Language Evolution: The Windows Approach addresses the question: "How can we unravel the evolution of language, given that there is no direct evidence about it?"
The LINGUIST List is dedicated to providing information on language and language analysis, and to providing the discipline of linguistics with the infrastructure necessary to function in the digital world. LINGUIST is a free resource, run by linguistics students and faculty, and supported primarily by your donations. Please support LINGUIST List during the 2016 Fund Drive.
Date: Tue, 09 Aug 2005 20:40:30 -0400 From: Elizabeth Pyatt : Subject: Paradigms in Phonological Theory
EDITORS: Downing, Laura J.; Hall, T. Alan; Raffelsiefen, Renate TITLE: Paradigms in Phonological Theory PUBLISHER: Oxford University Press. YEAR: 2005
Elizabeth J. Pyatt, Penn State University
This volume is a series of articles focusing on theoretically defining how output forms can be remodeled based on a paradigm within the Optimality Theory model. Linguists working with morphologically complex languages, especially their historical aspects, are all familiar with changes in word forms showing the effects of ''paradigm leveling'' or ''analogy'' in Neogramammarian terms. The challenge for modern theoretical linguists has been how to consistently model these changes so that the data can be accounted for without necessarily resorting to a series of ad hoc accounts.
One of the appeals of Optimality Theoretical accounts of morphophonological phenomena is that it postulates that grammars include a variety of competing constraints which, depending on which constraint has priority, causes different output forms to be realized. An account of paradigm leveling can be devised by invoking the notion of paradigm identity constraints requiring forms to have similar forms outranking competing phonological or phonetic constraints or outranking constraints on preserving underlying original lexical input forms.
As this collection of papers show though, formulating the constraints so that correct output forms are produced is still a complex task. Linguists interested in the nature of morphological paradigms and issues of how morphology and phonology interact will find in-depth discussions of a variety of issues including how paradigms are defined, how base forms are determined and how different constraint rankings change which output forms win. These articles cover a wide range of languages including Hebrew, Latin, Jita, English, Hungarian, Spanish and others. This volume also includes theoretical articles by McCarthy and Kenstowicz.
SUMMARY (WITH SOME DISCUSSION)
The first article is the introduction by editors Laura Downing, T. Alan Hall and Renate Raffelsieffen. The article provides an overview of how paradigms in were analyzed in linguistic theory from 19th century on. For much of the time, paradigms were viewed as peripheral or an epiphenomenon in phonological analyses. Morphology was not ignored altogether, as can be seen in the theory of lexical phonology, but some paradigmatic effects were attributed to cyclic application of some phonological rules interleaved with affixation. With the advent of Optimality Theory, the relation of morphology and phonology was reevaluated and Downing, Hall and Raffelsieffen identify several theoretical proposals. To account for ''cyclic effects'', phonologists including Benua's Transderivational Correspondence Theory (1997) proposed that constraints could be evaluated recursively, thus replicating the interleaving of phonology and affixation. Candidates showing cyclicity or paradigm effects are said to be obeying a high order Output-Output Identity constraint (O-O-Ident) requiring the base form to be the same in both derived forms and underived forms. However this approach is questioned by some, including McCarthy (Chapter 8, this volume) because rankings must change between recursions to account for the forms, this diverging from the OT parallelism requirement.
Another approach discussed in the introduction is a non-recursive Base- Priority approach in which a morphologically complex form is evaluated with one set of constraints, including one of a family of Output-Output (O- O) identity constraints. Most of the papers in this volume use the Base Priority approach in which individual forms are evaluated individually against a base, or the related Symmetric approach (Kenstowicz, McCarthy this volume) in which entire paradigms are evaluated together. Interestingly, Base-Priority approaches do not require the output correspondences to be always between a ''non-derived'' and ''derived'', but can be between two derived forms such as a future and imperative form (Bat El, Chapter 3, this volume). Thus , Downing et. al. note that more leveling phenomena can be accounted for. A major theme, therefore, is determining which form is the base used to evaluate other members of the paradigm.
