊst/, taste /eɪst/, hoist /oɪst/, east /iːst/ roost /uːst/. In contrast, syllables ending in /sp/ or /sk/ are not found with diphthongs like /aɪ/, /aʊ/, /oɪ/, /oʊ/ or /eɪ/, and are only occasionally found with monophthongs that can be considered "long" such as wasp and (one pronunciation of) Basque.
Syllables ending in /nt/ can be found with /aɪ/ (pint) /aʊ/ (count) or /eɪ/ (paint); syllables ending in [mp] or [ŋk] cannot (except for in some accents with a sound change of /æ/ to /eɪ/ before [ŋ]).

So I think there may actually have been a two-part sound change in Middle English, rather than an immediate vocalization of [l] to [w~ʊ].

First, a and o before a consonant cluster starting in [l] were diphthongized to something like [aʊ] and [oʊ], respectively. This would create Middle English forms like [waʊlk], [saʊlt], [foʊlk], [boʊlt] from earlier [walk], [salt], [folk], [bolt].
Then, to get rid of the ill-tolerated sequence of a diphthong followed by [lk] (a consonant cluster ending a non-coronal consonant), the consonant cluster was simplified by dropping the [l] (at least in some accents), resulting in [waʊk], [saʊlt], [foʊk], [boʊlt]. But the consonant clusters [lt] and [ld] were maintained after diphthongs because they end in coronal consonants.

In Modern English, the Middle English diphthong [aʊ] turned into a monophthong with different values in different accents. (The Early Modern English value is sometimes reconstructed as [ɒː].)

Vowels other than a and o, such as in bulk, hulk, milk, silk, elk, were apparently not diphthongized in this context, so the consonant cluster [lk] remained intact.

words with "l" followed by a labial consonant

Something similar would have occurred with [lf] [lv] [lm]. As with /lk/, we see no simplification of these clusters after e or i: elf, shelf, shelve, elm, helm, film and so on contain /l/. Wolf represents Old English wulf with respelling of wu- to wo-.

Words with "al" followed by a labial consonant show complicated developments. In some accents, such as British "Received Pronunciation", the al in words like half, halve, calm is pronounced as /ɑː/. This is not the usual outcome of the Middle English diphthong [aʊ], and it suggests that these words might have undergone an additional sound change, like [kalm] > [kaʊlm] > [kaʊm] > [kɑːm]. A change of [aʊ] to [aː] in this context might have been motivated by the fact that [ʊ] and labial consonants are both rounded, so the roundedness of [ʊ] might have been misinterpreted by a listener as just being part of the following consonant sound.

In American English, half and halve typically have /æ/, which could be the result of simplification of [aʊf] to [af] without lengthening. Something similar may have occurred in words such as laugh(ter) and draught/draft. Calm has a large variety of realizations in American English, some of which include an /l/.

For some reason, the word scalp (with p) shows /ælp/ in Modern English. I'm not sure how to summarize other words ending in -lp such as palp, gulp, pulp, poulp.

Other words I don't know how to summarize: gulf, golf; -olv- words like solve revolve etc.

words with "silent l" followed by a coronal consonant are irregular

Tchrist's answer mentions a few words spelled with LD that are pronounced without an [l] sound: could/should/would and for some speakers solder. I don't think any of these are related to a regular sound change.

Solder, a word with many pronunciations (including some with [l]), comes from French which had its own sound change of l-vocalization that applied before any consonant. The L was restored in the spelling and for some speakers in the pronunciation.

Should and would are probably cases of irregular loss of a consonant in a commonly used function word. For comparison, the consonant [k] has been lost in the past tense made of the common verb make; this is not part of any general pattern of [k] being lost. Could is an unetymological spelling by analogy with the spelling of should and would.

Solution 2:

_{Please Note}

_{In the following preamble leading up to the explanation, I am self-reporting how this all works for me personally so you can see where I’m coming from. I recognize that not everyone around the world talks the way I do. I am a native speaker
with the Inland North dialect of North America who was raised in the long lost rural southeastern Wisconsin of half a century ago before mass media inundation. I therefore have
the father–bother merger but not the cot–caught merger.
I also lack the Northern Cities Vowel Chain-Shift found in younger, more urban speakers from that region.}

Usually we keep the L

When it both follows a stressed vowel and also precedes another consonant in that same syllable, phonemic /l/ is normally retained in pronunciation compared with the written form.

With one exception, it does not matter whether this is a tense vowel or a lax one, nor whether it is a phonemic diphthong (or even a phonetic diphthong, a notion I’ll ignore here for clarity).

