SPANISH PHONOLOGY by Jones Wesley Harris B. A., University. of Georgia (1953) M, A., Louisiana State University (1962) SUBMITTED IN PARTIAL FULFILLMENT OF ‘THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPEY at the ‘MASSACHUSETTS INSTITUTE OF ‘TECHNOLOGY September, 1967 Stgnacure of Author»... -.. on ea oo O08 poop ORp Departebat of Modern Languages, August 21, 1967 Certified BY. eNO ee : Thesta Supervisor Accepted bY - ee ee aT Chairann, Departnental Comittee on Graduate StudentsSpanish Phonology James Wesley Harris Submitted to the Department of Modern Languages on August 21, 1967, in partial fulfillment of the requirement for the degree of Doctor of Philosophy. The phonological component of a transformational generative grammar of a dialect of Spanish is studied in the framework of the phonological theory whose most recent and comprehensive formulation appears in Chomsky and Halle, The Sound Pattern of English. The purpose of this study .s twofold! first, it strives for deeper in- sight into the widely studied facts of Spanish prouunciation by pre. senting a theory of these facts in the form of a generative grammar second, this study confronts the current theory of generative pho- nology for the first time with the results of sn extensive and de- tailed investigation of the sound structure of Spanish. It is shown on the one hand that, in general, the phonoiogical theory proposed by Chomsky and Halle, including recent innovations, is strongly sup- ported by the data of Spanish, and on the other hand that serious inadequacies remain. Several controversial and heretofore unresolved problems of Spanish phonology yield to insightful analysis in this theory. For example, the question of certain assimilatory phenon- ena before glides finds a natural and simple solution, and substan- tive clarification is achieved of the phonological and phonetic rep- resentations of the various r-type phones. Chapter I presents a general introduction to the investigation. Chapter II provides a close examination of rules involving the phonetic features of nasals, glides, voiced obstruents, and nonlateral liquids. Chapter III considers the wide range of phonological pro- cesses involved in the rich inflectional system of the verb. Chapter IV investigates various additional consonantal alternations, and con- cludes with an ordered list of all the rules discussed in the study. Chapter V is a historical excursus which traces the evolution of the medieval stridents in two dialects. Thesis Supervisor: Morris Halle Title: Professor of Modern LanguagesCHAPTER 1, CHAPTER I. CHAPTER III. CHAPTER IV. Table of Cont Introduction Some Consonantal Phenomena 1. Background 2. Nasal Assimilation 3. Glide/Corsonant Alternations 4. The Voiced Obstruents 5. The Nonlateral Liquids Verb Forms 1. Introductory Remarks 2. Present Indicative 3. Present Subjunctive 4. Participles and Infinitives 5. Imperfect Indicative 6. Imperfect Subjunctive 7. Preterit 8. Summary of Rules 9. Tense and Lax Vowels 10. Vowel Alternations in Third Conjugation Verbs 11, Syntactic Excursus 12. The Future 13, Further Clarifications and Summary of Rules Additional Consonantal Alternations 1, Introductory Remarks 2. Athematic Nominalizations 3. Productive Patterns 5 8 8 10 22 an 6 76 6 78 a 83 85 87 om 95 167 123 128 1384. Other Nominalization in -cién 5. Nominalizations in -siéa 6. Real and Apparent Anomalies 7. Suomary of Rules 8. Additional Alternations 9. Concluding Remarks and Summary of Rule: CHAPTER V. Historical Excursus: Reflexes of the Medieval Stridents BIBLIOGRAPHY BIOGRAPHICAL NOTE 156 165 169 175 176 192 206 218 224GHAPTER 1_- Introduction This work presents the results of an investigation of the sound struc- ture of Spanish within the framework of transformational generative grammar. Its purpose is twofold. First, it strives for deeper insight into the widely studied facts of Spanish pronunciation by presenting a theory of these facts in the precise form of a generative grammar. Such precise and explicit formulation can lead to the discovery of serious gaps in our under= standing, and may have the heurietic value of suggesting a principled way to #411 these gaps. Secondly, this investigation makes available a large and detailed body of data of the appropriate sort needed to confront the univer- sal phonological theory whose most recent statement is found in Chomsky and Halle (1968). This phonological theory, perhaps the most elaborate and richest yet postulated, is based primarily, though by no means exclusively, on a deep and massive study of the phonological component of a transforma- tional generative grammar of English. At present this theory can best be tested, sharpened, and enriched by examining it in the light of phonological studies of ocher languages, which approach in both scope and depth Chomsky and Halle's study of English. At various points ia the exposition, obser- vations are made regarding the empirical adequacy of the current formilation of the phonological theory in question. It is shown on the one hand that recent theoretical innovations are strongly supporzed by the data, and on che other hand that serious inadequacies remain. Detailed proposals for the solution of such difficulties ere net made in the present work. This would be premature. Rather, the observations made here are intended as a contri- bution to the large body of reasonably clear data vhich mist underly future theoretical revisions.This study does not purport to be exhaustive. Rather, it contains what I believe is a fairly comprehensive account of the rules for the central and crucial phonological processes of Spanish, plus a minutely detailed examination of certain areas in which there ie near certainty that all che relevant data have been considered, ard which, furthermore, promise to contribute significantly to the understanding of the sound structure of Spanish and to phonological theory in general. For example, considerable attention is paid to the question of the phonological and phonetic representation of glides and of the flapped and trilled varieties of x in Spanish. On che other hand the distribution in phonetic representa- tions of tenser and lexer allophones of vowels is completely ignored. As another example of an intentionally neglected topic, no rules are proposed to account specifically for the alternations illustrated orthographically in che double-underlined letters of £1/ellos, aguél/aquellos, doncel/doncetla, and desdén/desdefiar, since these four pairs of examples exhaust the cases to which such rules would apply, and are of no further interest. Also not includ- ed is a full account of many morphophonemic phenomena which have tradition- ally been considered irregular or exceptional, for example, irregular pre- serit and present tense forms of verbs, While apparently irregular and ex- ceptional forms are of value in dealing with questions whose solutions are underdetermined by the data provided by clearly regular forms, it seems to ne misguided to attach much weight to irregularities until at least the out~ Line of a theory of regular cases is clear. Exclusions of che sort just men- tioned do not contradict the claim of comprehensiveness of major phonological processes made for this study. It {s only in the light of the total grammer at detailed exploration of particular areas can te carried out with a reasonable assurance of relevance. Lack of such assurance greatly reducesthe interest of recent work in Spanish phonology such as Foley (1965), Sableski (1965), and Saporta (1965). The dialect of Spanish described is that of educated speakers fron Mexico City, specifically, my own (virtually, though not technically, native) pronunciation and that of a few Mexican friends, This dialect was chosen out of necessity, There does not exist for Spanish anything equiva- lent to, say, Kenyon and Knott's A Pronouncing Dic-ionary of American English, and the state of the study of Spanish dialectology is deplorable by che standard of such work in other major languages. Therefore the only body of data accessible to me in sufficient breadth and detail is that which Tam familiar with from first-hand experience. Information about other dialects plays an entirely peripheral role in this study, The extent to which the results of this {nvestigation say carry over to other dialects is, of course, an empirical question, and mist be left to future research. Aword about transcription: For the sake of readability, as little or as much phonetic decail is indicated in examples as the discussion requires. As a result, transcriptions range from standard orthography, which is Largely unambiguous and well known, through mixed representations Like haBrd and arotelsler (where confusion atgit result from not knowing whether a given character is a letter or a phonetic symbol), to conventional phonetic transcription.(GHAPTER IL jome-Consonantal_ Phenone: 1. Background The facts to be discussed in this chapter are well known; so much so that many of them are to be found in almost any elementary treatment of Spanish pronunciation. Knowledge of the facts, however, has not generally been accompanied by understanding of them, and careful studies (several of which will be cited below) have ended up in irreconcilable disagreement and puzzlement. Stockwell, Bowen and Silva-Puenzalida (1956) argue that certain of the phenomena to be discussed here might be handled by crib (ing these] alternations to free variation et this [‘phonemic] level, presumably control- led by stylistic chotce... [‘Thig] solution is easily eliminated because it is equivalent to saying that the variation cannot be described systematically even when it can be... At any rate, we would not object to recognizing free variation in an instance of thissort, where there is fair jurance that the variation will actually be statable ona stylistic level... It is, of course, recognized that a0 one has, to our knowledge, actually described the conditioning on this higher level" (pp.408~-409). Evidence will be pre- sented here that the correct solution to the problems referred to does fact involve recognition of levels of style. The following are impression- istic characterizations of the four styles which will be taken into account. qa a. Largo: very slow, deliberate, over-precise; used, for instance, when trying to communicate with a foreigner with little competence in the language, or when correct- ing a misunderstanding over a bad telephone connection.>, Andante: slow, careful, but unaffected; used when one is minding his p's and q's but not selfconsciously. Typical of, say, teaching a class or delivering a lecture in a large hal! without electronic amplification. ce. Allegretto: moderately fast tempo, casual, colloquial. In many situations one might easily alternate between Andante and Allegretto in mid discourse or even in mid sentence. 4, Presto: fast, careless or sloppy. We will be concerned primarily with Andante and Allegretto, but will aencion Largo and Presto at various appropriate points. The dialect under consideration, educated Mexico City speech, is generally thought of among Latin Americans as hyper-careful and sometimes slightly affected. This can ic interpreted as meaning that in thie dialect, the span of differ- ences between Andante and Presto is smaller than in some other dialects, educated Havans speech. There is in fact much less difference betweea Mexico City Andante and Havana Andate than between Hexico City Presto and Havana Presto, as the following example will show. Q Mexico City Havana, Largo ( mismo] [afemo Andante [mis*no] (ais*ag Allegretto [nis*ao] [mifmd] (CA] = voiced Ci] ) Presto (mf emg] Caf: ma] As hinted at above, a finely detailed phonetic transcription of a sufficiently long discourse might reveal considerable mixing of styles.40 I agree, however, with Stockwell et al, that it is wrong to toss everything into the Limbo of free variation, however defined, since, for example, {nfsmo] and (mizmo] simply are not free variants: the difference can be described systematically, as will be shown, The crucial point is the following: certain alternations can be correctly accounted for only by recognizing descrete levels of style. As the exposition proceeds, the vague descriptions given in (1) will be replaced by the sets of rules which explicitly characterize each style, and a few tentative observations will be made concerning the notions “style” and "stylistic level," as these are reflected in the formal differences among the sets of rule 2. Nasal Assimilation Im the examples to follow, [1m] represents a labiodental nasal, Co] represents a dental - not alveolar - nasal, and [n] represents an alveolar nasal. Standard orthography is used otherwise, except for(0] - 2.1 The following is the data for Largo. @) labio- Nasal before: bilabial dental dentai alveolar palatal veler Obstruents campo cuanto canso —rageho ayea cambio cuando saqga ajenio Liquide onra Masals inmenso innate Further details and observations: a. "Normally," that is, before vowels and finally, n is alveolar (n] .ey b. Bilabial [m], alveolar (n], and palatal (J contrast intervocalically: mo, ano, afic. On the other hand, labiodental (m], dental [n], and velar [y] occur only as assimilatory phenomena, c. Navarro Tom4s (1965) states that palatal [#] occurs before palatals, e.g. rallcho, but with the qualification that "la pronunciacién lenta y silabeada puede hacer que la n mantenga en estos mismos casos su forma dpicoalveolar, mis o menos palatalizada, sin asimilarse por tal siguient entero al modo de articulacién de 1a pa! " (p. 133). Note carefully palatalizada, "palatalIZED" not palatal, "palatal." Yany writers, possibly copying Navarro carelessly, have stated flatly chat only [7] occurs before palatals. This is patently false in the dialect and style under consideration. In the phonological theory which characterizes palatais as [=diffues, -grave], we can only say that this nasal is [+diffuse, -grave], and leave the qualification "ads 0 menos palatalizada" up to g-ary phonetic detail rules. d. In Largo there is no assimilation across word boundaries, Thus while ‘nb, “mt, etc., are impossible within a word, the following do occur (where space indicates word boundary): [un beso], [un kakto], etc. €. There is no assinilation before glides within « word: nuevo [qvefo]. aiete [ayeto] . On the basis of the above data, we might formlate rule (4) if we {gnore the bilabiai-labiodental and dental-alveolar distinctions, and if we assume that the nonassimilated n is [4diffuse, -grave]: +obstruent « - 7 c /—] -mee crave Comat] eee] odiffuse12 We wust use [+grave] rather than (sgrave] in order to exclude assimi- lation before palatals. We have used [+obstruent] in the environment in order to exclude assimilation before nasals: [inaogso], aot [immenso]. This is probably incorrect, but irrelevant to the present discussion.» The above, however, does not exhaust the data. Consider words Like redimix, "to redeem," redentor, "redeener," redencién, "redemption," and consumir, "to consume," consunto, “consumed, “consuneién, "consumption." Evidently we have underlying stem final m in thesewords. Thus the nasel assimilation rule(s) must have not only the effect of (5), dut also that of (8): qs) af Je } lave CS) an /__|[dencai] In other words, since dentals are [-grave] we can no longer state nasal assimilation for Largo as in (4). Among the conceivable alternatives to (4) are the following: ” ( ; +obstruent snext rule] / ~diftuse Tgrave [nasa] . . _ +obseruent] adiftuse (xe) (T= 8)> (a= 4) or (a= -)> (t= 4) (T#B)D (x#¥) Alternative (7), in addition to theoretical difficulties involving [-next rule] ,° assunes that there are no underlying af clusters, which may or may not be correct; alternative (8) contains an extremely complex set of condi- tions, We will leave this highly unsatisfactory situation for the moment, and turn to the data for Andante. 2 Nasal Assimilation in Andante. Within a word, the data is identical to that of Largo, That is, assimilation occurs before ell except palatel obstruents, but does not occur before glides and other nasals. Across word boundaries, the assimilatory phenomena of nasals are not at all straightforward. The following passage from Navarro Tomas (1965) is, to my knowledge, unique in its lucidity and attention to phonetic detail, and deserves to be quoted in full: "Nasal ante labial. --En contacto con las consonantes p, b, la a final de una palabra anterior se pronuncta corrientenente a, sin que en este sentido pueda advertirse diferencia alguna entre expresiones como, por ejemplo, con padre y comadre, pronunciadas anbas [kompaSre] , 0 entre con placer y complacer, pronunciadas (complaér]. Lan final mantiene, sin embargo, su propia articula- cida, ..., cuando por lentitud o vacilacidn en el lenguaje aparece desligada de la consonance siguiente. Suelen darse asimisno, segin la rapidez con que se hable, formas intermedias de asimilacién en que lan, sin perder enteramente su articulacidn alveolar, resultaaa en parte cubierta por 1a oclusion de los labios [emphasis mine, J.#.]. Zn la conversactén ordinaria, 1a transformacién de la n en @ ante las oclusivas bilebiales p, b, se produce de una manera regular y constante! (p.89). To this is appended the following footnote: "Los gramaticos han discutido extensamente sobre si lan ante p, b, se pronuncianog. La realidad da apoyo, como se ve, para varias opiniones. Todo depende de la forma de pronunciacién que se tome por ba I take Navarro's "lentitud o vacilacidn" to be characteristic of Largo, where there is no assimilation across word boundaries. On the other hand, his "conversacién ordinaria" corresponds precisely to Allegretto, where, letting the exposition get ahead of itself, there is full assimilation across word boundaries. That leaves the "formas intermedias de asim{lacion," which I claim should be assigned to Andante. I propose the following: the nasal assimilation rule, stated in terms of binary features, is the same in Andante as in Largo, whatever the correct form of this rule is: later scalar rules assign a greater degree of assimilation across word boundaries in Andante than in Largo. In particular, in the case of nasals before labials, the binary rule of nasal assimilation does not apply across word boundaries, hence it does not apply to representation such as conpadre and un#beso.