The Alphabet. 

Latin vowels ( A, E, I, O, U) are divided into long and short ones. Letter “I” stands for two sounds: a vowel (eg. “videre”: i as in “sick”) and a consonant (“maior”: i as y in “surveyor” ). The same applies to letter “U”: sometimes it is read as a vowel (“lugere”), sometimes (after Q and, in many cases, G) it resembles a quick English “w” sound (“equus”, “sanguis”) and is counted as one sound combination with the letter preceding it. The symbol “U” also designates a purely consonantal sound “v”, which on the lips of the Romans had the sound of w, such as in the word “valere” (N.B.: in ancient times both U and V were expressed by way of the same symbol V. In like manner, J was unknown). As has been mentioned, all of the five vowels can be long or short. There also exist in Latin a number of diphthongs: ae (Eng. “eye”), oe (“boy”), au (“owch!”), eu (not present in English). Their quantity is always long. The arrangement of long and short vowel sounds within a word determines on which syllable accent falls: if the second from last syllable is long, it bears the accent; if it is short, the third-last syllable is stressed (The Penultimate Law). Apart from being long or short by nature, vowel sounds can change their quantity depending on whether the syllable they are in is open (followed by 1 consonant or none) or closed (followed by 2 or more consonants). A closed syllable is said to be long by position (whatever its nature), even if the vowel inside it is short. Some complication arises with the palatals “L” and “R”, which can leave the length of a syllable unchanged when they figure as the second consonant following a vowel sound (we encounter both short and long A in “sacra”; however “LL” and “RR” always make a closed syllable). When either “R” or “L” precede a consonant, the vowel enclosed is invariably long. 

Practical significance of the syllable rules lies especially in the poetic domain. For the most part Latin verse is syllable-length- (vowel-quantity)-based, as opposed to the modern beat structure in many (western at least, to my knowledge) languages. Thus, the stress described by the Penultimate Law (see the above paragraph) has virtually no bearing on the diction of Latin poetry where natural accentuation must always yield to the requirements of the meter ( I invite discussion on the subject of tonal pronunciation in Latin and present-day efforts to recite Latin verse; I would especially like to hear from someone speaking a language that preserves quantity differentiation of syllables). 

Something else needs to be said in addition to the above concerning the stress-definition problem in Latin. Most linguists agree that there existed a secondary accent on the first syllable of a word, and that originally it had been the primary one. This seems plausible in the light of the otherwise apparently inexplicable and strange feature of even stressed vowels (if we are to rely solely on the Penultimate Law) to undergo significant transformations which reveal the same processes as are common to unaccented syllables. This phenomenon can be referred to as vowel-degradation (more on this follows below). 

As centuries went by, Latin accentuation changed dramatically. The seeds for the destruction of the notion of syllable quantity (vowel-length) in the language were sown ironically by the very Penultimate Law which itself vitally depended on quantity differentiation! As soon as stress settled on a particular syllable in the Classical period, the accent's position began to affect the length of its own syllable (by lengthening it in violation of true etymology) and above all, that of its neighbors (by shortening them, since they now received less tonal emphasis). Such change was especially remarkable in popular speech, which a couple of centuries into the new era lost all idea of vowel quantity. Thus, the later language which itself soon passed smoothly into the Romance tongues, possessed the kind of accentuation now prevalent in the West (stress based as compared to quantity based). 

Vowel Intensification: 

I have purposely “pillaged” some authoritative works, and made a resume of the most notable characteristics of Latin phonetics. I am especially obliged to John Peile’s “Introduction to Greek and Latin etymology”, a compressed and comprehensive guide. I shall start with vowel intensification (the technique of conveying different notions about the state of an action (its completeness) to the same root by modifying its vowel, which our forefathers accomplished by passing a stronger breath, the “a” sound, before the main vowel, thus turning it into a diphthong) and compare its course in the original language and in Latin. There are 2 steps of vowel intensification, the principle remaining the same for both (adding a breath in front of the vowel in the first step, and yet another one in the second). In the first column are the 3 original IE vowels (A, I, U), and each row illustrates the processes relating to that vowel scale. The dot (.) after a vowel signifies its long character. 
 

