Twenty-First Century Psycholinguistics: Four Cornerstones

Free download. Book file PDF easily for everyone and every device. You can download and read online Twenty-First Century Psycholinguistics: Four Cornerstones file PDF Book only if you are registered here. And also you can download or read online all Book PDF file that related with Twenty-First Century Psycholinguistics: Four Cornerstones book. Happy reading Twenty-First Century Psycholinguistics: Four Cornerstones Bookeveryone. Download file Free Book PDF Twenty-First Century Psycholinguistics: Four Cornerstones at Complete PDF Library. This Book have some digital formats such us :paperbook, ebook, kindle, epub, fb2 and another formats. Here is The CompletePDF Book Library. It's free to register here to get Book file PDF Twenty-First Century Psycholinguistics: Four Cornerstones Pocket Guide.

It includes information about theory and methods as well as background on the field, on modern methods such as brain scanning, and on future directions in research. Users without a subscription are not able to see the full content on this page.

Contact us:

Please subscribe or login. Oxford Bibliographies Online is available by subscription and perpetual access to institutions. For more information or to contact an Oxford Sales Representative click here. Not a member? Sign up for My OBO. Already a member? Publications Pages Publications Pages.

Subscriber sign in. Forgot password? Don't have an account? Sign in via your Institution. Sign in with your library card. Related Articles about About Related Articles close popup. Psycholinguistics by Shelia M. Kennison , Rachel H. Introduction Psycholinguistics is the field of study in which researchers investigate the psychological processes involved in the use of language, including language comprehension, language production, and first and second language acquisition.


General Overviews Since its rise in the s, the study of psycholinguistics, despite being a subspecialty within the broader field of cognitive science, has involved a wide range of topics. How to Subscribe Oxford Bibliographies Online is available by subscription and perpetual access to institutions. Jump to Other Articles:. Women, Psychology of Work Well-Being. Oxford University Press. All rights reserved. In Special issue: Psycholinguistics as a case of cross-disciplinary research. Guest edited by William Bechtel.

Synthese DOI: Altmann, Gerry. The language machine: Psycholinguistics in review. British Journal of Psychology This article serves as a fitting historical piece for anyone interested in the background to psycholinguistics as well as the history behind major contributors to the field in its early days. Blumenthal, Arthur.

The emergence of psycholinguistics. This article discusses the back-and-forth difficulties experienced by the separate fields of psychology and linguistics as they formed the field of psycholinguistics. Clifton, Charles, Jr. Sentence and text comprehension: Roles of linguistic structure. Annual Review of Psychology — This review contains many citations regarding linguistic structure, such as works on the role of prosody, semantics, and memory.

Cutler, Anne, ed. Twenty-first century psycholinguistics: Four cornerstones. Mahwah, NJ: Lawrence Erlbaum. This book provides information about the interdisciplinary development of psycholinguistics as well as sections on biology associated with psycholinguistic ability and methodology. Fundamentals of psycholinguistics. Malden, MA: Wiley-Blackwell. This text provides an overview of psycholinguistics, including first and second language acquisition. It also provides information about the biological bases of language processing.

Challenges for 21st century linguistics

Harley, Trevor A. Although numerous tests have been devised for soliciting the critical intuitions, instrument PPs remain difficult to categorize e. Following linguistic tradition, ungrammatical sentences will be preceded by an asterisk in the following examples. However, like typical adjuncts, they allow pro-form replacement John will eat the cake with a fork and Mary will do so with a spoon. One might conclude that there is no sharp distinction between arguments and adjuncts —such a possibility is discussed below. For example, Steedman's 77 Categorial Grammar assumes that "all PPs, even those that would normally be thought of as modifiers rather than sub- categorized, are in fact arguments.

In summary, there are at least two controversies within formal linguistic theory that psycholinguistic data may speak to. The first is whether there is in fact any distinction between the lexical specification of arguments and adjuncts. Secondly, if such a distinction is to be main- tained, psycholinguistic data may help resolve the debate over problem- atic cases such as instrument PP's.

Syntactic structures are built incrementally during sentence comprehen- sion, and new constituents are attached to the developing structure via competition between lexical alternatives. Constraints from any level of representation can influence competition, but the relative frequency of lexical forms is especially powerful: Just as more frequent meanings of semantically ambiguous words are accessed more easily than less frequent meanings, so more frequent syntactic forms are more easily accessed.

Thus, lexically specified structures exhibit lexical frequency effects. Consider the following example.