* Then an n-ary phonetic rule assigns to the nasal in question the feature ‘4 Labial] where 4 possibly equal to 0). argo “Suargo { u tandance ” Navarro's description of the other nasal assinilations are not as de- tailed as the one quoted, but, in general, they support ay proposal. 1 will cite one mo:15 "En el grupo nm la articulacién de la primera consonante, en la conversacion ordinsria, va generalmente cubierta por la de la a: la lengua realiza, de manera mas 0 menos completa, el contacto alveolar de 1a n; pero al mismo tiempo la = forma su oclusién bilabial, siendo en realidad el sonido de esta altima el dnico que dcusticamente resulta perceptible: innévil [1™méai1], conmigo [kotmfvo], ...; en pronunciacién lenta, anbas articulaciones, my By produciéndose sucesivamente, resultan claras y distint: (p. 113). 2,3 Nasal Assimilation in Allegretto: The data are sufficiently different from those of Largo and Andante to warrant stating them completely, even though this entails some repetition. (Space indicates word boundary.) Oy labio- Nasal before: dental dental alveolar palatal velar Obstrvencs campo emfético cuanto — canso rancho ania cambio cuando gagga ajegie um peso um foco un taco un saco. un charco ug cacto um beso un dios uy gato ug juego Liguids onra ug ron un lage {nmenso ingato um mexicano up gopal Glides muevo[ mes of aiel{ayet] oo quevo[aweso| gteto (nyeto] 1g huevo Cuqueso] ea atelo [uayelo]16 Thus we see that there is assimilation before nasals, and that word bound- aries are irrelevant in the sense that the same assimilations occur acro: word boundaries as within words. Alternatives (7) and (8) could be adapted for Allegretto by subscituting [+consonantal] for [+obstruent] > and in- serting (#) {n the environment, were it not for the w glide. We must add, apparently, to either alternative, a rule with the effect of (10). (10) _ -vocalic [ terave ~consonantat [+masai]>| -diffuse! / __# | tzeave There may be some notational device which would permit the incorporation of (10) into (7) or (8), although this would be hair raising -- note that the £ {8 aot optional in (10). Obviously something is wrong somewhere. If only the [w] in huevo, but not those in muevo and Ruevo, were velar conson- ants rather than glides there would be no difficulty, This is not just wishful thinking. We will turn presently to a discussion of glides in Spanish (about which more ink has been spilled to l S$ purpose than about any other topic in Spanish phonology); first we mst resolve some of the pre- sent difficulties. 2.4 Let us refer to as T, the phonological theory in which che features [dtfeuse] and [grave] provide for a four-way distinction among points of articulation for consonants. In T, there ts no way of making the bilabial- labiodental and dental-alveolar distinctions which have been observed in the nasal assimilation daca of Spanish. Let us refer to as Ty che theory Proposed by Chomsky and Halle (1968) in which these distinctions, end others, are handled as illustrated in (11):WwW ay a a2 aay coronal See ee anterior tee eee high - et tee back eee Mstributed + = + = + + (We have been using the diacritic subscript [ ] all along to indicate the (-disertbuted) nasals (p] and[n].) In T, the assimflations represented by (nf (nd, ind. (pd, and(gh] wit be acounted for ££ (12) 18 a proper part of the nasal assimilation rule, a2) [ high [echigh } [nasa] | back back ‘distributed | Tatsertburea | The problem remains, however, of excluding assimilation before palatal (¥] (with the features [+coronal, -anterior, +high] as opposed coli], which ts [-coronal, -anterior, +high]) while permitting it in all other positions, One possibility is to say that assim{lation occurs in the envir- onment =--[anterior, - 4 | «anterior | canterior facorscat deoronal Yhigh t Yhigh Sback Sack satetrsbuced palaceibuted afore proceeding, let us review the source of the difficulty. Masale ssimilate so that they become entirely homorganic with a following conson- ant except that before [%] che nasal which occurs is [p] and not [8]. th ty che [+atefuse, grave] nasat [9] does not Secone che [-dttfuse, -zrave] nasal [a] before the [=diftuse, -gravel obseruent (J. ta hie theoretical fcameword there is nothing further ¢ t can be said. In Tp, however, there is something further that can be said: in T, there is another theoretically possible nasal n* with the feacures [-anterior, 4coronal, +high, -back, +distributed]. What has been insisted on ie that the nasal which occurs before palato~slveolar {%] is not palatal [A]. 1 am quite willing to be- lieve that this nasal is n*, that is,a palatcalveolar nasal with the same point-of-articulation features as[€J. This {s perfectly compatible with Navarro's (rather vague) articulatory description of this nasal as "mas omenos palatalizada." If this is correct, then all the mythical rules present- ed so far in this discussion can be discarded, and one can state simply that nasals assimilate completely bezore all consonants, as in (15): as) +consonantal [ xanterior anterior ecoronal coronal (+mase1]4| vhigh foe thigh sback ‘back These considerations provide evidence of a very strong sort for the correct~ ness of T, as opposed to T,. Facts which cannot te accounted for in T are cs accounted for in Tj: the bilabial-Labiodental and dental-alveolar cont: can be expressed in T, but not in T,; and further, the possible existence of phonological m& is no longer « problem. In fact, (15), which is expressible only in, states positively (and correctly) that {m¥] cannot occur phonetically, while in T) this fact must be stated ad hoc, there being no independently motivated explanation. In T, tricky theoretical issues do not arise (such as those involving [-next rule] and complex conditions on rules), and real simplification is made possible (as opposed to the illusory complications which merely reflect the gain in phonetic accuracy of T, over T,). 2.5 Summarizing, rule (15) may be proposed as it stands as the nasal assimi- lation rule for Largo and Andante, For Allegretto and Presto, we may add optional # or = to the environment.° More explicitly, the nasal assinila- tion rule for Allegretto and Presto may be stated as (16): as +consonanzal [wanterior ] | xanterior Z4nasat}s| Thigh Yaigh | Sback | sback 1 | Sdiseributed| Lediseeibuced |20 2.6 The Assimilation of 1. In the dialect under study, 1 is "normally" alveolar, that is, it is alveolar initially, intervocalically, finally, and before labials, labiodentais, palatals, and velars. But before true den- tals, viz. £, d, it becomes dental, i.e. [+distributed], As a first approxi- mation, we might propose the following rule: [ teonsonancat | tacterior 1—[4dtscributed]/ --- | coronal | 4diseributed | a7) Notice chat it is necessary to specify both (+anteriod and [ +coronai] tn the environment: 1 does not become [+distributed], i dental, before [+aisertbuted] labials, palato-alveolars, or velars: [felpa, kolta, palko] not [felpa,*kolda, *palko]. Te will have been noted that the L-assimtlation data is just a subsec of the nasal assinilation data, Rather clearly there is a significant Lin- gistic generalization here: noncontiquant sonorants become homorganic with 4 following consonant, within the limits set by universal constraints (there are labial and velar nasals, but not labial and velar 1's). This generali- zation will not be captured if both (15), with its Allegretto variant, and (17) are in a grammar, In order to capture this generalization, one might propose the follow- ing. There is the nasal assimilation rule (18) which assimilates nasals with a following consonantal segment in major point of articulation. aay . - eel | +consonantal { sassat }s[ corona! cencertor | peoronalaa -seg] In Allegretto, the environment of (18) would of course be (78 - ‘Then a separate sonorant assimilation rule (19) assimilates nasals and in features of phonetic detail with a following consonant with which they already agree in major point of articulation: qs) f sunt] ; santerior r +eonsonantal | Yhigh ‘d@nterior| | ecoronal +sonorant aback | pcoronat Yhigh | scontinuant | | @distributed ba - 7 |_ Sdistributed It is clear, however, chat (18) and (19) fail to accomplish what they were invented to accomplish. Ceteris paribus, a grammar containing (18) and (19) will be less highly valued than a grammar containing (15) and (17) under the theory of evaluation proposed in Chomsky and Halle (1968). These facts seem to we %o point to a major deect in current phonologi~ cal theory, it should be possible to state a phonological rule as something Like (20): (20) ssrmmuare: [ i, a] [+consonantat] which would be interpreted to mean that 1 and n take the values of the features, within the limits set by universal constraints, of any following consonantal segment. Thus (20) would yield, e.g. labial and velar nasals, put aot labial and velar 1's before Labial and velar [ +consonan:al] seg~ ments, More generally, it seems that phonological theory should be revised ro allow rules which aention processes such as "assimilation," "palataliza- tion," "strengthening," “lenition," "vocalivation,” "consonantalization," ecc., and to include an evaluation measure such that such rules are in general less “costly” than other types of rules. The issues involved are22 complex, and we cannot go into them further here, However, the literature abounds in suggestive facts,” and it seems to me that the Spanish data just presented provide prima facie empirical evidence that some revision of phonological theory along the lines vaguely suggested is warranted. For the present, we will retain the rules of nesal- and l-assimtlation as in (15), (16), and (17). 3. _Glide/Consonant Alternations In the Literature of Spanish phonetics and phonology, segnents gener- ally called glides have either been passed over Lightly, or they have caused considerable confusion and controversy, as in Bowen and Stockwell (1955), Saporta (1956a), Bowen and Stockwell (1956), and Stockwell, Bowen, and Silva-Puenzalida (1956). A minimal introduction to some of the probe lems was given above at the end of Sec. 2.3, Let us proceed by stating the phonetic facts. The most accessible descriptions are the following: 3.1 King (1952) As an allophone of /i/: "[g], occurring between vowel and consonant oF between consonant indéa, aiads, Sten, and vowel: /signsia, 1 re/" (p. SIE). As an allophone of /u/: "[y], occurring between vowel and consonant or between consonant and vowel: /lasuérte, antiguo, kuidédo, oyuslo, déudas, automdbil/" (p. 52). As 4 separate phoneme: “Ya is the class of labial semivowels, including only: [w] with varying amounts of voiced velar friction [emphasis mine, J.t.],3.2 23 occurring initially and medially before vowels: /awa, manawa, Eiwdwa, suwante, suswiskis, laswértas/. In certain styles of speech [gy] and [w] are in contrast: /séngudntes, sdnwdpos/" @. 53.8 As a seprate phoneme: “/y/ 4g the class of palatal semivowels, including: [ J], with varving amounts of voiced palatal friction [emphasis aine, 3.1], occurring initially and medially before vowels: /dya, yégo, subyuga, losyéba/; [ 3%], with varying amounts of voiceless palatal friction (emphasis mine, J.H.], occurri:ig before pause: /mebdy#, dyi, mdy#/" (p. 53). Bowen and Stockwell's (1960) pedagogical statement: “Spanish /y/ at the beginning of a word is often considerably more tense than English /y/, but not started as a stop consonant, rather Like the /d/ sound, as the English /j/ is, The reason ic is sometines heard as English /y/, sometines as English /J/ is precisely because tc is partially similar to both, but not identical with either: it is, 80 co speak, between them" (p, 31). "Spanish /w/ is similar co Spanish /y/ im the relative tenseness of its pronunciation when it is che first sound in a word. Where initial /y/ suggests /dy/ or /j/, initial /w/ ts often heard as /gv/, and a few vords are actually Listed in the dictionaries with two variant forms, for exaaple, huaca and guaca" (p. 82). Bowen and Stockwell (1955): "Cy] before a vowel is an on-glide from high=front tongue position: "¥] is the same combined with palatal friction [emphasis mine, J.H.]:2h [] before vowel is an on-glide from high-back tongue position with Lips rounded; [ ¥] is che same with velar friction “emphasis mine, J.H,]. The sane syubols after a vowel denote off-glides to- ward high-front or high-back position" (p. 400, note 2). 3.4 We see that these scholars -= and they are typical -- have assumed that the various fricative type phones are glides, not true consonants, with some sort of funny phonetic admixture of fricative noise. Although one does not know what decisions would have been made tf these writers had been working within @ phonological theory which countenanced distinctive features, it seems never to have occurred to anyone that these fricative phones might be, at some stage of derivation, true consonants. This is just the claim that I am going to make, and we will see that various phenomena (e.g. Allegretto nasal assimilation before [w]) find natural and simple solutions within the current theory of distinctive features, Let us first establish the following uniform transcription: king oven and Stockwell danceforsh g y y i 3 4 w a Y i* x The intended interpretation of these symbols with respect to che features relevant to the present discussion is as follows:25 (21) yy yo ow consonantal = = + + s+ anterior - 2 ee coronal eos eee high + + FF + back soe eH round ee eH continuant toe RR voice + ote +H + Let us now reexamine King's examples (Sec. 3.1) in the light of (21). For simplicity we will consider first oniy the [-back] phones. (In the following, space stands for word boundary, the ad hoc symbol @ stands for pause, total cessation of phonation. ) Examples with [y]: syénsya, la syudéd, dy dos, dyga, ayre Examples with [YjJ: @¥ a, ¥,ég0, sub=t,Sga, los Yyéba c. Examples with [xj]: me box,#, dx, mix, one has but to recognize boundaries to see chat {y], [¥], and [x] are in complementary distribution, as follows: (22) a. [xm] occurs in the environment __ # 2B |J— a bs [aq] oesoes tn oe etre | fl c. [ yJoccurs elsewhere Implicit in King's data is the fact chat unstressed( i], 1ixe [y], does not occur between vowels: [&¥yaJ, but not [ *éya] nor trisyllabic [tata]. An additional relevant phenomenon which these data do not reveal is that there are alternations between the unstressed simple vowel ¢ and the dishthong26 YS: pensamténto, "thought (noun)," pensdémos, "we think," pyénso, I think." Further, there are alternations between unstressed e and[ ¥,é}: held [eld], "it froze," hiela[ Yjela], "it is freezing"; and eri [Yyerra], "(he) errs." These examples can ei ily be multiplied, It is imediately obvious that stressed e¢ diphthongizes to ye under certain condi- 9 tions, and that the resulting y is changed to[Y,] in the environments (22). AL1 the data of this section can be summarized with the following inforn= al rules. Curved lines indicate the (partial) orderin; (23) Av Eye (under certain conditions) (- C4, -seress]oy / (-} v -yot/ | / Yom /_# ( For the moment (23) is simply another way of organizing the data. As the ex- position continues and more data are accumulated, chese rules will be modified rather drastically. 3.5 We consider now the [+back] phones: Examples with (w]: la swerte, antiguo, iwidddo, of,welo, déwdas, awcomobil Sxamples wieh [Y]: 4a, mandya, Ly Ma, suy "ince, sue p*Lacts, las yértas As far as these examples show, [wjand [1 | have exactly che same distribution Jana [¥], respectively. the absence of a[ back ]27 analog to[/x,] need not concern us now. Likewise, [w]j and unstressed [uJ do not occur between vowels: [ af¥a], but not | *awa] nor trisyllabic [ *éu There are also alternations between unstressed 9 and the diphthong we: soltdmos, “we release," swélto, "I release"; and between unstressed o and [46]: oldr, "odor," olémos, "we smell," huele [yYele], "(it) smells. Henceforth upper case O represents o's which diphthongize under stress. follows: We may now extend (simplify) the rules of (23) {e yé (6) a {¢ = (under certain condi:ions) (8 Nee { — [ sconsonantal’) - y b. | thten > [ovwemtte] 1 {yo 1 vocalte]—+ [ teonsonancat] {a : (Rf) 3.6 We now summarize the discussion up to this point, and clarify a few matters: a, The phonetic description and transcription of all segments in Secs, 3.1, 3.4, and 3.5 which are glides at sone point of derivation are those of phonetic output of Andante only. Said another way, | y],[¥,],[w} and [y*] do-tn fact occur in tha final phonetic representation of Andante, as in che examples given. heretofore in the literature have been considered as >. Segments whi: glides plus frteacive noise, sanely [¥,] [x], and[r™], are not glides but rather true consonants, The specification of these segments as true con- sonants {s clearly aecessary on phonetic grounds in any phonological theory iculatory che relationship between phonetic transcription and a: in whic! and acoustical data is taken seriously. Moreover, the decision =o creat28 these segments as true consonants will be further justified below in terms of the set of phonological rules of the Language. c. The distribution of the segments in question can be completely deter- mined by rules which have access to boundary symbols. It will be seen be~ low that these rules are in part independently motivated; that is, they would be necessary even if the segments under disctssion did not exist in Spanish. 3.7 The issue of [gw] and[¥"] raised in Sec. 3.1: In some dialects there is a contrast betwen the initial sound of words like guante, guapo, etc.. and the initial sound of words like huéle, hudvo, ("egg") etc. On the other hand, in the dialect under study, and tn Andante style, no such contrast is possible, All these examples, and all words like them, begin with[¥"]: [yénte, fépo, Yéle, 1%éso]. Only in Largo are words written with {nitial gu¥ pronounced with an initial stop. Now it is clear that the phono- logical representation of huéle is /Ole/ (because of olor, olemos, etc.) and at of budvo ts /Ovo/ (because of gvidicto, ovifdrme, oviparo, ovulactén, gquério, etc.). But the phonological representation of the initial segnents of guante, guapo, etc., is not so clear. The only alternations that could provide any evidence are those of the stylistic variants Largo (gwante, gwapo] versus Andante [y*énte, Y¥spo]. One might take the position chat the Largo variants are simply “spelling pronunciations," and hence have no bearing on phonological representations. I will, however, tentatively and rather arbi> trarily take the position that guante, guapo, etc. have underlying /g¥/ in all styles, !