Scale \ Steps	Step 1: IE	Step1: Latin	Step 2: IE	Step 2: Latin
A	a + a = a.	e	a + a. = a.	o
I	a + i = ai	ei (ai)	a + ai = a.i	oi
U	a + u = au	eu (au)	a + au = a.u	ou

 
Diphthong Simplification: 

The Latin diphthongs obtained in the table were themselves susceptible to change from an early date. Below are their usual transition paths. 
 

The diphthong:	What it changed to:
ai (later written ae)	e. and often reduced to i.
ei	i. in literature, e. in the country & common folk
oi (later written oe)	u. though also ei (-> i.) in terminations
au	o. also rarely u. (accu.so, clu.do)
eu	u.
ou	u.

Note: ae, oe, au were preserved unchanged longer than the others. 

Vowel Change: 

Simple vowels also possessed a tendency to become simplified. This is referred to as vowel-degradation. This change occurs mostly in unstressed syllables. The steps for each vowel are: 

A -> o -> u -> e -> i 

O -> u -> e -> i 

U -> e -> i 

E -> i , e 

I -> e 

In fact, vowels in Latin did not play as important a role as they did in ancient Greek, and were more freely allowed to decay. However, as we shall see below, Latin was more tenacious with respect to its consonants, and this is its counterbalance to the Greek language, where neglect and willful avoidance of interference with the harmony of its vowels permitted frequent omission of consonants. 

Continuing with the discussion of Latin vowel dynamism, let us turn to its next property. There exist bonds of affinity between certain vowels and some consonants. That is, as soon as a vowel had skipped down its natural degradation ladder to a sound which had such affinity between it and the following consonant in the word, the process would go no further. Sometimes, vowels are even raised one step if the tie with a particular consonant resulting from such a change is exceptionally strong, which regularly occurs at the last steps of the ladder, especially in the case of E and I, both being weak and close sounds, though never in the instance A or O (we thus see that “I” can represent the weakening of practically any letter, while A, in contrast, must be always the original one: see the table above). Some most striking affinities are: 

U before labials (especially L): cul-tus (<- coltus); 

I before dentals (especially N): cecini (<- cecani); 

E before R and closed syllables; also in final position and in the last syllable before M (which was very weak in itself, a fact reflected in poetry elision rules). This is explained by the fact that the articulation of E, more than that of any other vowel, resembles the rest-position of the organs of speech. Examples: monuere < monuerunt; ipse < ipsus; ille < ollus. 

In reduplicated stems, the vowel of the root usually appears as E or I. Most verbs do not appear reduplicated in the Classical period since, at the time when accent fell on the first syllable, one of the doublets (the second one) allowed its weakly stressed vowel to become effaced, the lone consonant then becoming assimilated (merged) as well. Example: retetulit -> rettulit. 

Vowel Assimilation: 

There are 2 general properties relating to vowel assimilation which are observed in Latin as well: 

1) when 2 vowels come into actual contact, they tend to approximate each other; 

2) when 2 vowels are separated from each other by a consonant, they tend to become identical. 

Vowel Loss: 

U is lost only before L: vehiclum <- vehiculum; vinclum <- vinculum. 

U is also lost in -ulo when preceded by N or R, which consonants then assimilated themselves to the following L, producing -ello, -illo, -ollo, -ullo: ocululus -> ocellus, sterula -> stella, homonulus -> homullus, coronula -> corolla, stirula -> stilla. 

Sometimes -ulo was added to a word already containing it: popululus (popuo + ulo) -> pupillus; oscululum -> oscillum. 

E is lost before R, especially in suffixes -ero, -bero, -tero (-> -tri); also in likes of “pat(e)ris”, “ag(e)ri”. 

I is lost quite frequently practically in all positions. 

Consonantal Change: 

As regards the nature of Latin consonants, one of their more important properties is the fact that they never underwent any “satem” corruption: the Latin language is considered a pure “centum” descendant of IE. The changes that did take place are illuminated below. 

A hard sound in Latin could often become a corresponding soft one, but not vice versa (since a hard sound is more difficult to enunciate) as is seen from below: 

K -> G (cloria -> gloria; curculio -> gurgulio); 

T -> D; 

P -> B (bibo <- PA (stem reduplicated); Burrus <- Pyrros); 

Some other important changes of the consonants: 

D -> L (lingua <- dingua; lacrima <- dacruma; olere <- OD (as in “odor”)). 

