Printer Friendly

A fresh look at quantity sensitivity in Spanish *.


The present article aims to examine, with the help of a made-up-word test, the psycholinguistic reality behind the generalizations assumed to govern the stress system of Spanish nonverbs, paying special attention to quantity sensitivity. I will argue that the apparent quantity sensitivity of Spanish nominals is a historical heritage, not an active constraint, and that the stress system of Spanish does not show rule-based behavior; stress is rather the lexical property of words, and the system works on the basis of analogical extensions.

1. Introduction

The debate about the quantity sensitivity of Spanish stress goes back as far as the eighteenth century, when Larramendi pointed out that such words as limiste (1) `shawl' and Fromista <proper name> exist, so it is evident that in Romance "the pronunciation in no way depends on syllable quantity" (Larramendi 1729: 348). (2) Roca (1988, 1997) claims that stress assignment in Spanish is rule-governed, and he also argues against the quantity sensitivity of the Spanish stress system. The argument he brings is based on loanwords like Robinson. Roca represents diphthongs as underlyingly heterosyllabic vowel sequences, which explains the lack of words like *astrona[w]ta since the "three-syllable window condition" requires that stress in Spanish must fall on one of the last three syllables.

Other linguists, on the other hand (e.g. Harris 1983, 1992; Dunlap 1991), while also claiming that stress assignment in Spanish is rule-governed, say that the system is sensitive to syllable structure, more precisely to syllable weight. This means that if the penultimate syllable is heavy--that is, it is closed by a coda consonant or contains a falling diphthong--the word cannot bear stress on the antepenult.

In more recent works--like Aske (1990) and Eddington (2000)--the stress system of Spanish is not viewed as a rule-based system. These authors claim that stress is the lexical property of words and the system works on the basis of analogical associations. I will refer to this approach as "patterns in the lexicon" (following Aske 1990). Neither Aske (1990) nor Eddington (2000), however, studies the active quantity sensitivity of the Spanish stress system.

The rest of this article is organized in the following way: in section 2 the presentation of the basic differences between rule-based and analogical models follows. The detailed description of the made-up-word test is given in section 3 and the analysis comes in section 4; the conclusion ends the study. The present article does not deal with words containing diphthongs; (3) only words containing closed syllables will be discussed.

2. Rules and/or patterns in the lexicon

2.1. Rules

There has been a long discussion among scientists (especially psychologists) concerning whether our cognitive functions, including language, are rule-governed or work on the basis of analogical extensions, that is, associations. The question is whether the constructs developed by linguists have cognitive status, and if they do, what form this knowledge has in speakers' minds. A well-studied test case is the processing of regular and irregular morphology. (An interesting account is given by Jaeger et al. [1996], for instance.)

By "rule-governed system" a symbol-manipulating mechanism is meant in which a rule operates whenever its structural description is met. In linguistics, representatives of the classical generative approach, such as Chomsky and Halle (1968), see language as a system of symbols and rules. It does not matter how restricted the application of a rule is, it must take effect whenever its conditions are met. (Let us abstract away from rule ordering now.)

In generative grammar, rules have a central role: they account for all regularities. (Minor rules account for "irregularities." The fewer minor rules are needed for the derivation of a form, the more regular the form is thought to be.) Speakers are supposed to extract the same generalizations from the data available as linguists do. Therefore stems of the regular type are stored in the form of lexical representations--idiosyncratic underlying forms--and rules that apply to lexical representations (or to the outputs of other rules) generate the surface forms. There is a great variety in the extent of abstractness allowed in underlying forms.

In rule-based approaches, similarly to any phonological process, words receive their stress patterns via rules, in this case a stress-assigning algorithm. Rules can be formulated in such a way that they explain ALL of the data correctly. In many cases special lexical marking is needed too. In order to achieve this, rules refer to abstract formal properties like extrametricality and other diacritics. In Harris (1992) a word-final vowel and an inflectional consonant are extrametrical. "The rightmost (metrical) stressable element is accented" (Harris 1992: 4), unless the word is a lexically marked exception to this rule, like for instance sabana `sheet' and habil `able'. Harris considers words with a stressed word-final vowel (e.g. cafe `coffee') marginal. In Dunlap (1991) the algorithm builds a moraic trochee on the right edge of the word: ba(rato) `cheap', fu(sil) `rifle'. In Dunlap's account proparoxytonic vowel-final words and paroxytonic consonant-final words have their final mora lexically marked for extrametricality: (saba) <na>, (mobi) <l> `mobile'. Nonverbs with a final stressed vowel like cafe `coffee' are analyzed as having an unspecified consonant at the end of the root, which is still part of the derivational stem: ca(feC). Even in these approaches, which assume that it is not costly if the surface form is derived by a large number of rules but it is disfavored to have a lot of items to remember, a lot of lexical markings--the psychological reality of which is questionable--are needed besides the stress rules to account for the right forms. The economy of lexical storage, however, is not thought of in the same way now as in earlier generative theories (see the description of the "patterns-in-the-lexicon" approach in section 2.2 below).

In the traditional view, "regular" forms generated by rules should outnumber exceptions. In principle, however, we should not be surprised to find languages where rules only apply to a minority of words, while the majority are irregular (lexicalized) and have to be learned. What we mean by the term "rule" need not be statistically general, but it must apply when invoked--and may be invoked only rarely. Indeed, Marcus et al. (1995) report that German is such a language. They argue that the notion of "default" is an operation that applies not to particular sets of stored items or their frequent patterns, but to any item if it does not have a precomputed output listed. The occurrence of default cases might have various reasons: the item is unfamiliar and dissimilar to familiar items--and therefore not generalizable from familiar items. The results Marcus and his fellow researchers achieved is presented briefly in section 2.3.