° on che somewhat dubious grounds that tt would be slightly odd if closely related dialects had different lexical representations for the same items, and extraordinarily odd if different styles of che same dialect showed this disparity in the lexicon. If this is correct, then rule (25) is29 needed to account for the data of Andaute. This rule is to be ordered before (24c). (as) ows t__ Thus the derivations of guante and huevo would include the following steps: eYdnte 6v0 we (aha) . «sy ” (2b) Mente iNéve Presumably, (25) ts missing in dialects which contrast the initial sounds of guante and huevo, and in Largo in the dialect under study. 3.8 The data given so far for Andante are not quite exhaustive. For some speakers, the words hiena, "hyena," and hiato, "hiatus", are both trisylla- ste: [téna] and [idto], and form nearly minimal contrasts with yema and yate, which are [¥,éma] and [Yate]. There is one -~ and to my knowledge, only one == apparent (+back] analog, namely the formative which occurs in huir, "to flee," huida, "flight,"huidizo, " fugitive (adjective), "fleeing," and other words. Huida, for example, is trisyllabic [uisa], contrasting with dissyllabic giitra [r"{ra]. The question that hiena, hiato, Auida, etc., pre sent is why doesn't (24b) turn the initial high vowels into glides, which are chen turned into consonants by (24c). For reasons which are aot relevant che present discussion, it is conceivable that auida and the other words with che same formative have a consonant between the u and the i, which wouldblock (24b), and which would subsequently be deleted. Whether or not this is correct, no such possibility exists to explain the initial full vowel ion for assumiag of hiena and hiato, since there is no conceivable not: an underlying consonant immediately after the initial i, One might say that these are lexically marked as exceptions to (24b): alternatively one could assume for yema and yate the underlying representation /yema/ and /yate/ but for hiena and hiato the representations /iena/ and/iato/, and further, change the first environneat of (24b) to /C_VY. This would com- plicate the rules minimally, and vould make hiena and hiato unexceptional. With only these two words (and possibly huida, etc.) involved, the decision is hardly an important one, Having gotten the data into the record, T leave the question unresolved. 1! 3.9 Assimilatory phenomena of consonants and the glides: Let us take as representative examples muevo, nuevo, un auevo and miel, nieto, un hielo, The rules that have been suggested will give derivations which taclude the following steps: sive novo un vo. fl_to in Elo we ve ve a ve ye (2a) cid Y, — (24e) (15) fails a? i seater * . Nasal Assimilation awivo udvo -—y"dvo yl aydto Uso Rule (15) and Scalar nasal assimilation (page 14) fail for muevo, nuevo, miel, nieto because at the time of application =he nasal in question {s not followed by a [+consonantal]| segment; (15) fails for un huevo and un hieloBL because the word boundary blocks it. Scalar nasal assimilation, on the other hand, does apply across word boundaries before [+tconsonantal] seg~ ments. With respect to [4% 7 é16] , note chat since [ ]is [-coronat, canterior] , the scalar nasal assimilation will be covard palatal [3] , without loss of the alveolar point of articulation. This is the intended interpretation of che representation [*], 3.9.1 There is one further set of data which will be presented, without going into great detail, to support the claim that the phones under dis- cussion which have been traditionally called glides are in fact true con- sonants: in Andante, s voices to z before certain voiced segments: 4. Before obstruents: degde, loz dientes (vs. hasta, log tientes) bs Before liquids: izgla, loz lagos, Izrael, loz ricos c. Before nasals: mizmo, loz monos, azno, loz nuevos Additionally, g voices before "glides" across a word boundary, but not within a word: loz yates, loz huevos, but desverto, resvello. Voicing assimilation never occurs before vowels, either across a word boundary or within a word: los oss. Now, if the initial segnent of e.g. yates, Auevos vere really glides, then che rule of s-voicing would be rather com plicated. In the first place, the environment before which assimilation occurs would have to include obstruents, liquids, nasals, and glides, while excluding vowels.!2 secondly, the rule would have to be rigged somehow co apply in the environments of (26) but aot in that of (27): (26) /__#0bservent /__Obstruent /__#uiquid /__biquid /__fGlideBe 27) /__Glide Presumably this could be done, but only at the cost of severe compli- cations to the extremely simple rule (28): (28) saz ‘—( [708] | Note further that if (28) is ordered after (24¢), vhich change glides into consonants in the environment #__, these complications would be an otiose duplication of part of the work of (24). Thus it is clear that the phones im question must be true consonants not only on phonetic grounds, but also because this is demanded by the internal organization of the rules. An even stronger case will be presented immediately below. 3.10 The Glides in Allegretto: The differencesbetween Andante and Allegret= to with respect to the glides are the following: (a) in the few words with AV and u¥ in Andante, [iéna], [ideo], [ui sal, etc., che corresponding glides appear in Allegretto, [yéna], [yéto] ,[wi sa], etc., and (b) where Andante has [Y,] and (YJ, Allegretto has [yJ and(w], respectively. At first blush it might appear that Allegretto lacks rule (24c), which in Andante turns (y] and [w] into consonants. This, however, cannot be the case. Observe that in Allegretto, doth [g] and [n] occur before [w], across a word boundary: sin giliza [siqvire] but sin huida [sinvi sa] and that both (z] and [8] occur before [y] and w] across a word boundary: las llenas [lazyénas] but las hienas [lasyénas_] tan glizag [lazelras] our tas nuidas [leswitas]3B These examples show that there is nasal assimilation and voicing of s before some instances of [y]and[w], but not before others.'? Now, since nasal assimilation and voicing of g occur before the instances of [y] and[w] which aze consonants in Andante, but not before those which are vowels in Andante, it is reasonable to suppose that the former are consonants also in Allegretto at che time the rules of nasal assimilation and s-voicing apply, and that Allegretto, but not Andante has a later rule with the effect of (29): (29) It is slightly embarrassing that the grammar proposed has a rule which turns glides into true consonants, (24c), and then another rule that turns the latter back to glides again, (29). But this is infinitely less embar- rassing than being unable to account for the otherwise mysterious contrasts in nasal assimilation and voicing of s illustrated above. The following partial derivations of huevo, huevo (versus nuevo), and los huevos (versus Suevo, "Swabian") will illustrate the application of relevant rules: (30) wevo un wevo awevo los wevos swevo (after 24a) {evo evo t¥evos (he) 3 (a6) Loz 8) eve uguevo Lozwevos av3h It was observed in the first paragré ph of this section that hiena, hfato, ete. have initial GY in Allegretto, rather than initial WW as in Andante. If (24b) is allowed to apply to these words, then (24g) will change into consonants the slides resulting from the application of (24b)5 then (15) and (28) will apply, giving incorrect results. To avoid this, (2hb) can be ordered after (2c). There will be no undesirable consequences in Allegretto. One may well wonder if this is not the correct order for andante alse. We will assume that it is, but change in che previously given order will be without permananet consequence, since these rules will still undergo considerable revision. 1_Further Observations and Summary of Rules 3.11.1 Rules (24 bye), (25), and (29) imply a close relationship among the segments [1, y, ¥,] and among [u, w, y¥, g¥J. Evidence is provided by these rules, even in their present tentative formulation, for the correctness of 1, as opposed to T,. Lot us compare the feature specification of the segments in question in the two theories: a ty wow gt vocalic ee ee consonantal -- aoa diffuse + at hoes grave - oe ee OR Bt y owe % yooh voca a ee consonantal ee ee anterior ee ee coronal Pe fee 3 Ga ee 7 23 ¢ bal :35 Lt is not necessary to belabor the point that T,, but not T,, reflects precisely the fact that the alternations Lay ~¥,, u-w ~1%, and wea” are essentially alternationsof major class features ([vocalic] and [consonantal]) and not of cavity (point-of-articulation) features. 3.11.2 Consider King's example hoyuelo (sec. 3.1, second paragraph), which in Andante is Lo ¥ywélo], not *Loy sé10] or *[o¥, H%é1q. Hoyuelo is diminutive of hoyo; it 1s presumably represented as oviOlo before, and as oytwélo after, the application of (24a). Now the protlem is to refine (24) 30 that it converts {y] to [Yj], but not (w] to CY" ] in this example. More generally, we must determine whether (24c) should apply to only the first of a sequence of two glides, or whether it should apply only to the y in either of the sequences yw or wy. Unfortunately, no clear examples of wy at this stage of derivation come to mind.” I will therefore assune, tentatively and rather arbitrarily, that the first alternative above is correct; that is, chat (24) should apply to the first of two glides. I adopt this alter- native rather than the other, since the result is a simplification rather than a complication of (2éc); the right-hand environment can be revised to { (-consonantal] | 4 a is shown in the following partial derivations, which sake only minimal This rule still does not operate correctly, however, as assumptions about superficial syntactic structure. Let us assume chat (242) applies cyclically: ap [oywelo] Clay] frevog ] Clay] Cuna] ] Cayuna] First eycle: nh nm hh (2he) j Second cycle: [ey reves] [ay uma (24g) fails36 Hay una should be [a ¥,uma] rather than *(ayuna]as in (31). We could correct this by changing the environment of (24¢) to /...__(#)+++ Yow let us assume that (24¢) is a rule of word-Level phonology, which applies only once in a given derivation. It is easily seen from inspection of (31) that under this assumption, and with any immediately obvious modification of the right environment of (24c), either hay huevos will be derived incorrectly as *{a ¥, JYevos], or hay una will again come out incorrectly as *[ayuna]. More specifically, if (24c) is a word-level rule and applies in the environment ._#, then hay huevos will be incor~ rect; {£ (24c) is a word-level rule and applies only in the environment /s+._(#) [Zeonsonantal], then hay una will be incorrect. one way out of these problems is to split (24c) into two rules, (32) and (33): #] (32) [-vocalic] 5 [+consonantal] [8°] _ [cage[l consonance (33) Evocalie ]+[+eonsonantal] / V__ sf i e This solution, while giving the correct results, nay seem somewhat inelegant, so say the least. However, evidence will be given in Sec. 5.24 that it is in fact correct. In Sec. 6.2 we will discuss the theoretical status of the ad hoe notation "#F." 3.11.3 There are several difficulties with rule (24b) which we will outline but leave unresolved in this section. consider che nearly minimal pair euidads [twis dso] and ciudad [syusds]. In neither sequence of diffuse vowels ui and iu (if in fact37 this is the representation before the application of (24b), which is not at all clear) ts either vowel stressed; hence, if the tvo environments of (24b) were unordered, the rule could apply to either vowel. The fact that (24) aust apply to the first vowel in both sequences sens to establish the order of the environments as they have been given. But we have gone too quickly: we have not actually shown that (24b) is in fact a rule of Spanish phonology. What has been noted is that, with a few exceptions, Spanish does not have sequences of the form CV,V,0 nor of the form CV»ViC where ¥, is [+high, -stress], but does have sequences of the form OGYC and CVGC. It is gratuitous, hovever, to assume that the only explanation for the absence of GV,V,G and V)VjC is that all such sequences are converted into GGVC and CVGC, respectively, by rule (24b). There exists, in fact, crushing evidence that chis assumption is false, to which we now turn. Foley (1965) proposed that stress in Spanish can be entirely accounted for by the familiar Latin stress rule. This is patently false, as will be shown in Chapter III, Sec. 9. It can be argued, however, that the Latin stvess tule is one of the rules that play a role in assigning stress in Spanish. In any event, neither the Latin stress rule nor any reasonable ad- Junct to it will assign stress farther to the left than the antepenultimate 5 syllable of a word. Now consider words like the following: » (34) a. laudano, ulico, dureo, cdustico, hidraulico, nautico, ufrago, cléusula, farmacéutico, terapéutico, enfitéutico, seléucidas b. alicuota, ventrilocuo, conyuge38 Ie is clear that if all the segments hich aze represented orthographically as vowels (and the y of cényuge) were represented phonologically 2s vowels at the time the stress rule(s) applied, then stress could not be assigned correctly: in every case stress would have to be assigned to the fourth vowel from the end of the word, If the stress rules are correct, then either (a) the segments written i, u, y which are contiguous to a vowel are not vovels in the lexical representations of these vords, or (b) there mist be a rule roughly Like (35) which is ordered before the stress rules: (36) [ -consonantal. y | Fata Jec-veeetie] / fe} Observe now chat (249), if it is im the grammar, is ordered after the stress rules, crucially. Therefore, if both (35) and (24b) are in the grammar, we are faced with the situation of having two nearly {dentical rules which can~ not be collapsed. We turn now to evidence in support of the existenc of (240). Zxauples Like the following can easily be multiplied: (38) a. amplio, amplydnos, amplyar vacio, vacyémos, vacyét insingo, insingénos, insinyér individgo, individydnos, individvdr b. pais, paysdno vai, aylc'co examples in (36a) are all verb forms. Amplio is first person singu- lar present indicative, amplyamos is first person plural present indicative, anplydr is the infinitive, "to enlarge.” The other examples in (363) are39 arrayed in the same manner. The examples in (36b) show stress shifts accompanying the common suffixes ano and -ero, Clearly the vowel-glide alternations in all these examples are determined by the position of stress, hence there mst be a rule like (24b) which applies after stress is assigned. In short, {f alternative (b) above is correct, then (35) must apply before stress is assigned, and (24b) must apply after stress is assigned; yet (35) and (24b) are apparently identical except that (24b) applies only to seguents which are [-stress]. The question now arises as to why the application of (35) does not remove all the cases to which (24b) might apply, e.gs, how do pais and badl escape (35)? One particularly simple answer vould be chat (35) does not exist. We will accept this answer until contrary evi- dence is presented.40 3.11.4 I now collect the rules proposed so far, renumbering them as (37) (37) a. (25) a —aw/t__ 8) ye b. (24a) i? , (under certain conditions) OF Nee + (30a) [-vocalie] [ +consonantal} C#] ~consonantal y a. (24b) +high — [-vocalic] / 4V. -stress —_ . [ feonsonantal ] [ «anterior santerior | acoronal, Acoronal Andante (15): [nasal >| thigh J | yhign ~ | Sback Sback ee @distributed. | @distributed [samen as: [rasapf ] wipe ] [ feonsonantal |] - 4anterior Andante (17): I++ [+disertbuced | /__ | coronal benim |_ tdiseributed Allegretto: 1s [téiseribsteal -_ (780° J 3+ (30b) vocalic |-> |" 4eonsonantal] /¥. we Pf ven 8a: +eonsonantal a. (2) so i_d. re) [eee ne ] 3 } Le Allegretso (31) | [} 94 a4, The Voiced Obstruents 4,1 The statement is nearly always found in school grammars that /b, d, g/ "gormally" occur as[b, 4, g] initially, after nasals, and sometimes after /1/, but as the nonstrident continuants [, %, Y] elsewhere. This state- ment is approximately accurate, but needs clarification. First let us clear up what is meant by "sometimes after /1/." In the dialect under study, only d occurs as a stop after 1; b and g occur as continuants: calvo, caldo, and algo are [kéigo, kéldo, 41¥0].'