Twenty-first Century Psycholinguistics: Four Cornerstones - Google книги

Both delegate and suggest can head either a dative or a simple transitive structure, but the dative form is relatively more frequent for delegate. This is illustrated in Figure 2. Lexicalized versions of both structures are accessed by recognition of either verb, but weighted by frequency. Thus, to the students would be attached more easily following delegate than suggest, because the PP is specified by the dominant structure in the former case. The subcategorization preference effects reported by Stowe, Tanenhaus, and Carlson , Trueswell , and others provide evidence for this type of lexical frequency effect.

If argument slots are represented in the lexical entries of their heads, but adjunct slots are not, only arguments could be attached using the tree-adjoining mechanism summarized above and illustrated in Figure 2. Given the structures in Figure 2. Under this type of a two-mechanism account, lexical. The upper portion of the figure illustrates a lexical unification mechanism for argument attachment. The lower portion of the figure illustrates that same mechanism cannot be used for adjunct attachment if adjunct slots are not lexically specified by the head i. This Argument Structure Hypothesis is relevant to linguistic theory because it makes the prediction that argument phrases will be processed differently than adjunct phrases.

One challenge for testing this Argument Structure Hypothesis is distinguishing lexical frequency effects from plausibility effects and other factors that might influence our dependent measure, such as reading time on the phrase of interest. In the current section, I will illustrate this problem using a finding from Spivey-Knowlton and Sedivy , and suggest the solution offered by Boland, Lewis, and Blodgett Because action verbs are more likely to be modified by a PP headed by with see Table 2.

However, an alternative account is based upon a difference in local plausibility. On reading The mechanic changed a tire with The plausibility account is consistent with the Argument Structure Hypothesis, because both NP-attached and VP-attached adjunct options could be generated by rule rather than lexically specified. Unfortunately, no dependent measure provides a pure index of lexical frequency effects, uncontaminated by other variables.

However, it may be possible to distinguish lexical frequency effects from other influences on reading time. TABLE 2. This approach maintains a distinction between the generation of syntactic structure and selection processes that operate when multiple grammatical structures are possible. The distinction between the generation of syntactic structure and syntactic ambiguity resolution is explicit in some parsing theories e.

Frequency effects in syntactically unambiguous sentences provide the strongest evidence for the lexicalization of syntactic knowledge, because the effects must arise during lexical access and generation of syntactic structure. To illustrate, compare the noun-verb homographs in 2 : play occurs most often as a verb, while duck occurs most often as a noun. Boland and Corley each found that encountering a lexically ambiguous word in its less frequent syntactic form increased reading time compared to encountering its more frequent form.

Thus in 2 , reading times for duck are faster than for play, because the syntactic context is consistent with the dominant form of duck, but the subordinate form of play. Importantly, reading times in syntactically ambiguous sentences i. Boland and Blodgett found additional evidence that lexical frequency constraints and discourse constraints impact sentence comprehension in different ways. In an eye tracking experiment that used unambiguous.

The architecture of the parsing model is given on the left, the representations generated by the model are provided for an unambiguous example center and an ambiguous example right. In the absence of any alternative structures, discourse congruency had no impact on syntactic analysis. Rather, the second pass effects were presumed to reflect an anomaly within the discourse level representation.

Together, these findings suggest that lexical frequency affects lexical access and syntactic generation, but discourse congruency does not. Instead, discourse congruency plays a role in ambiguity resolution syntactic selection and relatively late discourse coherence processes. Under this approach, we can minimize the influence of factors that affect selection processes by using maximally unambiguous contexts, as in 3.

Doing so should increase the role of lexical frequency relative to plausibility in syntactic processing. Thus PP adjuncts like those used in Spivey-Knowlton and Sedivy , should no longer be influenced by verb type. In contrast, lexical frequency effects should be found for VP- attached PP arguments, so we added some dative sentences in order to demonstrate a true lexical frequency effect. Consider the context The mechanic changed the tires If the next word is with, English syntax allows for two possible adjunct attachments of the PP headed by that preposition: modification of the VP or modification of the direct object NP.

If these attachment alternatives are both rule- generated and thus equally available, one must use pragmatic knowledge or some other mechanism to select the most likely attachment site. In this case, structural factors such as recency and complexity make VP attachment more accessible and would likely swamp the selection process. Even in such relatively unambiguous structures, if the verb takes an argument PP, we should see effects of lexical frequency. A PP like to the children should be read more quickly following delegate compared to suggest because the dative syntactic structure shown on the left half of Figure 2.