2.2. Patterns in the lexicon

The "patterns-in-the-lexicon" approach assumes that surface phonological forms are governed by patterns in the mental lexicon and by analogies to these patterns. According to this view, there is no need for abstract underlying representations as assumed in generative phonology, nor for derivations and ordered rules or ranked constraints that are supposed to represent speakers' phonological competence. All information is stored in the lexicon, which is highly structured, and speakers have extensive knowledge of the patterns in their mental lexicon. Therefore they can make very specific generalizations. More recent (post-early generative) research shows that speakers only extract patterns that are relatively transparent and productive, and the human mind has a large capacity for memorizing anything that is not absolutely predictable. (4) It seems that it is more efficient for our mind to retrieve an item from the mental lexicon than to perform phonological and morphological operations on new strings. So while earlier accounts assumed that storage of lexical items and computation are organized in subsequent steps, a plausible possibility is to assume that storage contains fully formed structures. And these structures--as Burzio (1996) explains--are only "checked" by the calculation of whether the form is correct.

In this model, computation is not required every time a lexical item is used, only when an item is STORED, and retrieval is "cheap." The retrieval of one lexical item facilitates the retrieval of other lexical items similar to it in some sense--as argued by connectionist researchers. (5) In the analogical model the phonetic form of utterances is obtained either by direct retrieval from the lexicon or by analogy with already existing forms. If analogy is used in the production of a form, a number of factors might count, such as the phonological shape of the word, or nonphonological information (e.g. word category, semantic properties, etc.).

One of the first analyses in this theoretical framework dealing with Spanish stress is Aske (1990). Aske argues that Spanish stress is not rule-governed but rather corresponds to patterns in the lexicon. The view is supported by two psycholinguistic experiments he carried out with the participation of 36 native speakers. In the first experiment he presented informants with a list of sentences, each containing a made-up word ending in -n. Six test words ended in -en and six control words in -{a, i, o, u}n. The native speakers read the sentences aloud. He wanted to find out whether informants would stress them on the last syllable or on the penult.

The results of the test support the prediction of the "patterns-in-the-lexicon" model. Speakers stressed -en words on the penultimate syllable in 43.5% of cases, while words in -Vn (where V [not equal to] e) were given stress on the penultimate syllable only in 3.2% of the cases. The difference is statistically significant ([chi square] = 8.289; p > 0.01). This means that speakers treat words in -en differently from words in -an, -in, -on, and -un, which reflects the actual patterns in the lexicon. According to the traditional rule-based system all consonant-final words should be stressed on the last syllable. Of course, there could be a rule stating that -en-final nonverbs are marked for stress retraction. In this case, however, made-up words with this segmental composition should behave uniformly.

Eddington (2000) prepared a study to test Aske's findings; the analysis is set forth within Skousen's analogical modeling of language (AML). AML is a model that attempts to reflect how speakers determine different kinds of linguistic behavior, such as stress assignment, on the basis of memory tokens. Eddington's model also assumes that when speakers are presented with an unfamiliar word, they access their mental lexicon and search for words that are similar to the word in question. They then apply the stress of the word(s) found to the unfamiliar word.

In AML the information is contained in a database of items representing the contents of the mental lexicon, and the database may be added to at any time. All affixed and unaffixed words are stored as wholes in this database. Apart from the database of fully specified words a mechanism is needed for searching and comparing words. First, we search for actual examples of the context in question and then move on looking for nearby examples. The probability that a word is chosen as an analog for the given context depends on the following characteristics (Skousen 1995: 217):
(1) a. proximity: the more similar the example is to the given context,
 the greater the chances of that example being selected as the
 analogical model;

 b. gang effect: if the example is surrounded by other examples
 having the same behavior, then the probability of selecting
 these similarly behaving examples is substantially increased;

 c. heterogeneity: an example cannot be selected as the analogical
 model if there are more similar examples, with different
 behavior, closer to the given context.

In this sense, the lexicon is viewed as a module where phonetically and semantically similar items are interconnected. All of the attributes of a given context partially activate all the words in the lexicon that have an attribute in common.

The database Eddington used contained the 4970 most common Spanish words selected on the basis of the Alameda and Cuetos Frequency Dictionary (1995). The variables included in the AML experiment on Spanish stress are the following: the phonemic content and syllable structure of the last three syllables (as no lexical entry bore preantepenultimate stress, it was not necessary to encode more than the final three syllables). Since Spanish stress is contrastive, especially in verbal forms, variables were included that indicated the person and the tense form. These variables also distinguished verbs from nonverbs. Under these conditions 94.4% of the 4970 words tested were correctly stressed by the computer compared to existing Spanish words. If verbs and nonverbs are only allowed to influence members of their own class, the number of errors varies very little. So information available about verbal stress does not destroy the stress assignment of nouns, which means that verbal and nonverbal stress assignment should not be treated as separate processes. (6)

Eddington repeated Aske's nonce word test with -Vn-final made-up nouns in AML. His results in terms of figures were quite different from those of Aske mentioned above. In the AML simulation all -en words are assigned penult stress, and all but one -Vn words were assigned final stress. The author's explanation is that in the AML simulation the highest predicted probability applies, but it is only a limited estimation of a Spanish speaker's mental lexicon. Therefore, AML captures the subjects' preference qualitatively not quantitatively.