The second article ''The Morphological Basis of Paradigm Leveling'' Adam Albright discusses paradigm leveling within one class of Latin nominal declensions as exemplified in the change of /hono:s/ 'honor-NOM' to /hono:r/, to be more consistent with oblique forms such as /hono:ris/ 'honor-GEN'. Historically, the root for 'honor' was s- final /hono:s-/ resulting in older forms like /hono:sis/ 'honor-GEN'. At some point, Latin /s/ surfaced as /r/ in intervocalic position, causing forms like /hono:sis/ to surface as /hono:ris/, while /hono:s/ 'honor-NOM' remained unaffected. Thus this noun had two forms of a root in its paradigm -- /hono:s/ (nominative) and /hon:or/ -- (all other forms). This was leveled in favor of the -r form via a O-O-Ident constraint. The question Albright asks is why the oblique -r form was chosen as a candidate base. Intuitively, it has been known that oblique forms (genitive, dative, etc) should be treated as the base form over nominative forms; hence Albright notes that Latin paradigm leveling effects should favor oblique forms. As evidence for favoring the oblique as the base over the nominative, Albright provides statistical evidence from corpus studies showing that oblique forms are more often written in texts than the nominative. In addition, Albright points out that later Romance borrowings like Spanish ''noche'' 'night', French ''nuit'' 'night' and Italian ''notte'' 'night' are descendents of the Latin oblique root for night , ''nocte'' /nokte/ with a /t/, not the nominative ''nox'' /noks/ with no /t/. Finally, Albright notes, that there are a variety of nouns in Latin where the oblique form cannot be predicted from the nominative. As a minimal set, he mentions four nominatives ending with -us which have differently formed obliques (''populus, populi'' 'people', ''corpus, corporis'' 'body', /manus, manu:s/ 'hand', ''genus, generis'' 'kind'). Based on statistical frequency and predictability across paradigms, Albright argues that Latin speakers considered the oblique to be the true base, and favored for leveling effects.
In the third article, ''Competing Principles of Paradigm Uniformity: Evidence From the Hebrew Imperative Paradigm,'' author Outi Bat-El discusses the forms of modern Colloquial Hebrew imperatives in relation to other members of the verb paradigm, in particular the future. Like other Semitic languages, Hebrew verbal morphology is characterized by combinations of a series of consonantal roots representing the verb lexeme combined with different templates of vowels, affixes and truncations for each verb form. Bat-El argues that the Hebrew imperatives are formed by truncating the initial segments of a future form where ranked markedness constraints determine how much is deleted. Hebrew verb tenses are marked for gender, so the masculine future is truncated to form a masculine imperative, while feminine futures are truncated to form a feminine imperative. In some cases though, Bat-El argues that the constraint hierarchy does not allow a winning candidate to surface from truncation of the future. In these cases, if a masculine imperative can be formed, then the feminine imperative will be formed from the masculine future base plus a feminine marker. Therefore, there is competition in which form will be used as a base. In Optimality terms, Bat-El analyzes this as Existence (a form must be realized) outranking Structural Identity (a form uses the same base). However, if not even a masculine can be formed, then no imperatives will be formed; that is Structural Identity out ranks Existence in this case. To account for this paradoxical change in ordering, Bat-El proposes that the future-imperative is a ''sub-paradigm''. As long as one imperative can surface based on the future, then the constraints allow both imperatives to be formed. Bat-El contrasts the formation of the imperative in Colloquial Hebrew with Tiberian Hebrew. At that stage, a number of markedness constraints prevent the imperative and future form from surfacing with identical bases, However there are no gaps in the Imperative, meaning Existence is respected throughout the entire paradigm. In Colloquial Hebrew, Output-Output constraints prevent alternate bases from surfacing, even if gaps are formed.
The next article is Luigi Burzio's ''Sources of Paradigm Uniformity'' which focuses on the relationship on ''closeness'' of two related forms and the tendency for paradigm leveling to occur. Bruzio gives the example of the English word ''larynx'' /'lær.Inks/ shows stress shift in the irregular plural ''larynges'' /la.'rIn.jis/, but not in the regular plural ''larynxes''. Similarly, Spanish inflected imperfect forms maintain consistent stress, but exhibits stress shift in the infinitive (1). Note that in the case of ''amábamus'' 'we loved', the effects of paradigm levelling have caused the stress to shift from the original position it had in Latin.