The lexical frequency effect arises because access to the competing argument structures is weighted by relative frequency. These predictions were tested by Boland et al. Importantly, lexical frequency effects were obtained in the unambiguous structures for dative argument PP's: Self-paced reading times for the PP following a high- frequency dative like delegate were faster than after a low-frequency dative like suggest.

The eye-tracking data were particularly informative because they offered additional details about the relative timing of the argument and adjunct effects. The lexical frequency effects for the dative arguments were apparent in the early eye movement measures such as the first-fixation and the first-pass reading times over the PP. In contrast, the attachment site by verb class interaction, replicating the Spivey- Knowlton and Sedivy finding on locally ambiguous adjuncts, was found only in the total time on the PP region. There were no first-pass effects during the PP for the adjunct stimuli in either the locally ambiguous condition 1 or the maximally unambiguous condition 3.

In summary, Boland et al. This suggests that arguments are attached using detailed lexical information that is weighted by frequency, while adjuncts are attached using more global syntactic knowledge. The contrast between the argument and adjunct stimuli observed by Boland et al. This finding ought to be considered, along with traditional linguistic tests, when evaluating the argument status of instrument PP's. Implicit arguments in listening paradigms Frequency effects are one consequence of the lexical specification of arguments. Such effects are an empirical marker of argument status that can be investigated in psycholinguistic experiments.

Another conse- quence of the Argument Structure Hypothesis is that recognition of a lexical head provides access to the thematic roles associated with frequently occurring arguments. This prediction is supported by reading experiments that have demonstrated that verbs implicitly introduce their arguments into the discourse, without the arguments being explicitly mentioned e. Converging evidence can be found within a listening paradigm. We tend to look at things as they are mentioned, if the mentioned items are in the visual environment.

This phenomenon extends to items that have not yet been explicitly mentioned. For example, Sussman, Campana, Tanenhaus, and Carlson found that listeners made an eye movement to an appropriate instrument a pencil on hearing Poke the dolphin but not Touch the dolphin. Even though no instrument was mentioned, listeners used their knowledge about the two verbs to decide whether to manipulate the dolphin with their finger or a pencil in a real- world environment. Listeners were also sensitive to contextual factors that altered verb meaning. For example, they looked at a potato peeler when asked to Peel the potato, but not when asked to Peel the banana.

Sussman et al. Normal conversation involves a great deal of strategic guessing about the speaker's intent, so this is not a problem if the goal is to study the output of the complete comprehension process. However, if there are some partially or fully automatized aspects of syntactic and semantic processing, the directed action paradigm is not ideal for studying the representations that result from those automatized processes alone.

For example, one might question whether the recognition of poke obligatorily introduces an instrument into the discourse. Encouragingly, there is converging evidence for the automatic activation of thematic role information from passive listening tasks. Altmann and Kamide found more and faster looks to the cake following eat compared to move, beginning prior to the onset of cake.

Altmann and Kamide concluded that the verb's thematic roles were used to pro-actively restrict the domain of subsequent reference. Even in a passive listening task, it is difficult to identify the cause of the anticipatory fixations, because both linguistic and general world knowledge could have contributed to the effect. An important question is whether the discourse elements that can be introduced by a verb are limited to members of its thematic grids.

In other words, do a verb's arguments hold a privileged status or are all related words and concepts accessed in the same way? If it is solely the verb's argument structure that is driving eye movements, then listeners should not look at a bed upon hearing The girl slept because bed cannot be an argument of slept. Alternatively, listeners might look at a bed because beds are part of a prototypical sleeping event and are thus conceptually related to sleep. Furthermore, discussions about sleep often include mention of a bed, so linguistic co-occurrence frequency is high and the co-occurrence of sleeping and beds in participants' actual experience is likely to be extremely high.

One might consider an account of Altmann and Kamide's effect that is akin to semantic priming—a conceptual, essentially intra-lexical, process. However in more recent work, Kamide, Altmann, and Haywood found that combinatory semantics rather than simple lexical relationships influenced eye movements. For example, when viewing a carnival scene, listeners looked at a motorcycle upon hearing The man rode Thus, knowledge higher-level than simple lexical associations must have influenced gaze.

Was it argument structure or real world knowledge, or both? I investigated this question using a passive listening paradigm Boland, Across three experiments, effects of both argument status and real world knowledge were found. The goal was to distinguish between anticipatory looks to target pictures representing potential arguments and anticipatory looks to pictures that were strongly associated with the verb, but did not have the linguistic status of argument.