Aske carried out the experiment with only 36 informants, which is also a limited sampling. In my view, in the analysis of nonce-word tests--including mine presented in the following section--several questions arise: (i) to what extent do the exact figures have to be considered? (ii) what kind of corpus should the results be compared with? (iii) can the mental lexicon be modeled by the most frequent words? (iv) how big should the database be?, etc. The most frequent words are often the most irregular ones in a language, which might influence the results. How can one take into account ALL the existing Spanish words of an average native speaker? I will not seek answers to these theoretical problems, but I think we should focus our attention on the major tendencies these tests are able to capture.

In the present article I do not want to argue in favor of the superiority of any nonrule model and I am not proposing an algorithm to account for stress assignment in Spanish. I do not have a clear view about whether a connectionist model would describe stress in Spanish better than an exemplar-based one, as in Daelemans et al.'s (1994) account of primary stress in Dutch, or whether Skousen's AML is the most appropriate model to account for stress assignment in Spanish. (For the basic differences between these approaches see Skousen [forthcoming].) I am only claiming (see sections 3 and 4) that the decisions native speakers make about the place of stress in Spanish cannot be accounted for by rules that have a well-defined output. That is, there are tendencies but the boundaries of the sets of words that the "rules" are supposed to account for are not precise; there is leakage across boundaries between different sets of lexical items. In principle, there may be systems that are analogy-based but still contain rules, or a rule in the sense that there is a default that is activated if no analogy can apply (see section 2.3).

Neither Aske's nor Eddington's experiment focuses on quantity sensitivity, so having words like remington <type of rifle> in mind, I wanted to put to the test the importance of the penultimate syllable--whether a closed penult really blocks stress on the antepenult. Although I do agree with Harris (1983, 1992, etc.) that the vast majority of Spanish vocabulary is correctly described by the generalizations he makes, the tests aim to find out whether the assumed quantity sensitivity of the system is an active rule or constraint, or its apparent existence is a historical heritage. (7) I also wanted to find out whether stress in Spanish can be correctly accounted for by a rule-based algorithm (as for instance by the rules proposed by Harris 1992) or is a system based on analogical extensions.

2.3. A system where rules and analogy interact

It has been mentioned above that Marcus et al. (1995) report that in German we can observe the interaction of rules and analogy. The authors argue that the notion of default applies to any item not listed in the lexicon when analogy cannot work.

The hypothesis was tested by a made-up-word test, where native speakers had to form the plural of nonexistent nouns. These nouns were either "rhymes," which means they rhymed with existing German irregular nouns, or "nonrhymes," words that did not rhyme with existing German irregular nouns. This distinction tests the hypothesis that novel roots are likely to receive irregular inflection if they are similar to existing words (due to analogy), while they receive the default (the rule-generated) pattern if they cannot be associated to existing words. The test words are presented in three contexts: root, name, and borrowing. In the "root" context the word is treated as a normal German noun in a neutral context. "Names" have a proper-name reading. It is important to make this distinction because it seems that in the "proper name" context the made-up word is treated as if it did not have a proper lexical input. Therefore, it cannot evoke similar roots from the memory, so the regular suffix is elicited. The third context is that of "borrowings." Borrowings are somewhere half-way between roots and names, since a new word can become assimilated to root status, especially if its phonological pattern corresponds to the phonotactic restrictions of the language. Unassimilated borrowings, on the other hand, are more similar to proper nouns and should take regular inflection, especially if they are not "phonologically familiar" to the informant. That is to say, we expect an increased preference for the default pattern in "nonrhymes" and in the "name" context.

The results Marcus et al. achieved indeed show that plural formation in German can best be accounted for by a dual model in which the "irregular" forms (in this case the majority of German plurals) are generalized by analogy to the already existing roots, while regular forms (the plurals with -s) are created when analogy and memory cannot work--nonrhyme with a proper-noun reading. This also means that surface phonological forms can be the result of a system in which rule and analogy interact.

In the next section I examine whether the stress system of Modern Spanish can adequately be accounted for by rules or rather works on the basis of analogies. We can also imagine that similarly to German plural formation the surface forms might best be accounted for in a dual model, which works on the basis of analogical extensions, and, when analogy cannot work, a default rule is activated.

3. Test

The experiment was a "paper-pencil" production test in which native speakers of Spanish were presented sentences that contained trisyllabic nonce words functioning as nouns. The participants had to indicate where the word would be stressed if it were a Spanish word. Four syllabic patterns were examined, (2). The first pattern (group 1) is CV.CV.CV; it is included in the experiment as a control group and point of reference. All the other groups contain at least one closed syllable. In this way the quantity sensitivity of the system can be tested as well.
(2) Syllabic structure

 1. CV.CV.CV.

 (C = constant; V = vowel)

Words were tested in two contexts. In the first, the test word appeared as a proper name (this corresponds to the "name" context in Marcus et al.'s experiment), while in the second, it had a common "native-like" reading ("root" context). One person received a word only in one context. That is to say, a given test word did not figure in both common-noun and proper-name readings for the same informant--it appeared either in the common-noun context or in the proper-name context in one test. The reason for presenting the test words in two contexts is that various experiments (as mentioned in sections 2.2 and 2.3) examining the production and processing of regular and irregular morphological forms--Jaeger et al. (1996), Marcus et al. (1995), etc.--found that a nonce word that is segmentally similar to an existing irregular word may be treated in morphological processes like this irregular word in the "common-noun" (CN) context, but not in the "proper-name" (PN) context. This means that a nonce word is interpreted by analogy with a similar irregular word in the CN context. On the other hand, the same nonce word, even if it is a homophone of an existing irregular noun, will show rule-based behavior in the PN context (8) (if the system is at least partly rule-based).