(1) Forms of Spanish ama 'love' a'ma-bas 'you loved a'má-bamos 'we loved' < ama'ba:mus (Latin) a'mar 'to love' < a'ma:re (Latin) a'mable 'loveable'
The intuition is that forms which are semantically and morphologically ''closer'' are more likely to show effects of paradigm leveling. Burzio proposes that ''closeness'' can be calculated as a gradient using Hebb's (1949) notion of entailment relationships between two items. Criteria for determining whether entailments exist between forms are closeness in meaning and sharing phonological (segmental/metrical) material. Forms sharing a number of entailments, may be ''attracted'' enough, in some cases the attraction is ''strong'' enough that paradigm leveling may cause paradigm leveling where phonological information in forms is made to be closer or even identical.
The next section is an OT account of English ''Level 1'' affixes (e.g. -ive) differ from ''Level 2'' (e.g. -ness) affixes. I had some issues with this portion of the discussion that I will discuss in the following paragraph. The final section of the article returns to the issue of morphological parameters for determining closeness. Burzio's view is that if forms are close enough semantically, and share some phonological material, then they may be attracted enough to cause paradigm leveling effects. One criteria is whether two forms share the same part of speech. For instance, the Spanish forms ''amabas'' 'you loved' and ''amábamu'' '' we loved' are both imperfect inflected finite verbs, while the infinitive ''amar'' is a non- finite verb and the related adjective ''amable'' 'loveable' are different parts of speech. Therefore the attraction between finite verb forms would be stronger, and more subject to paradigm leveling than between a non- finite verb and an infinitive or adjective. Burzio also proposes that marked categories may be more subject to paradigm leveling because they contain more entailments that must be satisfied. As an example, Burzio considers an Old English subjunctive paradigm that would [+subjunctive] entailment links missing indicative paradigm. As a result, Burzio a subjunctive form might be more liable to paradigm leveling, and that does appear to be a general linguistic trend.
As mentioned before, the middle portion of Burzio's article is an analysis of English Level 1 affixes that cause vowel changes and stress shifts versus Level 2 affixes which do not change the stem. His basic analysis is that because words ending with Level 1 affixes are semantically irregular, the entailments between forms are weakened to the point where additional phonological processes and alternate roots can override an OO constraint for stems. A corollary that Burzio proposes that ''Level 1'' affixes which cause stem changes change the stem to a less phonologically marked form. An example cited by Burzio is that a regular past tense like ''weeped'' /wi:pt/ contains a marked combinationo of /i/ plus /pt/ which are normally found ONLY in derived roots. This -ipt cluster is allowed only because OO constraints or entailments between ''weep'' and ''weeped'' force the root to be the same. In contrast, the irregular plural ''wept'' /wEpt/ contains the less marked /Ept/ cluster (also found in underived ''sept''). As I interpret this article, Burzio is predicting that Level 1 rules cause forms to surface in a less marked form than a Level 2 affix would.
Yet there are counterexamples of ''Level 1'' (or irregular) affixes creating more marked clusters, not less marked. For instance, Welsh has a number of plural affixes, the most regular of which is probably ''-(i)au'' as in the English borrowing ''iard'' /yard/, plural ''iardiau'' /yardye/. Welsh also contains a number of stem changing plurals where vowels change between the singular and plural. One pattern is a change of the diphthong ''ia'' /ya/ in the singular to the triphthong ''iei'' /yey/ in the plural as in Welsh ''iarll'' /yarL/ 'earl' to plural ''ieirll'' /yeyrL/. Historically, this is a preservation of a fronting rule triggered by Proto Celtic plural /i/ which was deleted in later states, The synchronic result is a cluster which is marked because it is a triphthong closed by two liquids. Note that all other derived forms of ''iarll'' (iarlles 'countess', pl. iarllesau, iarllwaed 'noble blood', show the same stem as singular ''iarll.'' In addition, there are very few cases of ''ieir''clusters in underived forms; most are derived from ''iar'' singulars (e.g. ''patriarch'' 'patriarch', ''patrieruch'' 'patriarchs' (regular ''patriarchau'' is also attested). Whatever analysis is given to this plural formation, it would be difficult to argue that the ''iei'' cluster is less marked than ''ia''. These facts suggest that ''Level 1'' alternations effects cannot all be derived from phonological markedness constraints within the synchronic grammar. Note that all Welsh data is available in the Concise Welsh Language Dictionary/Geriadur Prifysgol Cymru (2004).