Acceptability ratings insured that sentences with typical targets were judged to be more acceptable than sentences with atypical targets. Furthermore, typical targets were more likely to co- occur with their verbs. Importantly, there was no evidence that typical recipients had a higher co-occurrence frequency than typical locations — if anything, the opposite was true. The primary finding in Experiment 1 was that dative verbs prompted more anticipatory looks to potential recipients than transitive action verbs prompted to potential instruments or intransitive verbs prompted to potential locations.

The relevant time window for examining these anticipatory looks was from verb onset to the onset of the PP that mentioned the target. The argument status effect began about ms after verb onset, suggesting that it occurred soon after lexical access of the verb. Interestingly, listeners were just as likely to fixate the atypical recipient toddler as they were to fixate the typical recipient teenager. In both the typical and atypical conditions, the potential referent met the lexical constraints on recipients for that particular verb.

If verbs specify the syntactic and semantic constraints on their arguments, recognizing a verb would make available knowledge about that verb's arguments, and entities that satisfy the syntactic and semantic constraints could be identified in the current discourse model or the situational context. In the first experiment, the argument structure of the dative verbs introduced an abstract recipient, but only one potential referent was pictured—the same one that was ultimately mentioned.

This experiment produced clear typicality effects, suggesting that when more than one potential referent is pictured, real world knowledge is used to focus attention on the most appropriate referent.

  • Twenty-First Century Psycholinguistics : Four Cornerstones -
  • Contemporary Perspectives on Natural Law;
  • MIMO Wireless Communications: From Real-World Propagation to Space-Time Code Design?
  • Definitive XSL-FO?

This account is consistent with prior evidence that pragmatic constraints influence ambiguity resolution, but not the generation of linguistic structure Boland, The argument status effect was replicated in a third experiment, in which a single animate NP and the corresponding picture served as an argument in the dative condition 6a and as an adjunct in the action verb condition 6b.

No instrument was mentioned in the critical trials, though a prototypical instrument for the action verb was always pictured, and in filler trials, pictured instruments were mentioned. There were more looks to the target picture when it was an argument recipient than when it was an adjunct benefactor, instrument during the interval to ms after the onset of the verb.

Co- occurrence frequency does not provide an alternative explanation. There were no reliable differences in co-occurrence frequency among the dative-recipient, action-benefactor, and action-instrument pairs. Together, these findings demonstrate that linguistic constraints play a privileged role in guiding visual attention in this passive listening paradigm.

Furthermore, these argument status effects suggest an important distinction between adjuncts and arguments in terms of how verbs introduce entities into the discourse. A verb implies its arguments, but not adjuncts, before they are explicitly mentioned. In addition, these results suggest another experimental test of argument status. In summary, the results from reading and listening paradigms converge to support the view that arguments and adjuncts have a different status in parsing. In the listening experiments, verbs implicitly introduced their arguments, but not adjuncts, and visual attention was drawn to likely referents of those arguments.

This is to be expected if only arguments are represented in the lexical entries of their heads. These findings are relevant to two major issues in parsing theory: How is syntactic knowledge stored and accessed? What are the mechanisms for attaching new constituents to the developing syntactic representation? Are these results also relevant to formal syntactic theory?

In this case and many others, psychologists who study sentence comprehension rely on linguistic theory for insight into the nature of our mental representations and vocabulary for describing them. However, the insights don't flow as freely in the other direction. Formal linguists don't often try to account for phenomena that psychologists discover about the mental representations involved in language processing. This may be because formal linguistics has little to gain from cognitive psychology under weak transparency assumptions.

But what about the exceptional cases? I have suggested that assertions about lexical specification within syntactic and morphological theory are in fact claims about how linguistic knowledge is stored, accessed, or acquired. As such, some of these assertions may be tested more definitively with experimental methods than with linguistic intuitions.

If the experimental data are clear, and if linguistic theory makes note of them, the experimental paradigms reviewed above may be able to resolve some of the debates about the distinction between arguments and adjuncts. In contrast, psycholinguistic research cannot resolve purely structural debates about the geometry of the phrase structure tree or the nature of a derivation within syntactic theory, because these constructs do not generate straightforward predictions about processing.