In this test, made-up words were not especially designed to rhyme with existing irregular nominals--though in some cases they did. The reason for this is that real words with the same syllable structure as the test words have very diverse distribution, as shown in (3).
(3) Distribution of non-verbs in DRAE

syllable structure antepenultimate

Group 1. [sigma].[C.sub.0]V.[C.sub.0]V 2938
 e.g. ... taca 12.96%
Group 2. [sigma].[C.sub.0]V.[C.sub.0]VC 183
 e.g. ... tacan 3.53%
Group 3. [sigma].[C.sub.0]VC.[C.sub.1]V 1
 e.g. ... tanca 0.02%
Group 4. [sigma].[C.sub.0]VC.CVC 4
 e.g. ... tancan 0.56%

syllable structure penultimate

Group 1. [sigma].[C.sub.0]V.[C.sub.0]V 19418
 e.g. ... taca 85.69%
Group 2. [sigma].[C.sub.0]V.[C.sub.0]VC 791
 e.g. ... tacan 15.26%
Group 3. [sigma].[C.sub.0]VC.[C.sub.1]V 4522
 e.g. ... tanca 99.38%
Group 4. [sigma].[C.sub.0]VC.CVC 159
 e.g. ... tancan 22.40%

syllable structure final stress total

Group 1. [sigma].[C.sub.0]V.[C.sub.0]V 305 22661
 e.g. ... taca 1.35% 100%
Group 2. [sigma].[C.sub.0]V.[C.sub.1]VC 4210 5184
 e.g. ... tacan 81.21% 100%
Group 3. [sigma].[C.sub.0]VC.[C.sub.1]V 27 4550
 e.g. ... tanca 0.59% 100%
Group 4. [sigma].[C.sub.0]VC.CVC 547 710
 e.g. ... tancan 77.04% 100%

The data are computed on the basis of the 21st edition of the Dictionary of the Spanish Academy (DRAE). The list contains 42,093 nominals of at least three syllables (verbs are excluded). Each noun--even if it has more than one entry in the dictionary (homonyms)--is computed only once; unless the same segmental melody occurs with different stress patterns, in which case the differently stressed nouns count as different entries (e.g. sabana `sheet' and sabana `savannah'). In the processing of the data only the structure of the last two syllables were taken into account, as the syllables more to the left do not play a role in stress assignment according to earlier metrical accounts. Group 1 in (3) contains words where the last two syllables are open. Group 2 has words with an open penult and closed final syllable. Groups 3 and 4 are more interesting for us because they involve words with a closed penult, so these words help to determine whether quantity sensitivity is active in the system or not. In group 3, we find words with an open final syllable, in group 4 words with a closed final syllable. It is evident from the table in (3) that the most frequent pattern for vowel-final words is stress on the penult in the case of Spanish nominals, while consonant-final words tend to be oxytonic.

3.1. Hypothesis

We expect that if stress assignment is not strictly rule-based, speakers are likely to associate some nonce words from group 1 in (2) to existing proparoxytonic nominals (stress on the antepenultimate syllable) even if these nonce words are not intentionally designed to rhyme with existing words because of the relatively high number of this type of nominal in the lexicon (12.96%). On the other hand, the dictionary hardly contains group 3 and 4 nominals with stress on the antepenultimate syllable. So proparoxytones with a closed penultimate syllable seem to be practically nonexistent, (9) which might suggest at first sight that quantity sensitivity is still active, so no matter whether the system is rule-based or not, made-up words with these syllabic structures are not expected to be stressed on the antepenult. Note, however, that there are proper names with group 3 and 4 patterns, and Spanish speakers' mental lexicon cannot be completely identified with the RAE dictionary, since the living language has many more (probably hundreds) of English-originating proparoxytones with a closed penultimate syllable.

If the Spanish stress system is divided into a regular and an irregular subgroup--as are the morphological systems of Hungarian and German, for instance--as argued by Lukacs (1999) and Marcus et al. (1995) respectively--nonexistent words in the CN interpretation will be treated differently from the PN interpretation, because in the CN context, existing irregular words (if there are any) may exert analogical influence, while in the PN context they probably will not.

If the results show that the two groups are treated in the same way, we have to reject this type of division in the stress system, but the system can still be either rule-governed or analogical (lexical). We need not be surprised to see that the stress system of Spanish does not show a behavior parallel to the morphophonological system of Hungarian (an agglutinative language with rich morphology), or German, which also has a rich morphological system. If the system is exclusively rule-governed--since rules are productive and systematic--all the words of a given syllable structure should be assigned stress on the same syllable. There should be no exceptions, at least according to the stress rules proposed by the authors mentioned in section 1, because these rules are not sensitive to segmental content but to syllable structure only. If the whole system is lexical (analogical), it would not be surprising if we found that within a certain group (e.g. CV.CV.CV) informants' judgments are not uniform. In this case we would find frequencies in the distribution of stress within a given type that might reflect the actual patterns in the lexicon, and we would find no significant difference between the two readings: PN and CN.

I am not claiming that a rule-based approach does not allow exceptions in a system. In lexical phonology for instance, lexical rules are expected to have exceptions; only postlexical rules are exceptionless. In this case, however, we are talking about nonexistent words that have no proper lexical entry and therefore cannot be lexically marked as exceptions. If there were a default rule--as proposed in earlier accounts: penultimate stress for vowel-final nonverbs and final stress for consonant-final nominals (10)--I assume that speakers would apply it and would not fall back on the lexicon to guess, by analogy to existing forms, to which pattern a given nonce word conforms--especially in the PN interpretation.