Another question I had with the article was that I was not able to determine a reliable method for predicting ''closeness''. For instance, how ''close'' would two forms have to be in order for paradigm leveling effects to occur? It seems reasonable that a verb and a nominal (e.g. ''compel/compulsive'') would have enough semantic distance to not be subject to paradigm leveling, but what about irregular inflected forms such as ''keep/kept''). Burzio's article did not appear to have any kind of principle predicting the maintenance of irregular inflected forms (other than possible cases of phonological markedness). If anything, one might predict that common verbs and nouns are LESS likely to be irregular because of semantic closeness. Instead they are MORE likely to be irregular. One criterion could be that irregulars have different phonological roots, and therefore are not as ''close,'' but this to means seems circular. Irregulars are not as close because they are irregular forms. I feel this is an interesting conceptualization, but I was not sure how to apply it to other data.
The next article, Start Davis's ''Capitalistic'' vs ''Militaristic'': The paradigm Uniformity Effect Reconsidered'' is an extensive review of Steriade (2000) in which she notes that ''militaristic'' has an aspirated [th] in he stem while ''capitalistic'' has a flap [D]. Steriade originally argues that ''capitalistic'' is regular and that ''militaristic'' preserves the [th] due to paradigm optimization with ''military''. However, Davis examines other English data and argues that ''militaristic'' is actually the regular form, while ''capitalistic'' is preserving the foot structure of ''capital''. Davis presents a series of words in which the voiceless stops are aspirated despite that fact that they are non-word initial, not the beginning of a foot and in stressless syllables. Some of these include (2):
(2) English words with aspirated stops in unstressed syllables. Medi[th]erranan Winne[ph]egosis Nebu[kh]adnezzar
Since these are underived words, a paradigm leveling account could not account for the aspirated vowels. Instead Davis argues that these weak syllables are adjoined to the following two-syllable foot to create a ''super foot'' of three syllables. He uses data for expletive insertion to show that ''fucking'' can be inserted before or after the target syllables. According to this analysis, the initial /t/ ''mili[th]aristic'' is aspirated by regular phonology. However, ''capi[D]alistic'' (derived ''capital'') does not show aspiration. Davis notes that the /t/ in ''capital'' cannot be adjoined into a super foot because only syllable follows it; therefore the /t/ is a flap. In ''capi[D]alistic'', Davis argues that this foot structure is preserved, causing a flap to surface. The analysis of aspiration does seem to match both the data and my native speaker intuition, but interestingly, this is one case where a traditional cyclic analysis might work. It's also striking that Davis argues for a paradigm relationship between ''capital'' and ''capitalistic'' (with a Level 1 affix) whereas Burzio would argue that the relationship would be more distant so that paradigm leveling might not be predicted.
In the sixth article, ''Jita Causative Doubling Provides Optimal Paradigms''., Laura J. Downing discusses a phenomenon in the Bantu language Jita where the causative marker /y/ is attached to all affixes as in ''oku- gus-i:s-y-an-y-a'' ''to sell to each other'' (lit: cause each other to buy) where /y/ appears after the root ''gus'' (as /i:/) and again before the reciprocative and benefactive suffixes. Downing examines older cyclic analyses and autosegmental floating analyses before arguing that causative /y/ can be displaced via special alignment constraints and that doubling is a way to optimize the output of causative forms.