° Next, let us demon- strate that che "elsewhere" statement is actuslly crue. The pronunciations indicated in (38) are to be taken as illustrative of Allegretto. (I had to reach pretty far for examples in some cases; assimilated foreign words and substandard forms are marked with the symbol |.) (38) & Lt vy: haga nasa haya Le calgo fal alto a: hagla atldteres a¥lonerar ae drgol arée seYamasa haara patre alrio jaygo fnaysan jhayte aayerto asyestrar sivyendo r dewsa ())zewtma afuelo 2 atwero asgerso agtonen sus7lotal sugmarino asmiracion diaYnostico assur fo as°quirie jcluat ses” 24a Thirdly, let us clarify the statement that b, d, and g occur as stop [b, 4, g] "initially": an utterance such as Beatriz babea, "Beatriz slobbers," occurs as both [ beatrizbapea] and[beatrizpagea]. The former represents the more careful pronunciation, and the latte: the more casual pronunciation. Let us say that the former is Andante and the latter is Allegretto. Thus, for Andante, "initially" means both /#__ and /#__. (And probably also __. Twill assume that this is correct, although it {9 not known in all cases where = belongs.) In Allegretto, on the other hand, "initially" means only /#_. Notice that the bracketing of Beatriz pabea ts [5 [yp [ySeatriqy Jyp CupLvbabealy Jyp ] s- Hence the break between Beatriz and babea is at precisely the point at which the greatest possible number of # boundaries might occur. Still the initial b of babes is [A] in Allegretto. Let us concentrate on Andante for the moment. Let us assume, perhaps incorrectly, that the directionality of the b-f , d=, g- ¥ alternations is from stop to continuant, Under this assumption we will make a first approximation to a rule to account for the aiternations. We will call this rule "Spirantization." Now, since anything in the environment /#___ will automatically be in the environment /#__, and since we are assuming that “{nitially" in Andante also means /=__, we may say, as a first step: 8 ‘+ continuant \ |_ -strident EXCEPT / Essen [tense As a second step, we incorporate the fact that Spirantization does not occur bombay donde, and gany are [bomba, donde, gayga]s We extend the environment of Spirantization to43 [ -seg | EXCEPT/ ane _ [ +nasa1] As stated at the beginning of this section, only d occurs as a stop after 4A ({kaldo]); » and g occur as continuants ([kalgo, al¥o]). Thus we fur- ther extend the environment of Spirantization to [see -FB sxourt/ 4 (+nasat J (ecarena1) | os Observe now that there are generalizations which this formulation does not capture. If nasal assimilation, (37e), precedes Spirantization, chen che nasals before b, d, g will always have the same point of articulation as che latter. Now note that 1 and d have the same point of articulation, nasely (+coronal, tantertor], while 1 ond b, and 1 and g do not. Thus the voiced obstruents appear as stops when they follow certain homorganic seg- sents. We must look closely at the [+coronal] clusters with d as the second member, however, for it is not the case chat d occurs as a stop in all cases. The occurring sequences are nd, ld, but r§ and s§- On the ascuaption that 1 is [-continuant] while x is [tcontimant ], we may now say, refining our previous generalization, that the voiced obstruents appear as stops after aomorganic noncontinuants. Apperently, chen, we may reformulate the environment of Spirantization as -seg | 2] coronal coronal jee |}4h But this is unnecessarily complex: anteriority pleys no role since nasals will never disagree in anteriority with a following voiced obstruent, and 1 is always [anterior]. Eliminating [anterior], and stating the environ- 19 ment positively, the Spirantization rule for Andante can be given as (39): teontinvant |, {econesmanet a9) [rs enorent] 4 [Ronse orem strident (-scoronal) Recall that in Allegretto, "initially" means only # Thus in Allegretto, if Spirantization occurs in the enviroament X__, it also occurs in the environments Xf__ and X=__. Otherwise, Spirantization is just as in Andante. We may therefore state the rule for Allegretto as (40): (40) eontinuant! if continua t seg] [ “eetidone eer I $85] cco] 4.2 Recall that the last five examples of (38) contained voiceless segments which we transeribed as (af, 5% ¥*]: afptJourdo, ciufat], Y6"]qutesr, sel8°], and aly*]ea.7° tr is immediately apparent that the voiced continu ants [A, $, ¥] become voiceless before voiceless obstruents and in final position (more precisely, in the environnent WW). Further data, hovever, indicate that closer scrutiny is warranted. We will limit our observations so careful speech (Andante) since in more rapid speech (a) additional assimi- lations occur which complicate the picture, and (b) the fine distinctions discussed here are extrenely difficult co perceive (if they exist).”? Consider the contrasts illustrated in the following examples (for simplicity I give examples of dentals only; similar contrasts exist also inong labials and velars):b5 a. voiceless [t]: b. voiced [t4J: atmésfera, dtnico, fitbol c+ voiced [a]: cuando, donde d. votced[%]: admiro, padre, adldteres, amigdalas @. voiceless (84: adi izir, adseribir, adjunto (j =(x), sed, ataud The contrasts illustrated in these examples could be marked as in (41): «ty 5 e etoig 8 coneinuant == = = + + tense toto eee voice tore e @ specifications indicated in (41) are sufficient to mark all the concrasts and are quite plausible, recent instrumental investigations << gest that (41) might well be refined in the direction of greater acoustico- articulatory accuracy. I give now summary sketches of two relevant studies. Lisker and Abramson (1964) have measured the onset time of vocal cord vibrations in the vowels following the release of stop consonants in « num= ber of languages, including Spanish. From their seasurements it may be con- ciudad that the onset times of vocal vibrations fall into four distinct 22 categories. (a) onset of voicing precedes stop release (b) onset of voicing substantially coincides with stop release (c) onset of voicing lags moderately after stop release (d) onset of voicing lags considerably after stop release. Lisker and Abramson have shown chat in initial position, spanish [t]] falls into category (b) and Spanish [d] falls into category (a). The span of46 prevoicing of Spanish (d] is on the order of 100 milliseconds, which, al~ though Lisker and Abramson do not specifically point it out, is clearly audible under good acoustical conditions.” Kim (1965) has deseribed three phonemically contrastive sets of voice- less stops in Korean, which can be represented schematically as : voiceless, unaspirated 1p voiceless, lightly aspirated 3 | voiceless, heavily aspirated. Pressure measurements made by Kim indicate that f, and ty, but not ty, are produced with heightened subglottal pressure. Further, t;, falls into cate~ gory (b), virtual coincidence of stop release and vocalic onset; t, falls into category (c), with moderate lag of onset; £3 falls into category (4), with considerable lag of vocalic onset. Thus Korean t, and Spanish t are in the same category. By an odd coincidence, there ere Korean words phonetically identical to Spanish words (though of course different in meaning). For exemple, Korean tal, Spanish t The words "phonetically identical" were not used loosely: Mr. Kim accepts ay pronunciation of Spanish tal as an absolutely native Korean pronunciation of ¢)al; I find Mr. Kim's pronunciation of tyal utterly indistinguishable from a native Spanish pronunciation of tal. Thus tt {s possible - though hardly necessary - that Korean £1 and Spanish Ce J are produced with idential articulatory mechanisms and therefore should nave the same feature specification. Largely on the basis of the investigations of Lisker and Abramson and of Kim, Chomsky and alle (1968, Chapter 7, Seeticn 6.2) have proposed that47 Korean £1, fp) and ty be assigned the following feature specifications: tense toe oF voice toe - Heightened subglottalp + 2 - + pressure Glottel Con") + = > striction 1 cannot reproduce or even sumarize here the intricate argunent which Chomsky and Halle give to support these specifications. The pointmost ger- gane co the present discussion is the assignment of the feature [+votce] to nyoiceless" t,.°4 This feature is correlated with the nonspread position of 1 the vocal cords appropriate for voicing; but £1 is not "voiced" because of the tenseness of the supraglottal musculature ([+tense]), and glottal con- striction. Onset of voicing of a following vowel is simultaneous with re- lease of the glottal constriction, however, since the vocal cords are already in voicing position. In fp, on the other hand, there is a moderate lag in che onset of vocel vibrations since the vocal cords are not in voicing posi- tion when the stop closure is released. For the sake of argument, let us suppose that Spanish (thas the same feature specification as Korean £,, and see what rules would be necessary to ssign the specification correctly. For completeness, we will consider all che rules in the grammar which play a role in assigning features to the seguents sentioned in (41), including two early “sorphophonemic" rules which will be treated in more detail in Chapters III and IV. The first of these rules {48 [-sonorant] —s [+eensey__[ stonae”™*| This rule is needed to account for the alternations illustrated in the under- Lined segnents of describir -- descripeidn and legible -- lectura. ‘The next is: ‘-sonorant] —+ [-tense]/V__[#vocalic] (under certain conditions) ‘This rule is needed to account for the alternations illustrated in natacion-nadar, recipiente-recibir, and persecucion-perseguir. It is highly restricted, applying only to a small subset of the lexicon. The next rule is the Spirantization rule discussed in the previous section. After appli- cation of these three rules, the segments of (M1) will be specified as follows: t dood 5 8 continuant == = - ot F# vense + + - oe voice - + oF oF Now recall thut we already have in the grammar a rule which voices s in certain environments, which for convenience is repsated here as (42): seontoaneat | (a2) sz (78) | Hotce clearly mast generalize (42) so that tt will yield Ct4] and C6") in cus iypropriate environments. But there is a problem: $ voices befors voiced obstruents (de: 0), nasals (mis*mo, ag’no), and, N.B., liquids (is21s, aus2lo -+ a does not occur before [r]); noncontinuant obscruents, va the other hand, voice before voiced obstruents (coflak® bueno -- examplee49 are hard to find since p, t, k do not occur in final position in native words), nasals (at4ndsfera, tek@nico), but, N.B., not before liquids (aplicar, aprecio, atleta, atroz, aclarar, acreditar). Thus, apparently, tule (42) cannot be generalized to account for the voicing assimilation of noncontinuant obstruents and still maintain the simplicity implied in the present formulation. But aow recall that, ex hypothesis, ve mst tssign more features to [e] chan those listed just above. Suppose we insert the following rule before rule (42): 43) tvoice seg, [‘sonorant wisrressure | /__(( “$55 (see | splot. con. : cont inuant | | +tense Then rule (42) can be stated simply as (44): | +consonancal ] | avoice sonorant | -HsPressure [> [-voice ]/__# It {s easily seen that (43) and (44) assign che appropriate features cor- rectly to segments incompletely specified as in (41). We now make the following concluding obs ‘vations: a. We have found motivation internal to the grammar of Spanish, aamely the problem of the reformulation of (42) to account for voicing assimilation of noncontinuant obstruents (obviously 4 grammar would be less highly valued with two nearly identical obstruent voicing assimilation rules), which lead us to postu- late some entirely nonobvious feature which distinguishes some noncontinuant obstruents érom others.50 Attention to phonetic detail, in particular the simultaneity of voicing onset after[t], has suggested certain nonobvious feature specifications. c+ It turns out that the features suggested by 2 and b above are asonant with theoretical observations arrived at independently of the present data. Specifically, it is tense noncontinuants vinieh are assigned the features [+votce, +52, 460] in corcain environments. d. It is interesting that the occurrence of heightened subglottal pressure is coincident with syllable onset. For example, atleta with (t] is syllabified while atmésfera with (t7] is 25 L 2+ Instrumental investigation of the dialect under study, which is badly needed at this point, may confirm the proposals made in this section, or it may show that they are in need of re- finement and correction. In any event, we have surely disclosed an area in which recent theoretical innovations may be confronted with a valuable body of empirical data. The Nonlateral Liquids Te is well known that there is a phonemic contrast between the inter- vocalic segemencs ed x and those spelled rr. Minimal pairs are plentiful: pero/perro, caro/ amara/amarra, torero/torrero, and rr perhaps hundreds more. For ease of discussion, let us use che symbol (r] for the segment spelied x intervocalicaliy, and, temporarily, the non~ commital symbol R] for the segment spelled zr.ou Presumably, both {r] and [R] are liquids, distinguished from [1] (and from (J in the dialects which have this palatal liquid) at least in being nonlateral. In the theoretical framework which we accept, one must eventually determine what phonetic features fully characterize Cr] and{a], and, concomitantly, distinguish [R] from[ +]. This is, to ay knowledge, completely unexplored territoy, and the task does not prom ise to be a simple one. If it is to be accomplished in a principled vay, consideration must be given to (hopefully, the convergence of) inferences from (a) acoustico-articulatory data, and (b) the rules that involve [r] and [a]. 5.1 Acoustico-Articulatory Data: The segment (r]is simply a voiced apico-alveolar single flap. We may confidently assign to it at least che features [ +vocalic, teonsonantal, +sonorant, +voice, +coronal, tanterior, sdistributed, -strident] , and, with only slightly less confidence, 2 The specification of [8] is more vexed.” As 2 (+continuant, -tens first approximation, let us say that (R] is a voiced apical trill, or multiple flap, where "multiple" means greater chan 2. 4.B. that (rJ is exactly one flap; anything ore will be interpreted as (21, in any style of speech. It is safe to estimate that, on the average, the trill consists of 3 or 4 flaps. In highly emphatic specch, however, the rill is often quize prolonged, and may, in extreme cases, reach as aany as a dozen or so flaps. In addition co this fully trilled phone, there also occur in the dialect under study the phoaze vhich we now describe, which we have sub- sumed under the symbol (R]:a, In nonenergetic speech, particularly at tow volume, [R] followed by a vowel, sither vord-initially or intervocalically within a word, ceases to be a trill and becomes som sort of fricative. It seems that the articulatory phenomena involved are the following: the air stream does not act on the tongue with suffi- cient force to initiate or maintein a trill, for which consider- able energy is required. (The trilled [8] ts quite “tense,” using this as an impressionistic term.) This fricative is voiced, and quite "strident" (again, impressionistically). The ton- gue seems to be slightly retracted or retroflected. Let us use the ad hoe symbol [2] for this phone to distinguish it from fully eritied Ca]. >. Erratically, in pre-pause, but not pre-consonantal, position, a voiceless apical fricative occurs, which I vill represent as BJ. tis |] is distinct from the normal alveolar [sJ: comard# [eomé8], versus tomfsi# [conds]. [3], 1ike its voiced counterpart [2], seems to be slightly retracted. c. The cluster gr, whether divided by a vord boundary as in los eleos or within a word as in Israel (which, along with israelita, ate., is the unique example of morphene-internal sr), has a number of pronunciations. Navarro Tomés (1965, pp. 109-123) states that simply (R], with the s completely absorbed, is a normal Castilian pronunciation of this cluster. This pronunciation is not common, however, in the dialect under study; when it is heard, it is erally taken as an affectation. What does occur ranges from a voiced ¢ followed by a trilled[ &]to waat seems to be a53 single phonetic segment which I will transcribe [¥], because it seems to be very auch lixe che phone spelled ¥ in Czech. Now the question arises as to whether (2] and (¥] are phoneti- cally distinct. The closest I can come to minimal pairs are 1/Israel and irreligioso/es religioso. ‘hen I pronounce these as[ Ledt, Wadi] and[i2oitiyse0, S¥elixyseo], t feel that [2] and[¥] are different, but I cannot objectively hear any difference. What inference might we draw from these data about the features of che segments under discussion? First, (8) and [¥ ¥] are surely dent , theoretically as well as impressionistically. (Ir would be a con- ceivable but rather implausible extrapolation to say that the trilled(RJ is distinguished from the flap [r] in that [R] is [strident] while [r] is (-strident].) In any event, [r] and[R] should be distinguished by tense- ness: Cr] ts [-tense] and [XJ is [+tense]; Navarro (1965) states emphati- cally that “la tensidn muscular es en (R] macho mayor que en [r]" (p. 123). We have used the locutions "retracted" and “retroflec' " above. 1 assume chat in che segments under discussion, “retractedness” should be characterized with the features anterior, -high]. However, neither ay own kinesthetic sense nor Navarro's drawings and comments (1965, p. 121) are precise enough to permit us to assert categorically that some of the segmeacs under discussion should be specified as C-anterior, -high] and others not. It seems to be a reasonable guess, however, chat [R] is [anterior], since it seems co have the sam point of articulation as C+], sich is clearly alveolar, i.e. [+eoronal, ‘anterior, -distributed], as are che nermal and g. On the other hand, C4] and C2] {[¥] seam cooh differ from[s] and [5%] only in their slightly retroflected articulation; hence I will assune chat the former are [-anterfor, -nigh]. I give in (45) a first hypothesis, based on acoustico-articulatory data, concerning the feature specification of relevant segments. The specifications enclosed in parentheses are, of course, extremely tencative. ne (45) 2 vocalic . Go - consonantal sonorant anterior coronal distributed high contiauant tense voice strident - G Cette ed EF4T 4 + feeds fete vebeeee beet bedtbee +4 > 5.2 Let us now approach the problem from the other side, by investigating the rules chat (rJ, (RJ, etc., figure in. For ease of exposition, let us make a bipartite stylistic split, and consider more careful speech (Largo and Andante) separately from aore casual speech (Allegretto and Presto). Further stylistic differentiation seems to me to be pointless here. 5.2.1 I now describe ti discribution of flapped(r] and trilled(R] in careful speach, assuming that (¥)4[2] and [3] occur only in casual speech. 