3.2. Method

Each type of test word (for instance CV.CV.CV) is represented by four different nonce words, so that the possible influence of the segmental composition of the test word could be reduced. Test words are shown in (4)-(7). The tests were presented in written form, which might cause a certain intellectual reflexiveness, although I am not sure in what ways this could influence the results. I opted for a written test because I considered the amount of data more important--and in this way I could gather more data--than the danger of intellectual reflexiveness.
(4) Test A PN CN
 1. CV.CV.CV.
 2. CV.CV.CVC ca.ta.nes
 3. CV.CVC.CV ti.mis.te
 4. CVC.CVC.CVC Far.tin.ton bes.der.mon

(5) Test B PN CN
 1. CV.CV.CV.
 2. CV.CV.CVC me.ra.sel
 4. CVC.CVC.CVC Cad.bir.tor tron.pes.tar

(6) Test C PN CN
 1. CV.CV.CV.
 2. CV.CV.CVC Ca.ta.nes
 3. CV.CVC.CV Ti.mis.te
 4. CVC.CVC.CVC Bes.der.mon far.tin.ton

(7) Test D PN CN
 1. CV.CV.CV.
 2. CV.CV.CVC Me.ra.sel
 4. CVC.CVC.CVC Tron.pes.tar cad.bir.tor

Note that as Spanish allows only a restricted set of consonants word-finally, only coronals: d, l, n, s, r; and [x] and [j]. ([x] occurs only in one common word relo[x] `watch' and [j] is traditionally analyzed as the glide element of a falling diphthong. (11) I respected these phonotactic restrictions in the CN context, while in the PN context in test B Tedecat ends in -t, which appears word-finally only in relatively recent borrowings like robot `robot', deficit `deficit', in some proper names like Guitart, and in some speakers' pronunciation of carnet `card', chalet `cottage', etc. (These words are more frequently pronounced without the word-final -t, i.e. [karne].) I wanted to emphasize the "proper-name-like" character of Tedecat with the word-final -t. The following instructions were given at the top of each questionnaire:
(8) Instructions (12)
 Each sentence contains a non-existent word printed in bold. After
 reading the sentence carefully, please indicate where this word
 would be stressed (where the "strength of the voice" [la fuerza
 de voz] would be) if it was a Spanish word. (The accent mark shows
 the stress.) For example:

 Esta pacara me gusta mucho. `I like this pacara a lot.'

The same explanation was given orally, and it was emphasised that orthography does not matter.

3.3. Informants

Eighty-four native speakers of Spanish participated in the test. They were secondary school students and university students from Madrid, aged 16-22. Four tests were removed because they were incomplete, so 80 tests could be evaluated. I tried to examine a narrow age group because I assume that young people's language habits may indicate a future norm.

4. Results and analysis

Results are given in (9) and (10). The numbers show how many people put stress on a particular syllable.
(9) Results

 syllabic structure antepenult penult

PN context 1. CV.CV.CV 31 41
 2. CV.CV.CVC 12 14
 3. CV.CVC.CV 10 59
 4. CVC.CVC.CVC 24 9

CN context 1. CV.CV.CV 23 50
 2. CV.CV.CVC 17 9
 3. CV.CVC.CV 11 57
 4. CVC.CVC.CVC 15 11

 syllabic structure final N

PN context 1. CV.CV.CV 8 = 80
 2. CV.CV.CVC 54
 3. CV.CVC.CV 11
 4. CVC.CVC.CVC 47

CN context 1. CV.CV.CV 7 = 80
 2. CV.CV.CVC 54
 3. CV.CVC.CV 12
 4. CVC.CVC.CVC 54

(10) Frequences in percentage

 syllabic structure antepenult penult

PN context 1. CV.CV.CV 38.75 51.25
 2. CV.CV.CVC 15 17.5
 3. CV.CVC.CV 12.5 73.75
 4. CVC.CVC.CVC 30 11.25

CN context 1. CV.CV.CV 28.75 62.5
 2. CV.CV.CVC 21.25 11.25
 3. CV.CVC.CV 13.75 71.25
 4. CVC.CVC.CVC 18.75 13.75

 syllabic structure final total

PN context 1. CV.CV.CV 10 = 100
 2. CV.CV.CVC 67.5
 3. CV.CVC.CV 13.75
 4. CVC.CVC.CVC 58.75

CN context 1. CV.CV.CV 8.75 = 100
 2. CV.CV.CVC 67.5
 3. CV.CVC.CV 15
 4. CVC.CVC.CVC 67.5

It is obvious at first sight that there is not much difference in the distribution between the PN and the CN context. In row category comparison (chi-square test of homogeneity with two independent samples) between the two contexts we get the following results:
(11) PN vs. CN

syllabic structure [chi square]

1. CV.CV.CV. 2.14
2. CV.CV.CVC 1.95
3. CV.CVC.CV 0.13
4. CVC.CVC.CVC 2.75

None of these results is statistically significant, which means that the subjects do not treat common nouns and names differently from the point of view of stress. This suggests that Spanish stress should not be analyzed in a dual model where regular and irregular forms are processed differently (consequently where proper names and common nouns are processed differently). This means that either (i) the whole system shows rule-based behavior or (ii) the whole system is lexical.

As the article also aims to find out whether quantity sensitivity is still active in Modern Spanish, for the present discussion we have to compare the results of group 1 with those of group 3 and the results of group 2 with those of group 4, because groups 1 and 3 contain vowel-final test words, while groups 2 and 4 are consonant-final. In groups 1 and 2 the penult is open, in groups 3 and 4 the penult is closed. If quantity sensitivity is active, the closed syllable should count as heavy, so a closed penult should block stress on the preceding syllable. As has been mentioned earlier, the canonical pattern is penultimate stress for vowel-final words and final stress for consonant-final ones. Actually, in a strictly rule-based account these are the only patterns we expect.