An example is the extensive line of experimental research e. The Trace Reactivation Hypothe- sis is usually stated as the prediction that an antecedent will be reactivated at its trace site. For example, in the sentence, In which box did you put the cake? The PP is said to have moved out of its canonical position, leaving behind a trace, which is represented by the underline.

The fundamental problem is that recognition and coindexing of the long distance dependency is a complicated processing issue that has not been carefully addressed in the trace reactivation literature. Researchers generally assume that coindexing—and therefore reactivation—occurs at the linear position of the trace. In our example, coindexing would take place after the offset of cake, so priming of box would be predicted at that point in the sentence.

Unfortunately, because traces are phonologically null, the listener or reader does not perceive a trace directly. Therefore, recognition and coindexing of the purported trace need not coincide with its linear position in a sentence. If they are psychologically real, traces must be postulated on the basis of cues that may or may not be adjacent to the trace site. For example, recognition of put could initiate projection of a VP with slots for a direct object and a locative PP.

If so, a trace could immediately be posited and coindexed with in which box, leading to priming of box at put. Depending upon the strategy used by the parser, other alternatives are also possible. In short, a syntactic theory of traces makes no predictions about when or if priming should occur unless it is wedded to well-articulated processing theory that specifies how and when traces are postulated, as well as how previously encountered phrases will persist or decay in working mem- ory.

Because these processing questions are themselves controversial, it is difficult to see how psycholinguistic research can resolve syntactic debates over traces. Even if some psycholinguistic data do influence a few corners of formal linguistic theory, we are not on the brink of a revolution in linguistic methodology. Psycholinguistic data —and data from cognitive neuroscience for that matter—will always play a secondary role in formal linguistic theory, adjudicating between linguistic theories that are equally elegant and account for the traditional data linguistic intuitions from a variety of languages equally well.

This is as it should be, under the assumptions of weak transparency. Linguistic theory does not attempt to describe neural or behavioral patterns, but rather the knowledge state that gives rise to those neural and behavioral patterns. Linguistic assertions about lexical specification are unusual in that these assertions concern the linking assumptions between formal theories of linguistic knowledge and processing theories of how linguistic knowledge is stored, accessed, and used.

For the most part, the linking assumptions among the knowledge state, the behavior, and the neural activity remain underspecified in both linguistic and psycholinguistic theories. This chapter benefited from numerous conversations with my University of Michigan colleagues prior to the conference. I would also like to thank the attendees at the Four Corners workshop on the interface between psychology and linguistics for many fruitful discussions. Altmann, G. Incremental interpretation at verbs: Restricting the domain of subsequent reference.

Cognition, 73, Interaction with context during human sentence processing. Cognition, 30, Berwick, R. The grammatical basis of linguistic performance: Language use and acquisition. Boland, J. The relationship between syntactic and semantic processes in sentence comprehension. Language and Cognitive Processes, 12, Visual arguments. To appear in Cognition, pending revisions. Understanding the constraints on syntactic generation: Lexical bias and discourse congruency effects on eye movements.

Journal of Memory and Language, 45, Distinguishing generation and selection of modifier attachments: Implications for lexicalist parsing and competition models. Manuscript submitted for publication. Review of Skinner's verbal behavior. Language, 35, Language and mind. New York: Harcourt, Brace, and Jovanovich. The minimalist program. Clahsen, H. Antecedent priming at trace positions: Evidence from German scrambling.

Journal of Psycholinguistic Research, 28, Corley, S. A statistical model of human lexical category disambiguation. Doctoral dissertation, University of Edinburgh. Edelman, S. How seriously should we take Minimalist syntax? Trends in Cognitive Sciences, 7, Grimshaw, J. Argument structure. Jackendoff, R. Foundations of language: Brain, meaning, grammar, evolution. Kamide, Y. Prediction and thematic information in incremental sentence processing: Evidence from anticipatory eye-movements.

Journal of Memory and Language, 49, Larson, R. On the double object construction. Linguistic Inquiry, 19, MacDonald, M. C, Pearlmutter, N. The lexical nature of syntactic ambiguity resolution. Mauner, G. Implicit arguments in sentence processing. Journal of Memory and Language, 34, Finitary models of language users.

Luce, R. Galanter Eds. New York: Wiley. Phillips, C. Order and structure. Linguistics and empirical evidence. Pinker, S. On language and connectionism: Analysis of a parallel distributed processing model of language acquisition. Cognition, 28, Rumelhart, D. Parallel distributed processing: Explorations in the microstructure of cognition. I, Foundations. Schutze, C.