2. Both [r] and {R] occur intervocalically, as illustrated by pero/perro [pé&ro/péRo], etc. >. onty [2] sceurs vord-iniziatly and presumbly after 2. as an example of the latter case, subrayar, "to underline”, is35 [eupRa...], the underlying representation of which is taken to be /subera.../ . Note carefully chat "word-initially" must be interpreted as /#__, not as /#t_. For example, 1s ropa is [laRépa], not *(larépa]. Only [R] occurs after 1, n, ands. For example, alrededor, honra, and Israel have 18, nf, and sR, respectively. However, the following observations should be made: (a) Alrededor is UJ and this is the only example in the entire lexicon of Ir,” Matorieally two words, al and xededor (and this by metathesis from detredor, which in turn is also two words, de and redor, roughly). Thus there 1s no clear example of lr within a mor- pheme. (b) Honra, plus the related honrar, hontado, and a handful of proper names like Enrique, Manrique exhaust the in- stances of nr. (c) Israel, israelita, etc., are unique with s Thus these words might be considered exceptional, in the sense that there are few of them. They are not exceptional in che serse they would be if there were other forms with Ir, or, and sy instead of 1R, aR, sR . There are no such forms. Two further details: since bota final s and initial r are extremely coamon, the sequence s#r occurs very frequently, and Section 5.le sh d be borne in mind. Secondly, honra, etc. may be related to honor, honorable, honorifico, etc., although this is not at all clear. If so, then either the nr cluster in the former arise: from the delecion of a vowel, or, extremely implausibly, che aor sequences in the latter arise from the insertion of 0 in the ar cluster. This should be considered in assessing the “axception~ ality" of the forms discussed in this paragraph.56 4. Only {J occurs after consonants other than 1, a, s: compra, frito, ladrdén, tres, grito, crea, etc. only [R] occurs before consonants: arma, [ 4Rma], arte [aRee], 2 [éRse] , {rbot [ARI], ete. a onty [2] occurs in the savironment /__#. Cf. paragraph > above. It is crucial to observe that the environment in ques- tion there is /#__, not /d#__. In contrast, in the present paragraph the relevant environment is /_#, not /__#. For exauple, amor eterno, “eternal love,” is [amoreteRno], not *[amoReteRno]. If, however, the next word begins with a conson~ ant, [RJ occurs: amor paterno, "paternal love," iz [anogpateRno] - In utterance final position, [R] tends to devoice slightly if hg final syllable is stressed, and a little more than slightly if the final syllable is unstressed. In short, [r] and [A] contrast only in intervocalic position; post- consonantally, except after 1, a, 8, only [r] occurs; after 1, a, 8, pre- sonsonantally, and finally, only [R] occurs.(with the qualification of £ Juss avove). 5.2.2 The distribution of [r] and [a], phonetic details aside, (i.e., disre~ garding (3, 3, FJ) in casual speech is as described in 5.2.1, with the axception shac [2] sather chan (R] occurs in the environment __¢. Thus arma is (48ma] in more careful speech and (4rma] in more casual speech.3? +3 Lexical representation of (r]and[¥]: the phonological representa- tion of [rJ and [R] intervocalically is /r/ and /zz/, respectively. For example, pero and perro are /pero/ and /perro/. ‘The representation of imtervocelic [R] as /rx/ is based on just one familiar set of examples which are extremely convincing. The future tense forms of tas verb auerer ae quered, querrd4s, querrd, querramos, querrdn, phonetically [kers, ate, ss]. Te vill be argued ta chapter IIL, sections LL and 12, ehae these forms are durived from the following representations, irrelevant details ignored: /ker+ré, kertrds, kertrd/, etc. That is, the future end- ings, -ré, -rds, -rd, etc., are attached to the stem ker=, which brings to- gether the final r~ of the stem and the initial -r of the endings.°° Thus it {s unarguable that at least some instances of intervocalic [R] are de~ rived from /rr/. It is therefore reasonable to assume, until evidence is presented to the contrary, that all instances of iatervocalic [R] have the same source. It vas shown in 5.2.18 and ¢ that [2] is predictable in the environ- ments of (46): <4) a 7 [338] _— be e/a de /s__ Therefore we could write r in these environments in underlying representa~ sions, and get [R]by rules. Alternatively, we could write #r, L grr. It is clear that che second alternative must be rejected in favor of che first: there are literally hundreds of words vith initial [R], and none with initia {:]; cave ara a uandful vith (1a, na, sa],58 notion of “simplicity” might be rita [le, ar, sr]. Therefore some vag 4 which would demand that the first alternative be selected as avi yielding a more highly valued grammar. More to the point, however, is the fact that if we do not choose the first alternative, we leave unstated the fact that non-occurring *[ir, lr, nr, sr] are not accidental gaps in the lexicon. Section 5.2.le, £ and 5.2.2 indicate other environments in which the Securrenes of(R] is predictable, namely __C and __#C in Andante, and —# in doth Andante and Allegretto. It is presumably beyond question chat che underlying representation should be r in these cases, and the occurzences of [R] are given by rule. Suffice it to mention that phonetically final (R] is, in the vast majority of cases, intervocalic r at a higher level of deri- vation, as is shown by hundreds of singular-plural pairs like amor/amores, Aonor/honores, tambor/tambores, pajar/pajares, alfiler/alfileres, It is clear that the phonological representation of ¢.g. amor/amores is anore/ amorets, where 5 is the plural formative, and where the final e¢ of the sin- gular amore is deleted (see Chapter III, Section 4.3). Thus we must have les with the effect of (47): (47) rR ES coaeT / __(#v In short, we have shown that the distribution of [FJ and (RJ is pre- dictable except in intervocalic position. de have shown that in this posi- sion {x] and [R] are represented as /r/ and /er/, respectively; in all other positions (R] is sepresented as /r/. .2.4 Rules. Let us examine the environments listed in (46). In order so sutract som generalization from this List, we must compare the sequences59 lr, az, sx with che consonant clusters vates have Cr] zather than Ca] as tae second member. These are: pr, br, fr, tr, dr, x, gr. (As far as T known, xr occurs only in Jruschef [xrusté£]; *mr and *ér do not occur at y which figure in (46) are all dentals, ise. [ +coronal, all.) Since 1, a, santerioz], we need consider only tr and dr of the other Cr clusters. The set [i, n, s] are distinguished from[t, d] in that the former are [-distributed] while(t, dJare [+distributed]. Thus we may account for the occurrence of[[2] in the environments of (46) with rule (48): (48) y (3s roa ]irs teoronal 7 | -distributed Note chat (+coronal] is required in the environment of (43) in order to veclude £, vhich is also [-distributed]. Now let us look more closely at (47), which is only an approximation to the correct rule. First of all, we ust replace (47) by (49); 49) rR xcept / __(#) [-consonancal ] since [vocaiic] plays ao role in the environment as is shown by the follow- ing examples: Jatlos, Avol, out 22r0, Maryo. Now we aust state the envir- onment of (49) positively. 3ince [2] occurs in __[teonsonantal J, __(#) [vconsonantal] , and ___# in Andante; and in __## in Allegretto, we replace (49) with (50): (8) [4eonsonanea J] f (30) a. Andance: rR le :3i/_#60 One 13 immediately struck by che similarity between (50a) and (375), whicn, for the convenience of the coader, wa repeat here as (51): _ (#) Econsonantal | Gv Cvs J [vsonsonantat] / 1 f = / 05 of x affected by (50a) vill always be preceded by a vowel, and since there is no reason not to order (50a) contiguous to (51), (50a) and (51) would collapse as (52), provided that the X in (52) could be specified: (52) [ xconsonantal @ [xconsonantal]) svocalie 3 a/v | -lateral * But the coincidence doesn't end here: (50b) is matched by the Allegretto rea further, (37¢), second case of (51). version of (51); that is, led by the first case of (48), which we repeat here as (53a) is parall snich we repeat as (53b): 33) a. Ly, 2]+ Cveonsonaneat] [73 appreaching = must be kept in mis61 ingle feature [[+consonental] appears to the right of the arrow, which would seem to be roughly correct for just these rules. In other words, it seems that some or all of the features other than [+consonantal ] which are necessary to specify the changes y~¥, and wy", namely [-sonorant, Helayed release, -strident, -tense, +voice, ete.|, should be supplied by convention. ?2 Whether some or all of these features are supplied by convention or whether they must be stated in the rule itself, the point is that more features change than just (Gonsonantal|. Thus the complete specification of (51) and (53a) would include at least all the features mentioned in (55): -sonorant 7 (55) -consonantal| _, +consonantal -vocalic tdelayed release | /... ~tense Now, if (55) is to be combined with r + R/... as in (52) and (54), as it clearly must, one plausible and particularly simple way of specifying X would be X= [Wsonorant, xtense] where presumably other features would be supplied by convention. This being the case, for [y, w], (52) and (54) would obviously give [Yj] and CY] as intended; for [x], (52) and (54) would give the tense sonorant segment s* = [ (-vocalie?), +consonantal, +sonorant, anterior, +coronal, -distributed, -high, +continuant, +tense, +voice, -strident, -lateral], which is clearly distinct from all other seg- ments of Spanish. Now note that S* has the sam features postulated for [a] in (45), with the possible exception of (vocalic], This is a remarkable convergence: the features of [R] in (45) were chosen solely on the basis ofacoustico-articulatory data. The features of S* were arrived at solely on the basis of rule simplicity, Thus we may state with some confidence that [R] is either a tense liquid, or a tense sonorant consonant. At this point one could make an arbitrary choice, but I prefer to leave the ques- tion open, Let us return to the question of rule (48). We have transferred the first case of (48) to rule (54); we are thus left vith what may be stated as (56); edistributed [ +eoronel ] (36) ra] C8888fse) | / I do not know how to order rule (56). On the one hand, I foresee no disas- trous consequences of combining it with (54); yet it does not seem intuitively plausible that a rule so far from the end of the grammar should refer to the feature [distributed] in its environment. On the other hand, it seems that a case could be made for somehow combining (56) with rule (39-40), Spirantization. Note that for dental d, Spirantization applies after seg- ments other than 1 and a, while (56) should apply after 1, n, ands. In these environments, d (and b, g in the corresponding environments) and r seem to undergo som sort of "strengthening of articulation" (cf. comments at the end of Section 2.6). At our present stage of knowledge, however, there seems co be no principled way of stating this generalization, if in fact it is a significant generalization. Let us now cie up what loose ends we can. Since intervocalic /rr/ ends up as (RL indistinguishable from che [R] which derives from single x by (52) or (54), we apparently need a rule with63 the effect of (57), which must be ordered before (56): (37) rr>R since [R] is optionally realized in Allegretto as(%], and in some environments [3], we apparently need a rule with the effect of (58): ‘ssonorant ] (58) Allegretto: R|-anterior OPTIONAL Presumably the feature [+strident] will be supplied by some convention. ‘The [2] resulting from (58) will be correctly devoiced in the environment __ftt by rule (44). mmary of Rules and Further Observation 6.1 T now collect all the rules proposed so far, renumbering them as (59), and placing them in the proper order. In most cases the reason for the ordering is obvious; in a few cases the ordering depends on considerations to be taken up in Chapters I2I and IV. (59) a. (25) Mw / A f bd. (24a) } ce. (57) rr BR xconsonantal a. (54) Sorte ley pacers css] j cgeatte | comes | 7 Liss) T+tense 7 F seocona? 2. (96) Fa|Cvecalte) | / | ~dlsertiveea64 sconsonantal | y f£. (24) Ss — [-voealte J if — astress \v. [eanterior ] “‘teonsonantal r coronal eanterior Andante (15) [[4nasal ]| rhigh J. | coronal $back Vhigh Gdigtributed back @distributed a Allegretto Gs) Cinasatbt I_cseb 0 * J] | +consonantal = - +anterior Andante (17) 1% [+diseributed] / | +coronal +distributed he Allegretto 1) Cadiseributed) /__(735*)0 "J [gonorene] _, [Ssossinane] {Esconsinenl Andance (39) | -tense +strident [ecoronai] orm co. [abegretto (40) 1 u - Jf Jesh] r. [sconsonantal (#) [consonance 1) Andante (52) lfpevezetze 4 ffeseeeees ne i -lateral «tense =! \ + J alegretso yf * J>C °] me6.2 65 ~sonor: k. Allegeteo (58) n9[7E0n93°"* | oprzonan 1. (43) cont inuan +tense [tvotce J cap [+ : \4| Hispress | /__(|7j ) ‘sonoran| JL seconsee me anes ] -sonorant | [favoice]/__([7$$8]) [ Zeonsonancal a. (44) Saeatael t J [8°] svoice ] [-votce] /__ iit y a. Allegestto (29) w oi The ad hoc symbol M#, which represents silence, lack of phonation, appears at two distinct points in (59). The phonological theory of Chomsky and Halle (1968), aowever, makes no provision for the incorporation of such a symbol. The purpose of this section is to present evidence which suggests that Some such provision must be made. Note first that it would be possible to avoid mention of # by che srcifics of substituting rules of the form of (60) by sequences of rules of the Zorm of (61), but the incorractness of resorting to such an artifice is obvious, and aerely underscores the theoretical inadequacy under discus- ston: (60) (6) a A338 /_# b AB a AaB 3 4 /__[+segment]66 be A7 B B= A /+segment]__ Te might be thought that the symbol ii is equivalent to the J(uncture) Rint) proposed in Stockwell (1960), which marks the end point of intona- tional contours, or the boundary symbol Of proposed by Bierwisch (1966). This, however, is not the case: M# refers Literally co silence, nonphona- tion, while JP and Qf are associated with intonational phenomena such as retardation in tempo, rapid rise or fall in pitch, ete. For example, a two-sentence utterance such as s dos. Démelos, "...both of them. may have either of the pronunciations illustrated in (62). (Clozééz}ddémelos}] [loz66eféémalos 4] The symbol $ subsumes all the phonetic properties of the (falling) intona- los tonal contours of both sentences of Démalos; (62a) and (62b) are to be interpreted as having identical contours, Thus i may or may not accompany a terminal intonation. The voicing assimilation cf s is broken only by the intervention of total absence of phonational activity, as in (628), although the Spirantization of d may be blocked by the occurrence of (inaudible) # or = (Cf. Section 4.1). One might speculate that the phonational activity which accounts for the voleing of o in (62a) is the enset of vocal vibration which precedes the release of d (Cf. Section 4.2, discussion of findings of Lisker and Abramson). Ia any event, (62) shows clearly that the domain of phonological processes such as voicing assimt- lation is aot limited to the boundaries of "phonological phrases" or "Phrasierungseinheiten,” as has generally been believed.5.3 Observe that the sequence of rules (59) has the following properties: a. (59) may be divided inco three blocks: rules a-f are shared by Andante and Aliegracto; rules have nearly identical formilations in Andante and Allegretto; rules -n are either shared by Andante and Allegretto ( 1 and a) or exist oaly in egretto (k and n). b. Rules g - i are idential in Andante and Allegretto except for the optional occurrence of 7888] ta the environment of the Allegretco version of each. c+ In Andante, i is the last rule for witch the presence of [- | in a string of units blocks application of the rule: that is, 239] after i, all rules allow for the optional occurrence of [7 in the environment. d, In Allegretto, £ is the last rule for which the presence of sf] blocks applicacion of a rule:?? that is, after £, all is | in cuies :ithse allow for the optional occurrence of the environment, or apply in all environments. Thus after cule £ in Allegretto and rule 4 in Andante, che presence of youndaries (other than M@#) in strings of phonological units is simply irre- levant to the applicability of phonological rules. Clearly the formulations of the rules in (59) does not properly reflect this fact: if ve vrite rules with parenthesiced boundaries, neither do we recognize tie che boundaries are irrelevant (since we aust mention them in =! rules) aor does their irrelevance give us stapler rules. ‘thi 2 all of this suggescs chat one of che formal correlates of taa stylistic distinction Andante/68 Allegreteo ts not the appearance of optional boundary elements in chs rules of one style but not of che other, but rather that boundary elements are Jslocad in phonological representations at a higher level of derivation in Allegretto than in Andante. Specifically, boundary elements are deleted after application of rule £ (or ¢ - see noe 33) in Allegretto, and rule £ in Andante, and no rule need specify optional boundary elements in its environment. Chomsky and Halle (1968, Appendix to Chapter 3) aave proposed that the features of boundary elements be deleted from phonological representa~ tions by convention at the end of a sequence of phoaclogical rules. The observations made in this section suggest that this proposal must be re~ fined in som way,4 69 EOOTNOTES TO CHAPTER IT Forms like inmenso and innato should probably be roughly /in=menso/ and /insnato/. For discussion of the boundary =, which would block the application of rule (4), see Chomsky and Halle (1968, Chapter 3, Section 10). Fuller discussion is presented in Chapter IV, Section 4.2. See Chomsky and Halle (1968, Chapter 4, Section 2.2 and Chapter 8, Section 7). ‘This is automatically accomplished by the convention that # blocks a rule unless it is explicitly mentioned in the rule. For discus- sion see Chomsky and Halle (1968, Chapter 3, Section 1.3.1). See footnote 1. For discussion of the feature composition of soundaries see Chomsky and Halle (1968, Chapter 3, Section 1.3.1). See, for example, the interesting recent article by Fudge (1967). It is not clear what King means here. In fact, ( gy] and (w] are “in contr tt" only because of King's decision tc phonemicize them a certain way. The last two examples also occur as /sénwantes, songuépes/, which, by King’s own analysis, shows exactly the oppe-10. 2. 