Article seven, ''Paradigmatic Uniformity and Contrast'' by Michael Kenstowicz gives an overview of different types of paradigm leveling effects that can be found in data from a variety of languages. In particular, he distinguishes paradigm effects which cause forms to surface in a more phonologically similar manner, and paradigm effects which ensure that different members of a paradigm have distinct forms. An example of paradigm effects making forms more similar can be found in Spanish diminutive formation. Spanish diminutives are typically formed with the ending ''-it(o/a)'' (3a), but the ending surfaces as -''cit(o/a)'' when the root ends with a consonant. (3b)
For some nouns though, if the masculine ends in a consonant, both the masculine and feminine counterparts surface with ''-cit(o/a)'' even if the feminine ends with a vowel. For instance masculine ''ladron'' 'thief' has a diminutive ''ladroncito'', but the feminine ''ladrona'' also has diminutive ''ladroncita'' instead of predicted ''ladronita.'' Kenstowicz proposes that a high ranking ParadigmUniformity constraint forces feminine ''ladroncita'' to surface in parallel to masculine ''ladroncito''. The paradigm effects forces the forms to be more similar.
In contrast, Kenstowicz presents a number of cases from Slavic declensions where paradigm effects appear to block cases where two forms would surface in the same form. An example is the Trigad dialect of Bulgarian where unstressed /o/ is normally reduced to /a/ (4a). However, this is blocked in the neuter singular /o/ whenever the counterpart plural ends with unstressed /a/ (4b).
Kenstowicz proposes that a high ranking Paradigm Contrast constraint keeps these two forms apart. If the /o/ to /a/ reduction had applied as expected, the singular and plural of ''zorno/zorna'' 'seed' would have merged into one form ''zorna''. This is an excellent article, and in some ways I wish it had been the first. It clearly lays out what kinds of paradigm effects can be found and what some of the theoretical implications might be. The article also notes that much more work in this area of analysis is needed.
The eighth article, ''Optimal Paradigms,'' by John McCarthy is a theoretical article in which McCarthy proposes an Optimal Paradigm (OP) algorithm in which all members of a paradigm are evaluated jointly for faithfulness with each other. McCarthy first reviews other approaches and presents theoretical arguments for rejecting them. For instance. McCarthy argues that base priority approach like Transderivational Correspondence Theory (TCT) (Benua 1997) cannot work for consonantal root languages like Arabic because no ''root'' can be identified, yet he argues that paradigm effects can still be found McCarthy also rejects Universal Exponence (UE) (Kenstowicz 1996), because it overpredicts that derived forms can affect base forms (and McCarthy notes that there is little evidence for this occurring in linguistic data).
Instead McCarthy hopes to combine the best of both approaches into his Optimal Paradigm (OP) approach by evaluating all candidates of a paradigm together for Input-Output and Output- Output constraints. That is, candidates would consist of entire hypothetical paradigms where OP violations would be counted as one violation per form (5).
McCarthy defines an inflectional paradigm as containing ''all and only the words based on a single lexeme'' and specifically assumes they are ''flat'' although does allow for the possibility of ''subparadigms.'' perhaps similar to those argued for in Hebrew by Bat El (same volume). Other factors proposed by McCarthy include ''majority rules'' and attraction to the unmarked favoring leveling in favor of less marked segments. To provide empirical examples, McCarthy discusses several cases from Arabic verbal morphology.
In the ninth article, ''Paradigm Uniformity Effects Versus Boundary Effects'', Renate Raffelsiefen reviews English data where morphologically complex words may have different phonological properties and phonotactic constraints than those of simple words. The goal of this article is to distinguish boundary effects from paradigm uniformity effects and propose a set of criteria for distinguishing between boundary effects and paradigm leveling
Article Ten, ''Uniformity and Contrast in the Hungarian Verbal paradigm'' by Peter Rebrus and Miklos Torkenczy discusses the relation of two families of paradigm constraints and how they interact in some Hungarian verbal paradigms. The first, PAR, is a constraint specifying that the forms of stems and endings remain constant throughout a paradigm; the second, CON specifies that each member of the paradigm be phonologically distinct. One phenomenon discussed by the authors are ''anti-harmony'' effects in Hungarian suffixes. Hungarian suffixes normally harmonize with the stem vowel in terms of backness, but the harmony appears to be blocked in some cases. Rebrus and Torkenczy argue that anti-harmony occurs because harmonic affixes might create homophonous forms in some paradigms; thus harmony is blocked when contrast in person and number might be lost.