/ɑl/ — (none, because the stressed FATHER vowel never occurs before /l/ for me in the coda)
/æl/ — Alps, calc, calque, talc, scalp, shalt, Ralph, valve
/el/ — paled, baled, scaled, whaled, veiled, bailed, hailed, nailed, failed
/ɛl/ — Kelt, geld, belt, melt, smelt, meld, knelt, help, elk, whelk, elf, shelf, Guelph, delve, shelve, meld, realm, helm, wealth, twelfth, twelve, shelve, squelch
/il/ — field, shield, kneeled, yield
/ɪl/ — guilt, guild, built, build, bilge, kilt, killed, filled, filch, silk, milk, wilt, milled, film, filmed, kiln, sylph
/ol/ — colt, cold, hold, molt, mold, volt, told, sold, souled
/ɔl/ — Walt, waltz, fault, faults, salt, exalt, called, walled, galled, bawled, bald, golf, solve, alms, calm, calms, calmed, palm, balm, hauled, trawled, shawled, brawled, drawled, mauled, salsa
/ul/ — coolth, ruled, pooled, cooled, fooled, fuelled
/ʊl/ — pulled, hulk, sulk, wolf, wolves
/ʌl/ — cult, sculpt, skulled, ?skulk, sultan, nulled, hulled, bulge, gulf, false, pulse, bulk, bulb, gulled, gulp, gulped, culled, lulled, mulled, indulge
/ɑul/ — cowled, jowled, fouled, howled
/ɑil/ — wild, mild, tiled, smiled, styled, aisled, vialled, trialled
/oil/ — spoilt, oiled, toiled

But not quite always

However, in a few words only, mostly common ones, the /l/ phoneme is completely lost in that position compared with the written form:

/ɑ/ — solder (because I have the FATHER–BOTHER merger)
/æ/ — calf, half, calves, halves, salve
/o/ — yolk, holm, ?folk, ?Polk
/ɔ/ — talk, walk, balk, caulk, chalk
/ʊ/ — could, would, should

As you see, the words where the /l/ has “gone missing” from the pronunciation are mostly words with lax vowels not tense ones there, with yolk being the principal exception still in common use. Words like folk and Polk often lose the /l/, but not necessarily always.

So what’s really happening here?

There do appear to be patterns that could be detected. For example, notice how there are no occurrences of /el/ or /il/ followed by an unvoiced stop. Or how kneeled with a tense vowel becomes knelt with a lax one because the unvoiced stop checks the earlier vowel despite the intervening liquid.

The thing to remember here is that when we write phonemic /l/ — or /r/, or for that matter many other phonemes — we have deliberately clumped together any number of allophones that are all different sounds, but we’re trained not to hear them as different because they never result in a different word depending on which allophone shows up somewhere.

Consider let /lɛt/ and tell /tɛl/. Those two words have the same three phonemes; they merely swap the pair on the front and back with respect to each other. But when you do that, you have to select different allophones. You are saying the same phoneme but not the same sound.

The /t/ in tell is actually [tʰ], but the /t/ in let is just [t] or [ʔ]. And while for most speakers the /l/ in let is usually the voiced alveolar lateral approximant [l], the /l/ in tell may not be; it might instead be velarized alveolar lateral approximant written [ɫ] or [lˠ].

When you velarize the /l/ in the coda, it seems to get a bit mushy. When there’s a front vowel in front of it, as there is in tell, tale, teal, till, it may not velarize, or not much. Velarization happens in the back of your mouth. So when there is a back vowel in front of it, as there is tall, toll, pool, pull, that velarization is stronger. It even changes the low back vowel [ɑ] to the open-mid back vowel [ɔ] as in all, tall, fall, ball, call, Paul; almond, almost, balsam, balsamic. That even happens when I use the Spanish loanword salsa in English, where after naturalization from the original [ˈsalsa] in Spanish, it becomes [ˈsɔɫsə] like Saul or even [ˈsʌɫsə] like Tulsa.

The only word I know that retains an unmodified FATHER vowel [ɑ] there before final L is an import from India, the fiery dish variously spelled phaal or phall which is invariably pronounced [fɑl] or [fɑɫ], quite unlike how we pronounce our own word fall as [fɔɫ].

But you weren’t asking about words that end in ‹l›. You wondered about those that had ‹l› followed by another consonant. If that final consonant is a velar one, meaning unvoiced /k/ or voiced /g/, then the only chance you have to retain the dark, velarized ‹l› preceding it is if the vowel just before that is in the front of your mouth. That’s why we still have an /l/ in milk, whelk: it’s because /ɪ/ and /ɛ/ are front vowels.

But with a back vowel before your velarized ‹l› and velar consonant following it, your mouth has no chance to produce any kind of distinct /l/ sound. Hence its disappearance in talk, walk, balk, caulk, chalk, folk, Polk. Interestingly, there are no words in English that end in the voiced version, /lg/.

The words where the /l/ is most often lost in many but not quite all speakers are just that sort of combination, like folk and hulk. I also know speakers who’ve lost it in wolf, golf but those are back vowels anyway, which destabilize the L even with a front consonant to follow (as long ago happened to could, would, should).

So what’s the excuse for losing it in half, halves, calf, calves, salve, given that those have a front vowel /æ/ before the L and a front consonant following it, especially since talc, calque, elf, elves preserves it?

I don’t know. Maybe it’s like the people who’ve lost it in wolf and golf. But I do wonder whether it doesn’t have its origin in the TRAP–BATH split, where speakers who distinguish those two vowels have a front vowel for TRAP but a back vowel for BATH. If it was a back vowel when the L was lost, that might explain it.