70 site of what he seems to be saying, that is, that Cgy] and Cw] contrast only stylistically, not phoemically. We return to this below. Diphthongization will be treated more fully in Chapter 3, Section 13. For expository purposes, we will henceforth use the ad hoc device of representing as upper case E those e's which diphthong- ize when stressed. For further discussion of rounded velars, see Chapter IV, Section 8. ‘There are other instances of WW versus GV contrasts which have other explanations. For example, dueto, (dueto], "duet," versus duelo (éwelo], "mourning." For these, the representations /du = eto/ and {d0lo/ account for the contrast. For a different kind of examples, see Chapter III, Section 9.1. See, however, Chomsky and Halle (1963, Chapter 8, Section 4), where a feature system is discussed in which obstruents, nonsyllabic liquids and nasals, and glides share the feature [-syllabic] while vowels and syllabic Liquids and nasais are [#syllabic]. The havoc caused by such data isa miter of history: oven and Stock~ well (1955), Saporta (1956a), Bowen and Stociwell (1956), steckwell, Bowen, and Silva-fuenzalida (1956). Stockwell et al (1956) made the xtraordinary statement about the glides that "the language really is not very neat at this point" (p. 411). No amount of fiddling with “plus juncture" did any good. For example, ayuna, "fast (mot eating),"16. a and hay una, "there is one," are both Cayjuna] in Andante and (ayuna] in Allegretto. Stockwell et al. observe with consternation that if Layuna] is phonemicized as /aytuna/, the plus is in the right place for hay una, but in an embarrassing place for ayung, and so on- Aullido is Cawyyi6o] in Andante, but there is strong reason to be~ Lieve that the representation is not awyido at the time (24¢) applies. Incidentally, it is extremely hard to find relevant examples, The reader will note that the examples in (34a) contain only aw and ey- Aside from a handful of exceptions, some unique, there are no in~ stances of ey, ow, iv, uw, iy, uy in the language. Rules will be given in Chapter 3 which account for the absence of these sequences. B versus v is an orthographic, not 4 phonetic, distinction. For ex- ample, hierba, "herb, grass," and hiexva, "boil (present subjunctive)," are both Cyérfa]: There is no (v] im native Spanish words, however, in Afganistén, afgano, the £ becomes voiced before the following voiced consonant. Similarly un_chef bueno, “a good chef," may be pronouned with a voiced £. The voicing of £ is of infinitesimal importance per se, but it will fall out automatically from an independently aotivaced extension of the s-voicing rule (37h) to be given in Section 4.2, she eynbots a, 32, and yS seproaent partially of completely voice- less segments. We will ignore these seguents until Section 4.2. =sonorant] r 7 We postpone justification of [soner i rather than [;eonoren until Section 4.2.19. 20. a1. 22. 23. 7 The features (-seg, -FB] seem to have vanished in the transition to (39). This is due to the convention that the presence of # or ‘= in a string of units blocks the application of a phonological dé in the rule. For rule to that string unless # or = is mentio discussion see Chomsky and Halle (1968, Chapter 3, Section 1.3.1). Although club and Agfa are not of native origin, they are in coumon use. I have cited a im order to illustrate assimilations of b and g not otherwise found because of limitations on the distribution of b and g in native words. Incidentally, in the dialect under study the plural of club is clubs (klup%s J, not the expected ¥elubes. The reader is also advised that not all dialects share every detail of the data presented here. Lisker and Abramson distinguish only three categories, collapsing (c) and (4). The reason for the four-way distinction will be apparent below. On the other hand, initial English t falls into category (4), while English d falls into category (b). Thus Spanish ¢ and English d are in the same category. This would aelp to account for the fact that English speakers learning Spanish experience extreme difficulty in distinguishing (in isolation) pairs like tos/dos, ¢ia/dfa, Tezcoco/ descoco. For many years I have found it pedasogically helpful to call students! attention co the prevoicing of Spanish d.2. 26. 3 che_sky and Halle have proposed that the distinctive feature [voice], as opposed to the impressionistic terms "voiced" and "voiceless," be restricted such that [voice] characterizes sounds produced with @ glottal opening that is so wide that it prevents vocal vibration when air flows through the opening. The feature [+voice] character- izes sounds produced with an aperture not so wide to prevent vibration. In these terms, sound which are [+voice] may or my not be “voiced": vocal cord vibration will not result although the glottis is in (voice ] position if there is (a) closure in the supraglottal tract, and (b) sufficient tension in the supraglottal musculature to prevent expansion of the supraglottal cavity, such that the flow of air chrough the glottis is impeded --without air flow through the glottis there can be no vocal cord vibration. In some dialects, perhaps most, atleta is atcle-ta. I have not in- vestigated the range of data presented here in these dialects. It is an intriguing question what influence the many indigenous words with initial and final tL] clusters in Mexican Spanish (1! » tlapalerfa, L, Bopocatépetl) may have had on the syllabification of words of European origin. The graph rx is used only intervocalically; is used for (R] in other positions, e.g. initially where’ no cofitrast with (r] is p0s- sible: [Roto] is spelled roto. Thus the spelling system is both anambiguous and economical in this respect. Amancio Bolafio e Islas, professor of phonetics at the Universidad cional Auténoma de México, once claimed to 2e able to distinguish no less than 37 varieties of (R] in the Valley of Mexico alone (personal communication).28. 29. 20. 74 I disregard obvious cases of composition like na ‘The cceurrence of both [érmi] and [4Rma] , and so on, provides the basis for a devastating, though hardly unique, argument against the (taxonomic) phoneme as @ perceptual unit. Since [r] and [8] contrast intervocalically, they must be phonemically distinct: /r/ and /R/ or /rr/. (The familiar and rather idle controversy over whether [R] should be phonemicized as geminate /rr/ or as a unit phoneme /R/ has no bearing on the issue at hand.) Now if the pho- neme is @ perceptual unit, native speakers should be able to per- ceive immediately the difference between [Arma] and [éRna], since these are phonemically /arma/ and /4Rma/ (or /drrma/ --these trans- criptions are copied from the Literature, not invented.) In casual but often repeated experiments that I have made, however, phonetically untrained native speakers consistently fail to perceive any difference, even under optimal acoustic conditions and after the difference has been specifically pointed out. On the other hand, native speakers of English usually hear the difference at once, even though this sort of Phonetic contrast obviously plays no role in English at all. if one took such facts seriously, one would be forced to the conclusion tnat, far from being alert to phonemic distinctions in their own language, Spanish speakers are actually deafened by the sound system they have acquired. It will be seen in Chapter III that the forms in question devive w mately from representations which can be stated slightly more accurately as /kErtrBtaty/, /kirtrBtats/, /kir+rEta+®/. It would take us inceler- ably far from the present discussion to justify these representaticns31. 32. 75 here. The conditional forms querrfa, querrias, etc., are analogous to the future forms in that a suffix with initial x is attached to the stem with final x, giving /...rtr.../, which becomes phonetic [R]. If it could be shown that honRa and honorable are related synchroni- cally, and therefore share the formative which can be represented roughly as /honor/, this alone would be decisive evidence in favor of the first alternative. More specifically, the deriviation of honRa would of necessity contain the following steps: honorta + honrta > honRta. The converse situation is clearly impossible: honorrtable ~> honorable. However, the relationship between honRa and honorable is not clear enough to count as evidence for or against either alter- native. Some additional evidence against the second alternative is provided by the fact that geminate consonant clusters apparently occur only after vowels in lexical items. It would be odd if rr were the only exception to this generalization. It would be pointless to state how the marking conventions proposed by Chomsky and Halle (1968, Chapter 9, Section 2.1) would affect these rules, since these marking conventions are undergoing substan- al revision at this writing. Further investigation would probably show that ¢ rather than £ is the last such rule, and that £ should be included in che block of rules vhich are identical in Andante and Allegretto except for the optional securrence of [es] in the environment of the Allegretto version. I have not gone into this since the cyclical assignment of stress to stretches larger than a word seems to be involved.76 CHAPTER THREE - VERB FORMS L,_Introductory Remarks The discussion in Chapter II centered around certain well known and widely studied phenomena, which, in some intuitively clear sense, are essentially "phonetic" rather than "worphophonemic." We now turn to 2 study of the morphophonology of Spanish verb forms. These have been chosen for study in this chapter because they present a rich body of clear data. Tais material is "rich" in two senses. First, in that the Spanish verb is highly inflected, with many phonological alternations in the various forms. There are inflections for an infinitive, two participles, “formal” and "familiar" imperatives, present indicative and subjunctive, two past indicative forms (traditionally called "imperfect" and "preterit"), two past subjunctives, a future and a conditional. All of these are in current use in the spoken language; that is, none is a “literary"form like the passé simple and past subjunctive in French. All but the infinitive, che participles, and the imperatives are inflected for three persons, singular and plural. The second sense in which the data is rich is that while it is readily available in any school grammar, it has so far resisted insight- ful analysis in any Linguistic theory.” This data is "clear" in that vexing questions of morphological “re latedness" simply do not arise: members of inflectional paradigm sets are "related" by definition. 2. Present Indicative 2.1 First Conjugation7 @) "love" singular plural Ist person no ankmos 2nd person fmas 3rd person dma aman The root is am; the vowel waich follows the root in all forms ecept éno is traditionally called the "theme vowel," which, for the first conjuga- tion is a. We may regularize the paradigm by postulating for dmo the underlying representation amtato, and deleting thenetic a with rule (20. 2) ve os Ha Thus all forms consist of the root, followed by the theme vowel, followed 3 by one of the person-number endings 0, s, 6, mos, a.” Since all the forms of (1) are stressed on thepenultimate vowel, we may state, as a first approximation, the rule for assigning stress in verbs as (3): Q) V = Cistress] / __¢,V¢g#ly Stress assignment vill be ordered after rule (2). Second Conjugation (4) “eae cémo —comémos, cémes céme cémen The root is com; the theme vowel of the second conjugation is e. Everything else is accounted for as in the first conjugation.rE 2.3 Third Conjugation (5) "Unite" dno unimos nes The stem is un; the person-number endings are the same as in the first and second cmjugations, but the theme vowel alternates between stressed i and unstressed 2. A few exceptions aside, no Spanish word has an unstressed 4 high vowel in the final syllable.’ Therefore we may account for the theme vouel alternation by postulating underlying thematic i and the following very general rule: 6) v (high) / __o¢ cstress| e The derivation of the forms of (5) will then be as shown in (7): untito untits unti untitmos untim 6 @ é é t a @) e e e 6) 3. Presenc subjunctive 2) ame anémos cdma comémos dna unamos ives cémas nas ime amen cdma coman dina gnanNote that throughout (7), the person-number marker for [lperson, -plural] is @, rather than 9 as in the present indicative. It seems then that while there is a rule roughly of the form (lperson, -plural, +indicative] +9, there is no corresponding rule with ( -indicative] . Before attempting to account for what appears to be thematic ¢ rather than a for the first conjugation, and a rather thane and i for the second and third conjugations, consider the corsonantal alternations tLlustrated in (8): (8) a opako, "opaque"; opasidad, "opacity" sveko, "Swedish"; swesya, "Sweden" mistiko, “aystic(al)"; mistigtsmo, "aysticism" indikar, "indicate"; {ndige, "index" belga, "Belgium"; bélxico, "Belgian" andlogo "analogous"; analox{a, "analogy" conyugal, “conjugal”; cény:a2, "spouse" The examples in (8) show that Spanish must have rules with the effect cf 7. frsonsonaneai] malice Now notice that the alternations illustrated in (3) do not occur in regular verbs. For example, the present subjunctive of first conjugacicn saka- isgake- not *sase-; second conjugation protexe- has the subjunctive protexa- (as well as first person singular indicative protexo); third conjugation divixi- has the subjunctive dirixa- (as well as first person singular indi- cative dirixo). Now, since we already have the irdependencly motivated rule (2), we propose that the subjunctive forms be derived as illustrated in (10) with first person plural forms: (10) First Cont. Second Con. Third Cont Indic. __ Subjune. Indic. Subjune, Indic. Subjunc. saktatmos saktatetmos protegtetmos protegteta-mos dirigtiimos dirigtita+mos : 5 x x x x (9) ’ o ® 2) sakamos sakemos protexemos —protexamos diriximos dirixamos Thus we see by these consonant alternations that che vowel following the root in the present subjunctive is not an "altered" theme vowel, but rather, in imprecise terms, the “subjunctive formative." Were it not for these (and other) consonantal alternations, one might entertain the possibility of simply spelling out the theme vowel differently in the subjunctive than in che indicative and having no "subjunctive formative." For the sake of simplicity one would Like to have one single shape for the subjunctive marker in 211 three conjugations. But given the three different theme vowels for the three conjugations, it is hardly surprising that there should be more than one shape for the subjunctive marker. Given the independent motivation for rule (2), it is unlikely that 2 solution will be found simpler chan the one presented.4. Participles and Infinitives 4.1 The Past Partictple. ‘The term "past participle" is traditionally used in Spanish just as in English, namely, for the verb form which co-occurs in the perfect tenses with the auxiliary "have," haber in Spanish. Unlite French and Italian, Spanish never has genier and number agreenent between the subject and a past participle in the perfect tenses. ‘The past participles of the three conjugations are amédo, comido, and unfdo. Note that the thene vouel ¢ of the second conjugation appears as We will account for this with rule (10), which in Section 7 will be shown to be somewhat less ad hoc than it now sens: i igh as ao) [ed (taigh] / [past participle] ‘The past participle ending itself seems to be -do. However, there are a few irregular verbs in which the partigipial ending is attached directly co the root, without an intervening theme vowel. In these forms, the participial ending appears as puesto. We will say chen chet the participial ending is alway -to, and that the which occurs in regular forms is a result of rule (11): a) Cesonorant] > Cetense] / V__ [tvocalic] (under certein conditions) 4.2 The Present Participle ‘hat has craditionally been called the "preseat participle" in Spanisu corresponds morphologically (and in part syntactically) to the “present82 participle," or "sing form," in English.’ The forms for the three conjuga- tions are amdndo, comyéndo, unyéndo. These forms are invariable: there is never any number or gender agreement of any sort. The ending of the present participle itself is -ndo; ve must pick up a few loose ends in order to account for the vowels which appear between the root and this ending. At several points in Chapter 2 we referred to 2 rule, (59b) which diph- chongizes ¢ and § to yé and wd under certain conditions. We used the nota- tional device of representing as upper case B and 0 those instances of © and 9 which diphthongize under stress (reserving lower case ¢ and 9 for those which do not diphthongize when stressed). For expository purposes, let us hypithesize an ad hoc feature (DJ, memonic for “diphthongization," such chat Ce, +D] = and Co, 4D] = 0, and so on-® Now note that the theme vowel of the second conjugation must be rather than £, since it does not diphthongize when stressed; comémos, not *comyémos (and, incidentally, that the first conjugation subjunctive e is also not £: amémos, not *amyémos). Finally, since the present participles of the second and third conjugations are comyéndo and unyéndo, ve must postu late rule (12), which applies before the diphthongization rule, (59b): , Phase «zy w= bo / tado Note that (12), stated in its most general form will also assign the feature (+D] to che thematic a of the first conjugation, but this will have no phone- tic effect. Thus the derivation of tle present participles will be as in (13):83 (13) amtatndo: comte+ndo: untitndo 4 é £ @) 4 é é a2) ¥ ye (59) andindo conyéndo uayéndo 4.3 The Infinitive aay amar comer unit Note the final stressed syllable, an apparent exception to penultimate stress for verbs. But recall a rule, alluded to in Chapter 2, Section 5.23, which deletes in the environment __#. We provisionally give this rule as (15): : as) (a) -°/" Goce) | That is, 2 is deleted in final position after 2 vowel followed by zero or jugation subjunctive ¢. Literally hundreds of singular/plural pairs of nouns like red/tedes, horror/horrores, framés/franceses, fusil/fusiles, gan/panes provide ample justification for this rule.” Imperfect Indicative The imperfect" is one of the simple past tenses of Spanish; the ‘preterit is the other. The difference between the imperfect and the prau cerit is one of “aspect," in the familiar sense; the imperfect is past tense, imperfective aspect; che preterit is past tense, perfective as- pect. We assume that the phonological segments of the imperfect and preterit endings are supplied by “spell-out" rules of feature bundles which contain the features {+past] and [perfective]. 