Let us start with groups 2 and 4--consonant-final made-up words. The difference in distribution between these groups (CV.CV.CVC and CVC.CVC.CVC) according to Pearson's chi-square comparison is [chi square] = 5.57, which means that the difference is not statistically significant. (For the sake of simplicity I will only give the results of the PN context, as there is no significant difference between the two contexts.) This is partly due to the fact that the majority--about two-thirds--of our informants stressed these nonsense words on the last syllable as is expected in the rule-based approach.

However, if we compare the results with the predictions made by the traditional generative approach, the difference is remarkable. The rule-based model predicts that all of these words be stressed on the final syllable (with a small margin of error), because the algorithm says, "build a trochaic foot at the right edge of the word." As these words do not have an entry in speakers' mental lexicons--as they are nonexistent--they cannot be marked as exceptions for stress retraction. The results do not indicate the functioning of a systematic and productive "default" algorithm.

The "patterns-in-the-lexicon" approach predicts this diversity in native speakers' judgments. According to this model we expect tendencies: one pattern being more frequent than another. Not only the structure of the syllable plays a role in the decision, but other factors too, for instance, segmental composition. So if a nonsense word resembles an existing proparoxytonic word, it is more likely to be stressed like that word. Since retrieval is blocked, the form will be "derived" on the basis of analogy to this existing similar form. This model predicts that the results of the nonce-word test should reflect the actual distribution of different stress patterns in Spanish words. This is not quite borne out either--it is shown in (3) that about 80% (the average of group 2 and group 4) of consonant-final Spanish nouns in DRAE are stressed on the last syllable and hardly any are stressed on the antepenult, but as has been mentioned above, DRAE is not a living corpus to compare our results with. Another explanation that comes to mind is that many fairly new borrowings are consonant-final and a significant part of these belong to group 4 according to their syllabic composition. They contain a closed penult and a closed final syllable, and many of these words do not comply with the canonical stress pattern. Consider recent borrowings like badminton, `badminton' remington <type of rifle>, or proper names like Washington for instance. The nonexistent test words may also be felt to be "borrowings" or "foreign" and form a separate subgroup in the lexicon. This is where intellectual reflexiveness may play a role, but it still shows that CN context and PN context are treated alike and still does not mean that speakers would assign a stress pattern to these made-up words that violates the active constraints of the Spanish stress system. The situation is somewhat similar to English, where [+Latinate] words show special behavior in comparison with the rest of the vocabulary. It seems that consonant-final made-up words in Spanish are likely to "join" this group of borrowings in phonological phenomena such as stress assignment, the result of which is that they are more often stressed on the antepenult than is expected on the basis of the Spanish (core) vocabulary. A reviewer mentions that "foreignness" is not a very likely explanation since there are place names like Helsinki (very foreign indeed) that have a canonical stress pattern. In my view, this indicates precisely that stress is not assigned by a rule (which states that foreign names are marked for stress retraction, or should be stressed on the antepenultimate syllable), but the decision about the place of stress is made on the basis of associations to words in speakers' mental lexicons. Since contrary to rules, associations are not "all or none" but rather of graded strength, foreign names with a closed penult may become proparoxytonic or paroxytonic or oxytonic in Spanish.

A frequently heard argument in favor of quantity sensitivity in Spanish is that from loan words like Robinson "we learn only that Spanish speakers can mimic the stress contours of English [foreign] words" (Harris 1992: 15), but this does not prove the quantity insensitivity of Modern Spanish. However, in a strictly rule-based stress system, like that of Hungarian, for instance, where stress always falls on the first syllable of the word, such mimicry is much less likely. For instance, the name Speedy Gonzales is often segmented as Speedygon Zales by Hungarian speakers who are exposed to the "right" stress pattern in the song. Those who encounter the name first in writing tend to pronounce it as [`spi:di `gonza:les] even if later they hear it stressed differently. So if quantity sensitivity was an active rule or constraint in Spanish, it should cause stress shift in words like remington and Robinson. Note that the claims that quantity sensitivity is not active in Spanish and that stress is lexical does not mean that one cannot assume the existence of ACTIVE well-formedness constraints like that of the three-syllable window.

Roca (1999) also claims that it is unlikely that native speakers simply mimic foreign stress patterns. First, because some items (e.g. Liverpul) undergo complete metrical regularization (like Helsinki). Second, because in the case of simple mimicry they should be able to mimic forms like Scheveningen or Bratislava with pre-antepenultimate stress, which obviously they do not do, rather changing the stress pattern of these words to satisfy the three-syllable window.

An additional issue that has to be mentioned here is why speakers stress nonsense words in group 4 less frequently on the penultimate syllable. A possible explanation is that the first "borrowings" of the violating stress pattern, such as badminton, started to form a subgroup in the lexicon, and the "newcomers," due to analogy, were/are also more often stressed on the antepenult, if treated as a borrowing, and on the final syllable, if they conform to the canonical stress pattern, but not on the penult.