The other phenomenon Rebrus and Torkenczy discuss the definite versus indefinite contrast in the Hungarian present indicative which disappears in the past (i.e. only forms corresponding to the definite surface, but can be either definite or indefinite). To account for this, the authors propose that the CON(Person/Number) (Contrast in person and number) is ranked higher than CON(Definiteness). In cases where homophony might occur, such as in the past tense, the definiteness distinction is lost so that PAR(Person/Number) and CON (Person/Number) can be satisfied. Although this analysis is consistent, I wonder if another analysis for the loss of the definiteness distinction might also be valid. The Distributed Morphology framework (Halle and Marantz 1994) proposes that ''peripheral'' features may be deleted when the feature bundles become too complex; this process is called Impoverishment. For instance, many Indo-European languages like English show gender in third person singular pronouns (he/she/it), but lose gender in plurals (they) and non-third person forms (I/you/we). The Hungarian case where a peripheral feature of definiteness is lost in non-present tenses could be a similar phenomenon.
The final article is ''A Note on Paradigm Uniformity and Priority of the Root,'' by Suzanne Urbanczyk which analyzes data from verb paradigms in Halkomelem, a Salish language of British Columbia using McCarthy's (this volume) mechanism of comparing outputs across a paradigm. Like other Salish languages, the root vowel of Halkomelem verb stems is reduced to schwa, particularly in cases where a CVC root is followed by a CVC suffix. In perfective stems though, all forms surface with a schwa, even if the ending is not CVC. Urbanczyk argues that this pattern of overapplication is a result of a high ranking identity constraint requiring all members of a paradigm to have the same base form, even if results in marked reduced vowels in the stem. Urbanczyk also discusses the formation of the imperfective root from the perfective via reduplication. In other Salish languages, the imperfective reduplication preserves any schwa from the perfective, but in Halkomelem, the reduplicated syllable surfaces with the underlying root vowel. Urbanczyk's proposes that the root vowel is allowed to surface as phonologically expected because the imperfective paradigm is distinct from the perfective, so the paradigm uniformity constraint does not apply in this case.
This volume is a valuable addition to the research into the morphology- phonology interface. The data comes from a wide range of languages and all of them present serious challenges to any standard derivational theory. Paradigm effects appear to be a major family of constraints much like faithfulness and markedness. But as most authors of this volume would admit, the challenge is to provide a consistent account across all languages. One issue is determining the ''scope'' of paradigm effects. Do they apply only within inflectional paradigms (McCarthy) or across derivational forms (Davis)? Do complex inflectional paradigms have an internal structure (Burzio, BatEl) or are they flat (McCarthy)? With questions like these, future researchers will have plenty of analytical articles to write.
Benua, L. (1997, 2000) Transderivational Identity: Phonological Relations Between Words. New York: Garland.
Halle, Morris and Marantz, Alec. (1994) Some Key Features of Distributed Morphology, MIT Working Papers in Linguistics, 21: 275-88.
Hebb, D. O. (1949) The Organization of Behavior: A Neuropsychological Theory. New York: John Wiley and Sons.
Kenstowicz, Michael (1996) Base-Identity and Uniform Exponence: Alternatives to Cyclicity in Current Trends in Phonology: Models and Methods, J. Durand and B, Laks, eds., Manchester: European Studies Research Institute, University of Salford.
University of Wales (2004) Geriadur Prifysgol Cymru (Welsh Language Dictionary) Concice Online Version http://www.aber.ac.uk/~gpcwww/pdf/gpc0015.pdf
ABOUT THE REVIEWER:
ABOUT THE REVIEWER
Elizabeth Pyatt earned a Ph.D. in linguistics, specializing in Celtic phonology, morphology and syntax.