The use of Optimality Theory in Word-Formation

Table of contents

1. Introduction

2. Optimality Theory

3. Case studies
3.1 "The phonology of -ize derivative": Ingo Plag
3.2. "Phonological constraints on English word formation": Renate Raffelsiefen

4. Applying OT to -ity suffixation

5. Conclusion

6. List of Constraints

Bibliography

1. Introduction

The still rather young Optimality Theory (OT) has found its way into the linguistic discipline of word-formation triggering many new insights and new points of view. This has mainly taken place when analysing the creation of new words in the Englisch language on a morphological basis. This new point of view revealed new possibilities how the speakers decide on which form is right when coining a new word. But how far is this theory developed? Can we leave all formulated rules behind such as in the form of when x attaches to y then z must be applied?

This term paper will take a closer look at how OT is applied in derivations. After a historical and explanatory chapter on the theory itself two investigations by Ingo Plag ("The phonology of -ize derivatives") and Renate Raffelsiefen ("Phonological constraints on English word formation") will be viewed to understand how the theory is applied. Finally the -ity suffixation will be observed on the basis of the results of the preceding two chapters.

To attain a textual flow for this thesis the individual constraints used by Plag and Raffelsiefen are summed up and explained in chapter six. Numbers in brackets refer to the page in their article where they have defined the constraint, whilst definitions in angle brackets represent a summary of how the author uses and understands it (only in cases when they did not formulate a definition). The reader may observe that a few constraints are listed under different names; this is dues to the fact that there does not seem to be a uniform naming scheme yet - therefore the constraints are named according to the authors use.

For the fourth chapter the Internet was a main reference to build up a corpus of -ity derived words. A search with Onelook ® Dictionary provided a selection of 1068 tokens of different derivatives after sorting out phrases, loan words and those which have undegone even further derivation (i.e. Negation) from 8560. The choice to use this online dictionary search was made because the website searches through several types of dictionaries in the web, starting from the general ones to the more specialized ones (i.e. linguitical, or medical) as well as collections of neologisms (i.e. www.wordspy.com).

2. Optimality Theory

As Diana Archangeli points out the linguistic advances of the 1970s and 1980s were hoped to be improved by simpler representation in the years to come. However, "this simplification did not happen" (Archangeli 1997: 25). A leading theory at the time was generative phonology. At this stage the development from an input to its respective output was analysed via moving from the underlying representation via the morpheme concatenation and rules to the surface representation. Each of these steps checked the development of the new output via inviolable constraints. Although the general analytical strategy has been on the right track; at the same time, there had been growing dissatisfaction in two ways. First, despite continued innovation in theories of rules and of representations, certain types of data remained unexplained. Second, the prevailing belief about constraints - that they are inviolable - resulted in a continuing frustration with their role in grammar, for it is exceedingly difficult to find a constraint that is never violated. (ibid: 27)

In regard to morphology Kevin Russell points out that the naturalness of rules could not be separated effectively by a theory such as generative phonology, which "allowed highly unnatural rules to be on an equal footing with widely attested rules," therefore "it failed to accurately reflect the linguistic competence of human beings" (Russell 1997: 106).

OT is a rather young concept in linguistic dating back to a course by Alan Prince and Paul Smolensky at the University of California in 1991. They published their first detailed exposition of this theory in 1993 (Optimality Theory - Constraint Interaction in Generative Grammar), triggering a great interest among linguists observing different topics. The former use of the theory in phonology became increasingly apparent "to topics in morphology, syntax, sociolinguistics, psycholinguistics and semantics" (McCarthy 2002: 1).

Most theories of language apply rules to each respective input in order to obtain an output. OT tackles the problem from the opposite angle namely by valuing possible output-candidates for a certain input. To do this is uses a range of violable constraints which rules out those candidates not corresponding to the specific constraints-hierarchy of the language in question. As McCarthy points out, OT is based on the idea of UG (Universal Grammar) which states that all languages are based on a universal kind of database, giving each and everyone of them a choice of many elements (i.e. In phonology the different possible phonemes). Each language has developed an order of precedence for these in which some of them take a minor role (such as allophones in example of phonology) or seem to be eliminated altogether. This order of precedence makes each language unique. In terms of OT the representative constraint-hierarchy provides this order by preferring certain forms or elements above others. The idea of a general database can thus also be applied to the choice of constraints; i.e. each language is based on the universal constraints and forms its unique hieracrchy by ranking these (ibid: 7).

McCarthy gives the following graphical explanation of "Basic OT architecture" (ibid: 10).

input -> GEN -> candidates -> EVAL -> output

Having already encountered input, candidate and output in the paragraph above, the components GEN (Generator) and EVAL (Evaluator) are still left to define. The former "generates output candidates for some input, and submits these to" the latter being "the set of constraints, which evaluates output candidates as to their harmonic values, and selects the optimal candidate" (Krager 1999: 19). The generator provides all possible candidates for the chosen input, as no rules or constraints prohibit "any conceivable output candidate" (ibid: 20). The evaluator will them filter through all of these until one optimal output for the language in question remains. This is where the constraint-hierarchy comes into play, as it is practically impossible to find candidates which do not violate any constraints at all. Therefore the hierarchy places constraints which are more important to the language in question above those of which a violation is more tolerable. In this constellation it is possible that two or more constraints might reside on the same hierarchical level in cases when the hierarchical distinction between them does not provide any difference to the result. The evaluator will thus reveal the candidate which provides the most harmony i.e. violates the higher-ranked constraints less than others.