Now we will proceed to vowel-final words. In this group the difference in stress assignment between the first group with open penult and the third group with closed penult is statistically significant: [chi square] = 14.47 (p < 0.05). This is not surprising, if we compare existing Spanish words with these syllabic compositions. In a rule-based account we do not really expect anything but penultimate stress in both groups. The "patterns-in-the-lexicon" model predicts more proparoxytones in group 1 (open penult), which is borne out (38.75% vs. 12.5% in group 3); see (10). However, (3) shows that there is only one word in DRAE with a closed penult and stress on the third-to-last syllable. Why did 12.5% of our informants assign antepenult stress to CV.CVC.CV test words? I think the explanation lies, first, in not considering DRAE as an absolute point of reference, and, second, in the "borrowing"-like interpretation of these words. As there actually are recent loans with a closed penult and stress on the antepenult that belong to our [+borrowing] subgroup, any word that does not have a proper lexical entry might be interpreted as a word belonging to this group. So native speakers are likely to stress "funny-sounding" words on their antepenult independently of the syllabic composition of the word. This phenomenon (and the existence of such borrowings) unambiguously demonstrates that quantity sensitivity is not operative any more, and judgments about the place of stress are not based on strictly defined rules but rely heavily on the lexicon and the analogical influence of existing lexical items.

The assumption that Spanish stress is lexical and speakers' judgments rely on the lexicon is further strengthened by the results of a pilot test I carried out earlier. (Although, because of the low number of items examined, this point is in need of further research.) The pilot test included the nonce word muculo in both CN and PN context. In PN context one person out of 21 assigned penultimate stress to the word, while everybody else stressed the word on the antepenultimate syllable. Where the same word appeared in CN context, all the 21 informants stressed the word on the antepenultimate syllable. Again, there is basically no difference between the results in the two contexts. But why do people feel that this word should bear antepenultimate stress? I think because it is almost homophonous with the existing noun musculo `muscle' and very similar to lobulo `ear lobe', testiculo `testicle', which are proparoxytonic. (13) We must conclude that the system is not sensitive to syllable weight--at least if the syllable is closed by a consonant--and stress in nonce forms is assigned by analogy to existing lexical items.

5. Conclusion

The present paper makes two claims. The first claim is that stress in Modern Spanish is not quantity-sensitive, that is to say, the locus of stress does not depend on the syllabic structure of the word, at least not in the way proposed in earlier metrical accounts. The made-up-word test presented in sections 3 and 4 and the existence of fairly recent loans like badminton and remington show that words containing a closed penultimate syllable allow stress on the antepenultimate syllable. The apparent quantity sensitivity of the system is a lexical heritage (that is why the core vocabulary of Spanish conforms to it), but it is not an active rule or constraint. The three-syllable window, on the other hand, cannot be violated, so it is an active well-formedness constraint in the language.

The second claim in the article is that Spanish stress is lexical (analogical). The nonce-word test seems to prove that the decisions about the place of stress in forms that are nonexistent, hence cannot be retrieved from the mental lexicon, are made on the basis of associations to stored items, which are (probably) in some way similar to the unfamiliar word. In the test shown we cannot discover the operation of a "default rule" in the stress assignment of made-up words. The case of Spanish nominal stress shows that not everything that is regular (oxytonic, paroxytonic, and proparoxytonic stress) is necessarily rule-based.
Hungarian Academy of Sciences

Received 5 March 2001
Revised version received
6 December 2001


* I would like to thank Peter Rebrus, Peter Siptar, Miklos Torkenczy, and two anonymous reviewers for reading and commenting on earlier versions of this article. I am also grateful to Jose Ignacio Hualde for bringing various papers on the topic to my attention and always readily answering my questions. I also thank Ann Kelly for a great deal of technical help. All errors are, of course, mine. Correspondence address: Kiraly u. 28III/62, 1061 Budapest, Hungary. E-mail:

(1.) The word is obsolete. (Examples are given in their usual Spanish orthographic form and additional accent marks are provided.)

(2.) Although I do agree with Larramendi that syllable weight does not count, his evidence is not so strong since his examples all contain sC clusters in crucially relevant positions. It is well known that the behavior of s is problematic cross-linguistically. In Latin it sometimes seems to close the syllable, while in other cases it does not count for syllable weight. In English it behaves similarly; see Harris (1994).

(3.) See Barkanyi (2001) for the treatment of alternating diphthongs in Spanish and Barkanyi (forthcoming) for nonalternating rising diphthongs and falling diphthongs.

(4.) See, for instance, Marslen-Wilson et al. (1994); or Baayen et al. (1997) and the references therein.

(5.) See, for instance, Rumelhart and McClelland (1986).

(6.) The debate about the division of verbal and nonverbal stress in Spanish has a long history. In this paper I do not want to consider this polemic.

(7.) Roca (1990) also discusses the role of historical accidents in synchronic generalizations.

(8.) In Hungarian, for instance, a noun that behaves irregularly with certain suffixes as a common noun will behave regularly if it is used as a proper name: e.g. tukor [right arrow] tukr + ot `mirror, acc.' but as a PN Tukor [right arrow] Tukor + t.

(9.) In group 3 the only word found is tabarro <a type of wasp>, a very rare noun. This word is ill-formed in a quantity-sensitive analysis only under the assumption that the trill represents a geminate, as in Harris (1983). Group 4 proparoxytonic words in the dictionary are chancharras mancharras `pretexts' (which is a doublet), badminton `badminton', and remington `rifle'.

(10.) Roca (1997) calls this rule "rightmost stem accent."

(11.) Word-final [x] appears in very few words in Spanish and it is frequently dropped, so none of the nonce words contained this segment word-finally. Since the present article does not deal with diphthongs, made-up words containing a glide are not analyzed here.

(12.) Cada frase contiene una palabra inexistente que va en negrita, despues de leer la frase con atencion, por favor indica donde llevaria la palabra el acento (la fuerza de voz) si fuera palabra espanola. Por ejemplo:

Esta pacara me gusta mucho.