D -> R (occurs seldom, mainly in the prefix “ad-”, the tendency disappearing in the literary age: arbiter <- ad-bitere; so also ar-cesso; somewhat different setting for the change in meridies <- MEDH + dies). 

S -> R (the rhotacism law: this change happens whenever a single S sound is sandwiched between 2 vowels; the transition occurred between the 6th and 4th centuries BC). 

Mutations of the Aspirates: 

Aspirated consonants were modified into non-aspirated ones at an early stage by practically all languages in the European group. Here is the process of this development in Latin. 

BH: in the initial position most often turns into F (*bhrater -> frater; *BHU -> FU; etc.), although sometimes H (ex. harena, while the form fasena is attested in the kindred Sabine tongue; haba and faba exist side by side in Latin; other examples: herba <- *BHAR; mihi <- *mibhi), which indicates that H in Early Latin was a rather strong sound. In a middle position, BH becomes B, although in related italic languages BH turns into F even inside a word. 

DH: at the beginning of a word this aspirate acquires most often the F sound (fumus <- *DHU; fores <- dhvara in Sanskrit; firmus <- *DHAR (to hold)). In a middle position, DH becomes generally D, although in a few instances B ( F in other italic tongues, which suggests that DH turned into BH in early Italy, which in itself became B in Latin and F in its sister languages), as seen in ruber <- *RUDH (rufus in Oscan), barba from *BHARDHA. 

GH: when this sound began a word, it passed mostly into H, as is evidenced by hiemps <- *GHI; hostis <- *GHAS. Sometimes, into F: fervor <- gharma, the Sanskrit “warm”. However, in a number of examples, some factors helped the aspirate to stay at G (this occurs widely when the following sound is R, whose liquid nature absorbed the breath): granum, grandis. Quite often practice hesitated, as indeed is observed in the existence of such double forms as faedus and haedus, “goat”. To see how the aspirate behaved in a middle position, it is worthwhile to look at the stem *VEGH: the aspirate became H in vehere, but enough of the strong G sound remained in the formative stages of Latin to produce the participle vegtus -> vectus, the G assimilating itself to the T. Briefly stated, the path of change of *GH inside a word is less strict. 

Consonantal Loss: 

Initial: 

S is lost at the beginning of words, since Latin did not like jumbles of consonants there: (s)nurus, (s)nix. Sometimes, even ST was lost: (st)locus, (st)latus. 

The same is true for V: (v)radix, (v)lacer, (v)lupus <- vrika in Sanskrit. 

Some other sporadic changes: 

C is lost before V, L: (c)lamentum. 

G is lost before V: (g)venter, (g)vorare <- GVAR, (g)venire <- GVA. Before N: ((g)notus <- GNA, same for nomen, narrare). 

T is lost before L: latum <- tlatum <- TOL. 

D is lost before R, V: viginti <- dvi, bellum <- duellum. Also before Y in Diovis -> Iovis. 

P is lost before L: laetus <- plaetus, as in sanskrit priya; latus, compared to the Greek platos. 

Clusters of consonants separated: unlike the more authentic Greek “krad-”, Latin has “cord-”; FRAK -> farcio. 

Medial loss (mostly before spirants): 

before S: di(c)-sco, mul(g)-si, spar(g)-sus, miles (milit-s), sua(d)-si, ce(n)sor, co(n)sol (the last two forms common in Early Latin, though the nasals were restituated later), quotie(n)s, ru(r)sum, etc. 

before Y: ma(g)-ior, se(d)-iungo, pe(r)-iero, tra(ns)-iicio, etc. 

before V: bre(gh)-v-is, le(gh)-v-is, sua(d)-v-is, etc. 

Also before nasals: 

before N: lu(c)-na, pi(c)-nus, de(c)-nus, va(c)-nus, ce(s)-na, po(s)-no. 

before M: sti(g)-mulus, exa(g)-men, re(s)-mus, Ca(s)-mena. 

Loss is hardly found before momentary consonants, except the dentals: 

before T: in past participles: tor(c)tus, ul(c)tus (here loss occurs in groups of at least three consonants). 

before D: loss of S in iu(s)-dex, i(s)dem, di(s)-duco. 

In all these cases loss takes place at the last sound of the root or a prefix. 

Final loss: 

Latin tolerates consonants and groups of them at the end, though not any 2 same ones (os(s) <- ossis, fel(l) <- fellis), nor 2 mutes: lac(t), cor(d). 