A hierarchical ranking is represented by '»' when the constraints differ in importance i.e. C1 » C2 - with C being a universial constraint. This ranking is transitive as C1 will always dominate any constraint ranked below C2. Thus "If C1 » C2 and C2 » C3 then C1 » C3" (ibid: 21). Constraints working on the same level are summed up with commas: C1,C2 » C3 (Plag 1999:148). When accounting an investigation of an optimal ouput for a given input the following type of table is used:

Abbildung in dieser Leseprobe nicht enthalten

In this paper constraints on the same level are divided in the table by a single line, whereas the different hierarchical levels are separated by a double line. The optimal candidate is indicated via the arrow '->' whilst the violations of each constraint in question is marked with an asterisk '*'. Severe violations are indicates with an exclamation mark and multiple violations which might occur in more than one place in the candidate (i.e. in different syllables) are represented by an asterisk for each. Although the number of violations of a constraint is of no major importance "since better or worse performance is all that matters" McCarthy points out on the basis of Prince's and Smolensky's method of mark cancellation. "If and only if a tableau compares exactly two candidates, violation-marks that the two candidates share can be ignored or canceled, since those violation-marks contribute nothing to that particular comparison" (McCarthy 2002:6). Furthermore, in cetain contexts it might occur to join two constraints together when they seem to interact. In these cases they are summed up together using '&' - i.e. C1 & C2 - both in the constraint hierarchy and in the representative table, in which they take the place of one constraint. Plag points out that this idea of "local conjuntion of constraints" was coined by Smolensky:

According to this proposal, two constraints may form a composite constraint which is violated in some given domain. The conjoined constraint only has tangible effects if some other constraint(s) is/are ranked between the conjoined constraint and the individual constraints that together make up the conjoined constraint. (Plag 1999:181)

This meant that the constraints in question are represented twice in the hierarchy: once in their conjoined function, and then again separately in a position which is at least spaced two hierarchical levels lower or higher from the conjoined constraint.

This interaction between the two constraints shows how the conjoined constraint can be violated. For instance, if in this conjunction C1 may only be violated as long as C2 is not, one ends up with the result that the conjoined constraint is violated once C2 is violated.

When working with OT one will encounter two types of constraints, namely faithfulness and markedness constraints:

Faithfulness constraints require that outputs preserve the properties of their basic (lexical) forms, requiring some kind of similarity between the output and its input. Markedness constraints require that output forms meet some criterion of structural wellformedness. (…) such requirements may take the form of prohibitions of marked phonological structures (…) Note that markedness constraints refer to output forms only and are blind to the "lexical" input (Kager 1999:9f.)

These two types of constraints hold the results balanced between the extremes of either practically no change between input and output (faithfulness) or great differences between the two. Each language has weighed out the balance between these two components for itself.

To give the reader a concrete example of how OT works we will turn to Kager's examination of voicing contrasts of the coda in Dutch and English (ibid:14ff.): To investigate these contrasts he looked at the ranking of the markedness constraint *VOICED-CODA and the faithfulness constraint IDENT-IO(voice)¹. The Dutch language ranks the markedness constraint higher than the faithfulness constraint (*VOICED-CODA » IDENT-IO(voice)) whereas English emphasises faithfulness (IDENT-IO(voice) » *VOICED-CODA). Therefore both possible candidates win in one of the two languages each:

Abbildung in dieser Leseprobe nicht enthalten

3. Case studies

3.1 "The phonology of -ize derivative": Ingo Plag

In Ingo Plag's cahpter 6.2. "The phonology of -ize deriatives" of Morphological Productivity. Structural Constraints in English Derivation he observes with the help of OT how the suffix -ize is applied to nouns to create new verbs. After a global introduction to the theory he takes a close look at Raffelsiefen's constraint-hierarchy and her results before finally constructing and explaining his own hierarchy. Leaving his comments on Raffelsiefen's work aside his own hierarchy will be of major interest here. He summarizes his constraint-hierarchy with their effects on the output as follows:

FT-BIN, TROCH, NONFINALITY, IDENT-HEAD » R-ALIGN HEAD [sic!] Effect: noun-like stress, no stress-shift (r á ndomize, h σ spitalize) MAX-C » R-ALIGN-HEAD » MAX-V

Effect: deletion of stem-final vowels, but not stem-final consonants (m é morize vs. H σ spitalize)

*CLASH-HEAD » R-ALIGN-HEAD » MAX-V

Effect: no deletion of stem final vowels with disyllabic trochaic bases (d á ndyize)

*SCHWA-V, *CLASH-HEAD » R-ALIGN-HEAD

Effect: glottal stop insertion with disyllabic bases ending in shwa (m σ ra [?] ize) MAX-Ȉ'-V » R-ALIGN-HEAD

Effect: no deletion of stressed vowels (r á di ò ize)

*CLASH-HEAD » STRESS-IDENT

Effect: distressing of base-final syllables (á n [ø] dize)

*SCHWA-V, STRESS-IDENT & DEP » *CLASH-HEAD » STRESS-IDENT, DEP Effect: non-uniformity of distressing (á n [ø] dize vs. Gh é tt ò ize) IDENT-HEAD » *CLASH-HEAD

Effect: no stress-shift with iambic bases (ban á l ì ze)

*CLASH-HEAD » OCP-ONSET = MAX-C » R-ALIGN-HEAD

Effect: OCP-effects and their non-uniformity (f é minize vs. str ´ ychnin ì ze) (Plag 1999: 187)

Although he has developed a graphical arrangement of these constraints in form of a tree-diagram, it does not provide an individual guide as to which 'route' the candidates are to take. For instance, as one can see by the constraints listed above, many hierarchical orders move to *CLASH-HEAD proceeding to the next constraint.

[...]