5.1 Imperfect, First Conjugation (a6) andba amfbamos amébas amdba améban With the phonological representations amtatba, amatbats, amtatba, amtatbat+mos, amtatbatn, where ba is the realization of the feature com- plex [+past, -perfective], the only thing which requires comment is the antepenultinate rather than penultimate stress of amébamos. To accomodate this form (and the rest of (16)), we will provisionally replace (3) by an: an Vv = (lstress} /__([-perfective] Sq¥Cy#y 5.2 Imperfect, Second and Third Conjugation 18) comfa comiamos unfa —_unfamos confas unfas confa comfan anga un fan ‘The examples in (18) present two problems: che theme vowel of the second conjugation, and the form of che imperfect marker. We may handle the85 former by replacing rule (10) by (19): 1 Cipast, -perfective Ed =Erhigh) / __ {ets participle] as) L This rule will be further generalized in Section 7. In the first conjugation, che imperfect marker is ba; in the second and chird conjugations it seems to be just a. Let us assume ba for all three conjugations, and postulate the following rule, which deletes the b in the second and third conjugation: 120) bs Ott ‘The motivation for (20) is admittedly rather slim. é1though there are many instances of historical loss of intervocalic b, I am not aware of any clear synchronic evidence aside from that of the inperfects of second and third conjugation verbs. It seems to me, however, that the available evi- dence is convincing. Consider the following: (a) if rule (20) were not ordered after rule (2), then would have to find sone other way co account for the fact that rule (2) does not delete che theme vowel in comrita, comritats, etc., (b) iba, rather than *{a, the imperfect of ir, could be accounted for quite naturally by simply making ir an exception to rule (20). The same is true of a few substandard forms Like czeiba for standard crefa, che imperfect of creer, and (c) ceteris paribus, the “spell-out" rule for the imperfect marker vill be simpler if ic can be given simply as [+past. sgerfective] + ba for all chree conjugetions. Imperfect Subjunctive There are two sets of forms of the imperfect subjunctive, which have86 nearly identical syntactic functions. (In some dislects these two sets are entirely interchangeable.) It is important to note that the two sets of imperfect subjunctive forms do not correspond to the imperfect and preterit indicative in aspectual contrast (see section 5). Both sets of inrerfect subjunctive forms are imperfective aspect; thus bo:h are syntactically analo- gous to the imperfect indicative. The syntactic analog of the preterit (indicative) is the present perfect subjunctive (waich also replaces the present perfect indicative). For simplicity we will consider only the so- called "za" forms of the imperfect subjunctive, although the other set of forms could be handled by trivial extensions of the rules to be given. 6.1 First Conjugation (21) améra andramos amdras andra amfran LT assume that the marker of the imperfect subjunctive is given phonolo- gicl shape by 2 rule roughly like [¥past, ~indicative] + ra. Note that the stress contours of the examples in (21) are the same as those in (16), 1 assume that the forms of the imperfect subjunctive contain the feature erfective], hence stress will be assigned correctly in these cases by u 5.2 Second and Third Conjugations (22) comyéra comyéramos unyéca unyéramos comyéras unyéras, comyéra comyéran unyéra unyéran87 ‘Tne only comment necessary here concerns the theme vowels. We already have rule (12) which lowers high vowels and assigns to them the feature [4D]. Accordingly, we will replace (12) by (23), which will convert thematic ¢ and { to Z in the imperfect subjunctives, so that these theme vowels will end up as ye. (23) vo is] 7+ {rs This rule will be further motivated and generalized below. (24) amé. amdmos — comi cominos enf — unfmos amdste comiste unfste amd andron — comyd comyéron unyS_—unyéron Obviously the preterit presents 2 number of complexities not found 12 in che forms surveyed up to this point. The first person plural forms are the only ones which present no difficulty at all. Second conjugation nematic 1 in comfnos (as opposed to present indicative condmos) is read- 13 {ly accounted for by a simplification of (19), which is given as (25): (2s) [v1 = Chien: (pase participle] . [lo] 7 Stated {Cipeeed Note that preterit andmos end unfmos are identical co the corresponding pre- sene indicative forms. I see no alternative to postulating different person-number markers chan in the other tenses, with che exception of first person plural mos- if » and third le assume for the preterit, second personal singular88 person plural ron, then che second person singular and third person plural forms of all conjugations can be accounted for by (a slight simplification of) rules we already have. Of the second person singular forms, only comiste requires mention, and here thematic i ratker than ¢ is accousted for by rule (25). For the third person plurals, the diphthongs in converon and unyeron by the second case of (23) (and subsequent diphthongization + if this is simplified fron tra to __#rV. We therefore replace (23) with (26): (26) wr fate 1_+ (Rt Note that (26) must not apply to second and third conjugation infinitives, waich at an early level of derivation are comere and unire: the phonetic forms are conér and unfr, not *conyér and *unyér. This can be handled by ordering (15), which deletes final £, before (26), so that the thematic and £ of the infinitives is no longer in the environment __rV at the time (26) applies. Ne give a few sample derivations in (27) to illustrate some of the forms that have been discussed. (27) cometsce comtetmos- comtetron untitste © untitronuntiteE, 4 ‘ i (25) t t { { { £ (7) @ (15) é é (26) vé yé (98) comiste comimos comyéron uniste unyéron unde89 We must now account for the various vowels and the final stress in all the first and third person singular forms. In coayS and unyé tt is obvious that the y comes from i, but it is not so obvious what the 9 cones from. We need some vowel Y* such that comHtv* ard unti#V* can be stress- ed comtfiv* and untitv* by (17), such that the stress will be shifted to the right, and such that V* may be converted too. We must also prevent (2) from deleting the theme vowels before they can be turned into y. Re- viewing the cases where (2) must apply, we notice that in the present sub- Junctive of the first conjugation, (2) applies in che environment __+{V, -D]. That is, (2) applies to amtatet..., not tamat#t..., giving amtess, . Let us assume then that (2) must be replaced by (28). (28) v= O/ 4+ fe Now, siree (28) does not apply to combi#v* and untiti*, che vowel V* nust be some vowel with the feature (+D]. At this point we may observe chat with the unique exception muy, there #te no instances of *uy or *iv at all, althouga the sequences wi and yu are common. Now, since we have independent motivation for rule (59) which curns unstressed high vowels into glides immediately preceding and imnedi- ately following a vowel, we may give the following extremely general rule (whose correct formulation need not coresrn us now): wr fade] fata a L brseres: > (vstress J [+seress] 1 2 1 290 Given (29) and given that the stress in comH{4v* ard unt{+* (assigned by (17)) shits to comisf* and untit¥*, then Y* must be some high vowel. We have established now that Y* is some vowel with the features (high, 4D]. Since Y* is ultimately realized as 9, the simplest assump- tion is that Ve is also [tback], that is J. This assumption is supported by first conjugation first and third persons singular amé and amd: if comy and unyd are phonologically comet and untit, then and aust be amiatU, and there must be rules with the effect of (30): (30) a> 0 Before refining (30), let us examine the first person singular forms ané, comf, and un{. The apparently final stress aad the ¢ of amé will be accounted for if we find some vowel V' such that tae underlying forms in question are amtat¥', comte+¥', and untitY', chat aV' a> e, and that Yt will be deleted in the second and third conjugation forms. Since the theme vowels are apparently not deleted by (28), Y' must have the feature (+D]. Now i€ VU! is I, chac is, (1, 4D], we can expand (30) to (31): au o} (31) > { { al? fel de can capture the symmetry of (31), and achieve part of the desired effece with (32): (32) (Ned = flow] / consonantal] WH \ewoack} '— frhigh | Gander certain conditions) lpback J91 Since there are in fact aw and ay sequences in Spanish, (32) must be restricted in some way. This will be discussed below in Section 9.3. In any event, (32) will produce sequences which will ultimately be re- alized as ow and ey. But such sequences do not occur phonetically out- side a few exceptions. In fact, there are no sequences consisting of « nonlow vowel followed by an unstressed high vowel (or glide) which agrees with the first vowel in backness: ‘*1y, *ey, *uw, *ow. Therefore we may give the completely general rule (33): -consonanta (33) J-stress v | thigh =0/ |-low _ laback ack | Thus the derivations of amé, amd, comf, and unf are as shown in (34): (34) amtatI amta+U combetT unti+l £ (25) - o - s (28) fails a a { t (7) é é (32) 0 9 9 @ (33) ang amd conf unt We still have to account for the final 9 of comyd and unys. With che rules proposed so far, these forms vill be left incorrectly as comyd and uny. Since the final vowel of these forns has been shown to be /U/, we obviously need a rule with che effect of (35). (35) uso92 Since we already have rule (6), which lowers high vowels under certain conditions, we may collapse (6) and (35) into (36): (strese] ey @) (36) vi~ (high / (B] | ) Rule (36), particularly case (b), will be discussed further in Section 9.3. The complete derivation of the third person singular preterit forms is given in (37): @7) amtatU comet «unit i (2s) a : : (28) fails a £ { a7) é (32) 0 (33) a6 av (29) é & — (36b) y y (59g) aaé somyd says Our survey of the preterit forms is now complete. For the convenience of the reader, we now summarize the derivations of these forms, by conjuga- tion. First Cor jugation amratl amtatste «— amtatU «= amtatmos © amtstron 4 4 a i % a7) é 6 (32) ® » (33) ang andste and amimos anionSecond Conjugation 3 come+l comfetste —comfetU—cometnos. © comtetron ‘ ‘ ‘ ‘ L @s) £ ‘ fs £ 7) $ (33) é (26) yé (598) it 23) é (36) y (598) comf comiste comys comfmos comyéron Third Conjugation untitT untitste unit = untitmos © umtitron { t £ £ f a7) 0 (33) g (26) yé (59b) ig (29) 4 (36b) y (398)of Summary of Rules ie now collect and give, in the proper order (to the extent that this can be determined by the daca presented up to this point), as (38) all the rules discussed so far in thie chapter. (There is sone duplication of rules discussed in Chapter 2.) It must be stressed that several of the rules in (38) are not yet in final form. a) o vo) L frataty a. (10, 19 25) Cree -D | __Urpact] so fp onsonenta ack v e. (2, 28) vos ++ | @, 17) V=(Iseress} /__((-perfective JC VCgily 2. (20) a Vv), flow Feonsonanta I (32) lato * fback} /__frhigh (under certain laback conditions) f-consonanta . sacreea | wid (33) thigh b= of elo | [geack [ebeck| a a) [-sonorane ] = (-tense ] /V___f+vocalic] (under cercain conditions)95 1 fe Freoronal 4. as) Gol -6 wissen + v . fea ‘ado} js (2, 23, 26) -stres: +D J_+Aw é Kk. (1T-59b) {il - (under certain conditions) o had peo v v 1. (29) high inn > (-stress] [tstress] Eseress] | L 2 1 2 [-consonanta, i a. (21-592) [thigh = Gvocalic] / rstress v. Tense and Lax Vowels We have postulated the ad hoc feature (D] which distinguishes octher- tise identical vowels. Anyone familiar with the history of Spanish will aave recognized that several of the rules presented here are similar to aistorical rules, and that, given chis similarity, [oD] corresponds to {-atense]. For example, the vowels which diphthongize axe reflexes of Latin lax vowels, and it is in che reflex of lax ¢ which is deleted finally after 4 single dental consonant, ete. I have not taken the step, hovever, of identifying D] as (-tense] and (-D) as (+tense], because I do not believe96 chat this step can be justified on the basis of che data alone, although it may be justified by extrapolating from the data to general theoretical considerations. the only careful discussion known to me of the distribution of tense and lax vowels in phonetic representations is that of Navarro Tomés (1965, uw p. 35££) for Castilian, Navarro's description, hovever, does not carry over to any Latin American dialect I have ever heard. In particul King (1952), whose data is essentially identical to mine, states, in effect, that phonetically tense and lax vowels are in free variation, and he takes some pains to press the point (perhaps because his statements are at vari- ance with Navarro's vell known and highly respected work). King has over- simplified slightly, but in any event it seems to ne that a detailed study of phonetically tense and lax vowels in Mexican Spanish would be most unre- warding. This is apparently the position taken in Stockwell, Bowen, and Silva-Fuenzalida's (1956) extremely detailed study of another dialect, where discussion of tense and lax vowels is limited to the single statement chat “we have been unable to classify the distribution of lax (BJ and tense (+] and similar data well enough to include them here, except to assert chat they exist..." (p. 408). It is clear that there is absolutely no correlation between tense and lax vowels in systematic phonemic represen- tions and tense and lax vowels in phonetic representations. Therefore, without general theoretical constraints, the choice of 2 feature in terms of which distinctions are made which are necessary to capture significant generalizations is wholly arbitrary. We have mentioned at several points that the familiar Latin stress rule seems to play a role in predicting stress in Spanish (cf. Chapter 2,7 Section 3.11.3). It is perhaps from this rule chat the most cogent theoreti- cal argument can be adduce for the identification of [aD]as [-atense], since uyowel quaneity" (and the derivative notions "strong" and "weak" syllable) figures crucially in this rule. For the sake of argument, let us accept the ccrrectness of the Latin rule for assigning stress co categories other chan verbs.!) Recall now that rule (38d), the only rule we have proposed so far to assign stress, is quite unlike the Latin stress rule. In particular, (38d) assigns stress only to verbs, and it assigns stress in fixed positions without taking into account vowel quancity./° 1 will present evidence now that this is the correct vay to assign stress to verb forms (although (38d) itself may need to be altered). Consider the paired examples in (39), which show antepenultimate stress in nouns and adjectives but penultimate stress in (synonymous) verbs? G9) Noun, Adjective Yerb Noun, Adjective Verb continuo contindo préctica practica ndufrago naufrdgo crdmite tramfte eréfago trafégo erdnsito trans {to Animo anino catdlogo cataldgo fdbrica fabrica platica platica vélido valfdo triéngulo triangdlo ifgrima Lagr ima cdlculo calcdlo idscima last ina codgulo coagélo p4gina pagfna intérprete interpréte efeculo ciredlo donéstico donest feo sxédito (a)eredsto (in)édito edfto fategra intégra réplica replica legftima legitina oritica critifea solfeito colicfee 1fquido Liqufdo ‘ecfproco recipréco pertfeipe parcicfpe setfaulo estimilo equfvoco equivéco vinculo vinerlo sincope sincdpe préspero prospéro eitulo (en)titdlo orSximo (a)proxfno Svalo ovdlo Sxido oxide arédigo prodfgo98 (39) continued vbnito vonfto dspésito deposfto Férmule forméla ox{zeno oxigéno edaputo comptito eémodo (a)eonddo public publ feo rétulo rotdlo jbilo jjubflo nimere numéro siplica suplfca cfreel (enjearcélo the List in (39) can be extended considerably. It is abundanely clear that in order for the Latin rule to assign stress to the antepenultimate youel in a noun or adjective, the penultimate vowel must be lax, but this sane lax antepenultimate vowel is stressed in the related verb; therefore either (a) verbs cannot be stressed by the same rule as souns and adjec- cives, or (b) the stressed vowel in verbs is antepemultinate at the time stress is assigned. I know of no reason to suppose that (b) is correct. Let us assume chat (a) the identification of Cod] as C-atense] is theore- tically justifiable, (b) stress is assigned to nouns and adjectives by che Latin stress rule (ignoring the problems mentioned in note 15), and (ce) stress is assigned to verbs by rule (38d). Then che first case of (40) is just the Latin stress rule, and the second case is (38d): { easiegen ota#Ian @ (40) vo (stress) /__ ((CCepertectivel Cpu Gg#ly | () There are a number of unclarities in (40), but the following observations will suffice for the moment: first, the outermost set of parentheses in both cases is co allow stress to be assigned to monosyllabic nouns and verbs like eé and va; second, the parenthesized L in the first case is in- sended to express the fact that 2 consonant plus liquid cluster is “weak,” as is shown by {ategro, ailtiple, nd many other clear examples. 