(13.) A reviewer points out that speakers' decision might be due to avoidance of homophony with the vulgar form culo `arse', as muculo would sound like the mispronunciation of mi culo `my arse'. However, such avoidance of homophony is not typical within forms involving different bases (see Kenstowicz 2001).


Alameda, J. R.; and Cuetos, F. (1995). Diccionario de frecuencias de las unidades linguisticas del castellano [Dictionary of the Frequencies of Linguistic Units in Spanish]. Oviedo: University of Oviedo Press.

Aske, Jon (1990). Disembodied rules versus patterns in the lexicon (testing the psychological reality of Spanish stress rules). Berkeley Linguistics Society 16, 30-45.

Baayen, Harald; Dijkstra, Ton; and Schreuder, Robert (1997). Singulars and plurals in Dutch: evidence for a dual parallel route model. Journal of Memory and Language 37, 94-119.

Barkanyi, Zsuzsanna (2001). Alternating diphthongs in Spanish. In Selected Papers from the 5th Graduate Students' Linguistic Symposium, 1-10. Budapest: RIL, Hungarian Academy of Sciences.

--(forthcoming). The phonemicity of glides in Spanish: cambia vs. rocia. In Verbum: Analecta Neolatina 2001/2. Budapest: Akademiai Kiado.

Burzio, Luigi (1996). Surface constraints versus underlying representation. In Current Trends in Phonology: Models and Methods, J. Durand and B. Laks (eds.), 123-141. Salford: European Studies Research Institute, University of Salford Publications.

Chomsky, Noam; and Halle, Morris (1968). The Sound Pattern of English. New York: Harper and Row.

Daelemans, Walter; Duriex, Gert; and Gillis, Steven (1994). The acquisition of stress: a data-oriented approach. Computational Linguistics 20(3), 421-451.

Dunlap, Elaine (1991). Issues in the moraic structure of Spanish. Unpublished dissertation, University of Massachusetts.

Eddington, David (2000). Spanish stress assignment within the analogical modeling of language. Language 76(1), 92-109.

Harris, James W. (1983). Syllable Structure and Stress in Spanish. Linguistic Inquiry Monograph. Cambridge, MA: MIT Press.

--(1992). Spanish Stress: The Extrametricality Issue. Washington, D.C.: IULC.

Harris, John (1994). English Sound Structure. Oxford and Cambridge, MA: Blackwell.

Jaeger, Jeri J.; Lockwood, Alan H.; Kemmerer, David L., van Valin, Jr., Robert D.; Murphy, Brian W.; and Khalak, Hanif G. (1996). A positron emission tomographic study of regular and irregular verb morphology in English. Language 72(3), 451-497.

Kenstowicz, Michael (2001). Paradigmatic uniformity and contrast. Unpublished manuscript, MIT.

Larramendi, Manuel de (1729). El imposible vencido. Arte de la lengua bascongada [The Impossible Won. Art of the Basque Language]. Salamanca: Antonio Joseph Villargordo Alcaraz.

Lukacs, Agnes (1999). Szabalyok es kivetelek: a magyar morfologia szabalyos es rendhagyo alakjainak tarolasi es feldolgozasi jellegzetesegei [Rules and exceptions: properties of storing and processing regular and irregular forms in the morphology of Hungarian]. Unpublished MA thesis, Eotvos Lorand University, Budapest.

Marcus, Gary F.; Brinkmann, Ursula; Clahsen, Harald; Wiese, Richard; and Pinker, Stephen (1995). German inflection: the exception that proves the rule. Cognitive Psychology 29, 189-256.

Marslen-Wilson, William D.; Tyler, Loraine Komisarjevsk; Waksler, Rachelle; and Older, Lianne (1994). Morphology and meaning in the English mental lexicon. Psychological Review 101 (4), 3-33.

Roca, Iggy (1988). Theoretical implications of Spanish word stress. Lingistic Inquiry 19, 393-424.

--(1990). Diachrony and synchrony in Spanish stress. Journal of Linguistics 26, 133-164.

--(1997). On the role of accent in stress systems: Spanish evidence. In Issues in the Phonology and Morphology of the Major Iberian Languages, F. Martinez-Gil and A. Morales-Front (eds.). Georgetown: Georgetown University Press.

--(1999). Stress in the Romance languages. In Word Prosodic Systems in the Languages of Europe, Harry van der Hulst (ed.). Berlin: Mouton de Gruyter.

Rumelhart, David E.; and McClelland, James E. (eds.) (1986). Parallel Distributed Processing, vol. 1. Cambridge, MA: MIT Press.

Skousen, Royal (1995). Analogy: a non-rule alternative to neural networks. Rivista di Linguistica 7, 213-232.

--(forthcoming). Analogical Modeling.
COPYRIGHT 2002 Walter de Gruyter GmbH & Co. KG
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2002 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Barkanyi, Zsuzsanna
Publication:Linguistics: an interdisciplinary journal of the language sciences
Date:Mar 1, 2002
Previous Article:Is there a Spanish imperative intonation revisited: local considerations *.
Next Article:Spanish diminutive formation without rules or constraints (1).

Related Articles
SAUDI ARABIA - July 7 - Spanish Olive Oil Banned.
Crazy world: Poison in Euro coins.
KEVANON BRINK OF FREEDOM; Family pay pounds 100 to court.
CAMERA CLUE TO MISSING COUPLE; New twist as man who vanished `is seen on TV'.
Material characterization. (Instruments).
Material characterization. (Instruments).
Today's recipe.

Terms of use | Privacy policy | Copyright © 2019 Farlex, Inc. | Feedback | For webmasters