S is lost very often, a change to which the frequent vowel-only case-terminations are due. 

M starts to be dropped from at least III cent. BC, though not done away with until the Christian era. 

T also suffers loss in the final position from earliest times, though the practice never became universal or permanent in Latin (unlike languages such as the Umbrian): dede <- dedit, hau <- haut <- haud, dedro <- ded(e)ro(nt), patre(d) -- the archaic ablative form. 

Consonant Assimilation:  

a) Complete: 

First letter to second: 

supmus -> summus, sedla -> sella, lapid(u)lus -> lapillus, edse -> esse (D thus assimilated very often); petna -> pesna -> penna, cesna -> cena; flagma -> flamma. In these, the consonant was assimilated because the vowel preceeding it was initially short (lengthened in the process). When it was long, the same consonant was usually dropped altogether: sua(d)-vis (long a). 

Second letter to first: 

T in superlatives after S or R: dur-i(o)s-tumus -> durissimus; celer-is-tumus -> celerstumus -> celersimus -> celerrimus. 

S after R: fer-sem -> ferrem, torseo -> torreo. 

S after L: vel-sem -> vellem. 

V after L: mol(d)vis -> mollis, sol-uos -> sollus. 

Both letters pass into another double sound: 

T of -tus (the past participle ending) and -tor after a dental (T or D) of the root (not S, though: ustus, haustus): becomes SS: fid-tus -> fissus, cad-tus -> cassus, pat-tus -> passus. 

b) Incomplete: 

i) “euphonic change” (hard -> soft before soft, as in segmentum <- SEC, or soft -> hard before hard, as in actor <- AG); 

ii) momentary sounds to nasals: sab(i)nium -> samnium, sop-nus -> somnus; 

iii) T to S: after dentals R, L ( -tus, -tor become -sus, -sor): cursor, cursus, falsus, celsus, pulsus, though exceptions abound: artus, expertus, altus, cultus. On false analogy, words like the following were formed, even though no dentals are involved here: lapsus, lixus, fixus, fluxus. The same tendency to soften T to S is encountered in such manifestations as pultare -> pulsare, mertare -> mersare, mantare -> mansum. 

The Assibilation of C and T: 

It is generally thought (under the influence of the corrupted Romance languages, where it was true), that sounds “ci” and “ti” must both have been pronounced by the Romans as “si” (suspicion, ovation, generation). However, it is established that it was only in the final stages of the existence of the language that the difference between the two melted away. It was not until the seventh century AD, as we may judge from credible sources, that “ci” began to resemble “si” (although it must be avowed that in Italic dialects, the change is attested a thousand years earlier). As for “ti”, it was assibilated in the fourth or fifth century AD, at which time the transition was already accomplished universally among the humble classes and the provincials. Again, rare examples of this process are encountered even in Latin much earlier (before third century BC), but in the Classical age, such attempts were uncommon. The bottom line argument is that in the days of Cicero, “ci”, “ti” and “si” were all pronounced differently, namely: “ki”, “ti” and “si” respectively. 

Dissimilation:  

When two dentals occur together, the first one tends to become an S: equit-ter -> equester; edti -> est; claudtrum -> claustrum. 

Another rather curious demonstration of the princple of dissimilation is the existence of parallel suffixes -aris, -alis, both contributing the same modification of sense to the root-base. However, one or the other is preferred in each particular case depending on what consonants precede the suffix (repetition is avoided): vuLgaRis, popuLaRis; moRtaLis, lateRaLis. 

Lastly, some rather unremarkable and minor consonantal variations, quite foreign to the spirit of Latin: 

K -> KW (QU) -> P (because of a parasitic W sound creeping in; commonplace in Greek and italic languages. Examples: lupus (borrowed from Sabine), Petronius (from Quatronius), Pompeius (from Quinqueius). 

G -> GW (GU) -> B. A Greek process, unknown in Latin, although there is a rare form dating back from the common Graeco-Italian era: BOV <- gaus (sanskrit, “cow”). This root is found already modified in both languages. 

D -> B, a process hardly observed in Latin: bis <- duis, bellum <- duellum (“division”), bonus <- duonos. Here, it appears, the V sound played a role (it follows the B in all of the examples). 

The infiltration of parasitic (a.k.a. auxilary/inorganic) consonants is limited to a few examples: hiem-p-s, sum-p-tus, etc.