1 will allow other unclarities to remain.9 One might entertaia the idea of attempting to achieve somewhat greater generality in the assignment of stress, which would be reflected in a simpli- Eication of (40), along the following lines: let us assume that the vowel and ra is in fact lax Z. Then we might try in the [ -perfective] markers to collapse the ewo cases of (40) by somehow letting the ¥ in case (2) and the imperfective % fall together, eliminating case (b). This will not work, however: consider pairs Like ndufriigo (noun) vs. aaufrigo (verb) and créfigo (noun) vs. trafago (verb). Thus, although (40) clearly can be cleaned up a bit, I see no alternative to assigning stress co fixed positions in verbs without regard for vowel quantity. We have hypothesized that verbs are alvays stressed penultimately, except for the first person plural of the imperfect indicative and subjunc- tive. Henceforth, when "penultimate verb stress" is mentioned, the excep- tions just aade will be assumed to be understood. Now, there are clearly no verb forns with unstressed penultimate 2, ¢, 9. There is, hovever, a large set of verbs, at least 150, which might at first glance seem to have unstressed penultimace i. We will take cambiar as representative of this class.” The phonetic representation of the present indicative and subjunc- tive and nonfinite forms of cambiar are given in (41): a) Indicative biunc tive séubyo——kambyénos sefubye cambyémos xénbyas ieéabyes sémbya xéabyan kfabye edmbyen Inginitive st Participle © Present Participle kambydr kamby&6o iambydndo100 If e.g. kdmbyo, kdmbyas, were /kambio, kambias/ before stress is assigned, they would be incorrectly stressed as *eanbfo, *kanbfas, etc. The simplest assumption is that there is no penu: imate iat the time stress is assigned: phonetic y is also underlying y, the root being kamby- The “cambiar class" of verbs may be contrasted with another large class. We will take ampliar as illustrative of this second class.18 ‘The forms of ampliar relevant to the present discussion are given in (42). 42) Indicative Subjunctive amplfo _amplydnos anplfe anplyénos ampl fas amplifes anplia anpifan amplfe anpifen Past Participle Present Participle amply4So anplyéndo Thus the final seguent of stems of verbs of the anpliar class must be a vowel, not y. The following sample derivations illustrate the crucial dif- ference between the cambiar class and the ampliar class: (43) kambytatmos amplitatmos —kambyta. = amplita 4 ‘ é £ (40R) y (38n) kambydmos amplyamos kdmbya — amplfa Further supporting evidence for this treatment is provided by che following: verbs of the ampliar class may have related nouns or adjectives ith (a) penultimate stress; rociar, rocfo (verb) and rocfo (noun} vaciar,101 vacto (verb) and yacfo (noun or adjective). Thus the stem has final tense 4, and (b) antepenuleinate stress; ampliar, amplfo (verb) but dmplyo (adjective); variar, varfo (verb) but véryo (adjective), Thus the stem has final lax L. Most of the verbs in the cambiar class also have clearly related nouns or adjectives, e.g. cambio (kdmbyo ](noun). In contrast to the anpliar class, there are no antepenultinately stressed nouns or adjectives related co verbs of the cambiar class, e-g- *cambio. Complete lack of counterexamples in the 150 or so examples I have collected, which is quite a large sample, is rather strong support for the hypothesis that these verbs have stem final y. ‘There is still another (marginal) contrast to be accounted for. Consider the (so far as I know unique) verb piar, in which the £ is always syllabic, never y. Thus we have such contrasts as bisyllabic piar (pidr] versus monoyllabic guiar (gyér] (infinitives), trisyllabic pisdo [piso] versus bisyllabic guiado (yé8o] (past participles), and even the completely ninimal pair of tricyllabic piara (pidra] (imperfect subjunctive of piar) versus bisyllabic piara (pyéra] ("herd of animals, especially pigs"). Now ig verbs Like cambiar have stem final y, verbs like yaciar have stem final tense i, and verbs Like ampliar have stem final lax I, it is not immediately obvious what to do about che stem vowel of piar. There is also « synonymous noun pfo. I therefore propose an internal noun cycle for piar, roughly as follows: ferms is scraightforwara except for the question of the underlying117 representation of phonetic Iq short, stems of both the pedir and servir clesses of verbs may save eather high or mid front underlying vowels. In the absence of any alterna cions which indicate the contrary, we assume chat the vowel in quescion is 4, since this is the simplest assumption. 10.2.2.3 Members of the “herix class" are listed in (65): 465) ace ee .| - ee ee a faerie to ine oor fgute requeriz eae = | {illustrate wich near: 466) Past Parciciple Zeesent Participle nerir aerfdo nixidndo Presenc indicative nar {nos aygra agednos118 Inperfect hirigramos nerfa hervan niriéra — biridran Preterit nerf ner fnos herfsce hirié nirigron Thus the distribution of ve, 2, and i exactly parallels chat of ue, 9, and 1g illuserated in (54). That is, only ye occurs under stress; both £ and ¢ occur unstressed, but only ¢ occurs in the environment |. Accord= ingly, (55) is replaced by (67)+ «67) v- in sone environment aula (67), Like (55), is ordered before (381), diphthongization. We will jot be able to specify the environment of (67) until we have completed the yurvey of third conjugation verb subclasses. For clarity I give sample terivations of sone forms of herir in (63), where i scands for some non= Low front vowel68) hiett hirtitmos £ £ 4 e ayére ‘serimos 119 hirtitetaos i re) o (38) 4 (384) (67) (38k) (8m) (62) hirdmos 10.2.2.4 Members of the “sentir class" are listed in (67): 3 J re> sentir agrvir Jerbs of the gencir class 3 vertir contro- per- acrepentir ve exactly che same vowel alternations as chose of the nerir class. I have separated the two classes on the basis of the fact that che stems in the nerir class end in a single consonant, name iy x, while those of the sentix class end in @ consonant cluster120 With the exception of a few idiosyncratic (defective, ete.) verbs to be mentioned below, this completes the survey of subclasses of third conjugation verbs with front stem vowels~ 10,3. We have seen chat just three rules, (59), (67), and (62) =~ im this order -- must be added to an independently motivated granmar in order to account for all the voel alternations in the stems of third conjugation verbs, These three rules aust now be refined, and made to work properly. We postpone further discussion of (59) until after the treatment of the constituent structure of verbs in Section 11. Let vs now Look aore closely at (67). This rule mist apply to just che following verbs (and the classes they represent): mori, dormir, hari, sentiz. This means char (67) applies to stressed vowels in the following environments: 2 (4 consonantal] _ ) ancecior | gue (67) does not apply co all stressed vowels in the environments of (70) ind counterinstances in forms of the folloving verbs: none Frconsonancaial But consider the following: all the verbs waich undergo (67) also undergo (38k), diphthongization. We know that (38k) is a [+] rule. Let us assume then that (67) is also (+Q], and, of course, that all the verbs affected by (67) are [+]. Now, there is no reason to assign the featuce GQ] to any of the verbs in (71), so we may replace (67) with (72) : 2 | We now noce that (72) is strikingly similar to (38j), which, for the con- venience of the reader, ve repeat (wich trivial a v Ba ado] (73): stress] fst] hoot tens Ww Since both (72) and (73) must be ordered before (38k), let us investigate onal changes) as (73): the consequences of collapsing tie two former rules as (74): ) fv J rv [rsecesd = FRish] oy | J eae fd [esesmeoe] tsa \alow 7 \gBconj} +__#] phigh v db. (383, 53, v . pause [Sot#y 59, 104) clow tense] /__ [C+pase]| e. (388) {" . {| _ feeaceno] [etasel ree(tfuture] xanor 4. (38¢, 95) V+ Ott1 142 Cees nrvrege (384, 40) V+ Clstress]/__ (CEeperfective NOV ICoFly (15-592) > w/#__ (38e) b= 8 it v Low ~consorantal] (38£) tow = aback] /__ (ist 4@ aback consonantal] j-strese v (38g) thigh 97 |-tow | __ . aback | \oback, (38h) onorand] LQ * (-tense] / V__GHocalic] G8) [ested = a/v [feneerio] 5__¢ (384, 55, v w 67, 72, 73, stress| _ (“high 1) Ktense] /__ Freononance] (feensonsaee] lsonoranc \tanterior fe) ye ) Fstress| (38k) | 7 [rtense \ | +Q l we v v (381) | +high | [pee] > (-stress] [+stress] rstress| 1 2 1 2143 stense| 49 o. (38m, 56, 62) V~ C~high]/ # sail {41 | aoe eons onantal, p+ (96) G+ 4/ |+sonorant + -continuant qe (II-59¢) r>R onsonanta. re (11-594) voca Lic “._ fisonorant seg} lateral tense 7 (eee | +tense coronal s. (11-592) r-(-vocalic} / |-distributed| __ cons onantal| te (ET-59£) igh Cevocalic] / stress vw ue (45) Erase nonrightmost stresses in a word Rules (105y, ¥, X, ¥» 2, a', B', ¢') are (11~59g, b, respectively, with no change.ha FOOTNOTES TO CHAPTER IIT Second person plural (ycsotros) forms occur only marginally in the @ialect under investigation: educated speakers learn them in school, encounter them in literature, and occasionally use them - often in- correctly - in religious and other types of stylized discourse. Thus these forms have in Mexican Spanish roughly the same status as that of "thou" and "ye" forms in Standard American English. I see no reason €o include them in this study, although to do so would entail no major revision of the rules to be presented, as far as I can see. GE. Bull (1949, 1960), Hall (1945), Hockett (1947), Nida (1948), Saporta (1956b, 195: By Stockwell, Bowen, and Martin (1965), Foley (undated, 1965). I assume, without justification here, that these phonological segnents will be supplied by rules which can be stated roughly as (person, -plural]= 9, [2person, -plural] + s, (lperson, +piural] + mos, (3person, +plural] + n. There is no rule to give phonological reali- zation to the feature bundle (3person, -plural]. It seems to oe chat any theory is doomed to failure which countenances a person morphene and @ separate number morpheme. Not only is this "string-of-morphemes" approach contraindicated on syntactic grounds, but also an elaborate reductio ad absurdup counterargument on phonological grounds has been provided by Foley (undated).145 ‘The exceptions include the clitics su, tu, mi, words of Greek origin Like énfasis, dosis, the three Latin words espiritu, tribu, {apetu, affective words like mami, papi, and a few miscellaneous others like casi (which probably shouid be treated as a clitic). Rule (9) has only illustrative value for the present discu Ie will be revised in Chapter 4, Section 6. Further discussion of this rule will be found in Section 9.3 and Chapter 4. For further discussion see Section 11. The identification of (D] is discussed in Section 9. As is to be expected, there are some exceptions and complications which will not be dealt with now. ‘The familiar Latin stress rule has already been alluded to on page 37, and will be further discussed in Section 9. In the present connection it is interesting to note that the imperfect (penultimate) was long (tense) in Latin amabiimus, and the stress continued to be amabémos in Spanish until it shifted to amMbemos for no discernible reason (cf. Menéndez Pidal (1962, p. 276). 1 can see no reason to be- Lieve that this stress shift is to be accounted for by an unexplained laxing of the penultimate a rather than by an equally unexplained change in the stress rule itself. The uistory of the stress in amframos is parallel to that of amébamos (cf, note 10).12. 13, 146 This is hardly surprising in view of the history of the preterit forms. They are the reflexes of the forms of the Latin perfect, whose paradigm was already quite different from the other Latin tenses. Further, the historical development of these forms con- tain many points which are not understood by philologists. (cf. Mendndez Pidal (1962, p. 308££) Garcfa de Diego (1961, p. 229£)). Both the traditional terminology "past participle" and rule (25) suggest that the feature bundle which represents the ending of the past participle contain the feature (+past]. I am convinced that this suggestion can be corroborated by syntactic considerations, which, hovever, fall far outside the domain of this study. If the suggested analysis of past participles is correct, then the envir- onment of (25) is simply __[tpase]. Navarro explicitly points out, and gives examples, that "las diferencias de timbre que hoy se advierten en la pronunciacién de cada una de las vocales espaffolas, no tienen valor significative ni obedecen a motivos de cardeter histérico 0 etimolégico, sino simple- sente a circunstancias fonéticas" (p. 41). There are, however, a number of difficulties. For example, consider the 9 - we alternation in Venezuéla, gléno, where the apparently “lax" 9 diphthongizes when stressed. But it is not clear how the Latin rule can sign stress to the penultinate lax vowel followed by only one consonant in venezQla. The same difficulty appears in many other nouns, e.g. abyglo, agiiéro (cf. agordr), tropiézo, (cf. tropezdr), ete.147 16. The situation described here arose historically in two steps. The first step took place sone tine after the 13th century. Menéndez Pidal (1962) states that "los verbos cultos dislocaron el acento latino para hacer llenas (penultimately stressed] las formas latinas esdrdjulas [antepenultinately stressed]: asf [Spanish] recupéro, coléco, vigila y otros muchos; compérense las formas espaiiolas de {Latin} sipplico, imigino, detérmino, pébito, drrogo, aggtego, élevo, {ntimo, fruct{-, ampl{-, not{-fico [which are suplico, imag{no, determino, etc. in Spanish]. El cambio de acento latino no lo cfan an los cultismos del siglo XIII; Berceo pronunciaba significa, sacr{fica. El italiano conserva siempre la acentuacién cldsica: sacrifico, vivifica, célloca, stérmino, cons{dero, ete." (p. 274). The second step was the retraction of stress im the imperfect indica- tive and subjunctive, which was mentioned above in note 10. At that time, as will be recalled, e.g. amabémos and amardgos became andbanos and amdxamos, thus destroying the last vestige of relationship between verb stress and etymological vowel quantity. These are the facts, not an explanation of the facts. A true explanation for these stress shifts would be, it seems to me, an accomplisiment of the highest order of interest in historical lirguistics, but I have nothing to contribute. 17. A few other examples: Limpiar, aliviar, anunciar, remediar, rabiar, rincipiar, premier, odiar, obsequiar, negociar, ensuciar, entibiar, envidiar, estudiar, fastidiar, incendiar, injuriar, lidjar, asediar, calumniar, codiciar, columpiar, contagiar, desperdiciar, diferenciar, divorelar, elogiar, peneficiar.18. 19. 20. 21. 22. 148 There are about 30 or 40 verbs in this class, including vigiar, rociar, resfriar, pipiar, Liar, expiar, estriar, enriar, enlejiar, ciar, ataviar, descerriar, adiar, cuantiar, This class includes actuar, exceptuar, habituar, insinuar, situar, nticusr, desvirtuar, evacuar, fluctuar, greduar, individuar, genstruar, oblicuar, perpetuar, puntuar, usufructuar. See Chomsky and Halle (1968, Chapter 4, Section 2.2 and Chapter 8, Section 7) for further discussion of diacritic features, like [@]. which are @ property of entire formatives rather than of single segments. Although the point is often misunderstood, I take it to be self evident that Language learners do not have access to historical data as such. Therefore, it is not clear what claims can be made for grammars which make use of data which is not available to human beings who learn the language. It is entirely another matter when grammars based entirely on synchronic data happen to reflect certain historical Processes. There are a few oddities which cannot be used as data since neither I nor my informants known how to conjugate them. For example, abolir, vhich 1s defective. Native speakers diagree with prescriptive authori- ties and with one another atout which forms are in use and about the shape of the forms they use. Also, for some speakers, pVdrir has u im all finite forms, podride and pudrido in free variation for the past participle, podrir and pudrir in free variation for the infinitive.23. 2. 26. It is an interesting historical fact that at an earlier stage of the language, many of the verbs in (52) had phonetic o rather than u. Menéndez Pidal (1962, p. 273) mentions older ordir, complir, cobrir, sofrir, somir, and adocir for modern urdir, cumplir, cubrir, sufrir, sumir, and adueir. In the light of these examples it seems that the forms of pWdrir with o are simply archaicisms. ‘The following argument against (74) has sone initial plausibility, but is incorrect: it is fairly clear that at least some instances of # are reflexes of an in[+Q] formatives (cf. /ana/ in (+0) afio, "year," and GQ] anual, “anual"). Now certain verbs contain » seffir), yet they do not undergo (74), as they should if they are (+Q] and have (grufir, brufiir, muir, -seredir, teflir, heir, refi underlying nn. This argument fails because there is no reason what~ soever to suppose that all instances of & derive from nn, or even that the verbs in question contain un at any stage of derivation. See Lakof£ (1965) for a discussion of minor rules. Note further that some of the verbs that do not undergo (62) had, at an earlier stage of the language, phonetic ¢ in the stem, e.g. older escrebir and vevir for modern escribir and vivir (cf. note 23). Haber never means “have in the sense of "possess, own"; this is ener. Spanish is thus excluded from the large number of languages in which "have and the perfect auxiliary are the same (cf. Allen (1g64)).27. 28. 29. 30. 32. 150 For discussion see Chomsky and Halle (1968, Chapter 1, Section 5.1). The entire discussion of the future which follows carries over with only trivial modifications to the verb forms traditionally called "conditional," which will not be discussed separately. This is not quite true for the dialects that use the second person plural. This form of haber is habéis, but the corresponding form of the future is e.g. amaréis, not *tamarabéis It is of course vell known that the traditional explanation is cor- rect historically: until the 17th century the future was expressed by the infinitive followed by the present of habez. These were written as two separate words, and cbject pronouns could come between them (as is still the case in modern Portuguese). For example, 17th century dar me le has for modern me lo dards, "you will give it to me." But these historical facts do not conditute synchronic evidence. The purpose of this work was pedagogical, ané thus it must be judged on the basis of its pedagogical value, which, in ay opinion, is out- standing. Still, the theoretical apparatus within hich the pedago- sical exposition is framed is quite clear, and its correctness can also be judged. The range of relevant data is wider than what has been presented here. For example, the rules which govern the placement of clitic object pro- nouns in surface structures are relevant to the correctness of struc-