Argumenthood and English prepositional phrase attachment. Journal of Memory and Language, 40, Skinner, B. Verbal behavior. New York: Appleton-Century-Crofts. Spivey-Knowlton, M. Resolving attachment ambiguities with multiple constraints. Cognition, 55, Syntactic ambiguity resolution in discourse: Modeling effects of referential context and lexical frequency within an integration-competition network. Steedman, M. Surface structure and interpretation. Stowe, L. Filling gaps on-line: Use of lexical and semantic information in sentence processing. Language and Speech, 34, Sussman, R.

Verb-based access to instrument roles: Evidence from eye movements. Trueswell, J. The role of lexical frequency in syntactic ambiguity resolution. Journal of Memory and Language, 35, Getting Sound Structures. Acquisition data have never prominently figured in linguistics despite the fact that the ultimate goal of linguistics is to understand what consti- tutes knowledge of language and how this knowledge is acquired. In his recent GLOW lecture, Chomsky stressed that in order to answer these questions it is important to gain insight into how a lexicon is built up during acquisition, and what lexical representations look like.

Here, I focus on representations of sound structures in the lexicon. So far, child data have always been considered as external evidence in linguistics, just like results from psycholinguistics have been Boland, this volume. Consequently, I know of no linguistic theory that has under- gone changes based on new results in research on child phonology. At best, child language data have been used as additional evidence for particular linguistic claims.

On the other hand, linguists take learnability very seriously, as any grammar, being it syntactic or phonological, should in principle be learnable on the basis of the primary language data that a child encounters. Studies in acquisition of phonology have mostly been concerned with why children produce words differently from adults. In fact, phono- logical representations have not been central in generative studies of phonological acquisition, even though they have been so prominent in 'adult' phonology, which aimed at providing the most elegant and economic descriptions of lexical representations, using universal phono- logical units only.

Moreover, information that can be supplied by rules is often assumed to be absent in the representations, leading to abstract underspecified phonological representations e. In psychology, infant speech perception studies, have recently given rise to a new view of language acquisition Kuhl, In the seventies and eighties researchers argued that children pick up salient parts of the input first e.

Changes in the lexical repre- sentations served an efficient organization of the lexicon. Today, most psychologists studying language acquisition assume that children have detailed phonetic representations from a very early stage. By simply listening to language infants acquire sophisticated information about what sounds and sound patterns occur in the language, which of those patterns are frequent and which are likely to co-occur. Moreover, they do so long before they utter their first word. If infants already know so much about their language before speaking it, any discrepancies between this knowledge of the sound patterns of words and the actual way in which they produce them must lie in production skills, either due to underdeveloped or untrained articulatory routines or by processing limitations, such as limited memory, weak entrenchment of forms, etc.

Although it is often claimed that production plays a role in development, its role in understanding language acquisition is fairly limited in most current views. The usual assumption is that perception precedes production and production hardly influences perception of mental representations.

Thus, both psychologists and linguists are concerned with the manner in which the sound structure of words is represented in the mental lexicon. Psychologists are interested in the units that are used for speech recognition and speech processing. Linguists, in particular phonologists, are concerned with the form of phonological representations, the units of which they are composed and the phonological processes that relate different appearances of words. Their ultimate goal is to define linguistic competence. Psycholinguists strive for understanding how knowledge of language is used in perception and production, i.

Yet, a number of great linguists have explicitly assumed that linguistic competence should have psychological reality, meaning that it should be reflected in performance. Halle , for example, states the following: "Speakers find it difficult to memorize the stress contours of each word separately, but find it easy to compute the stress contours by means of rules".

Hence, stress need not be part of each individual lexical item, but can be computed by stress rules. Similarly, Kaye has argued that processing considerations are the ultimate cause of phonological phenomena. Many phonological processes, such as vowel harmony, have a delimitative function and help detecting morpheme boundaries.

Lahiri and Reetz go even further by arguing that speech perception highly benefits from abstract phonological representations in the mental lexicon: The less information is stored in the lexicon the less the change that it mismatches with the incoming acoustic signal or, put differently, the more likely a word is being recognized.

However, despite the fact that phonologists have often mostly implicitly assumed psychological reality of phonological rules and representations, seldom have they gone out of their way to prove this in a way that has convinced psycholinguists. On the other hand, psychologists have largely ignored results from theoretical linguistics. With the appearance of Optimality Theory OT in the early nineties the focus in generative phonology has been shifted from underlying representations input to surface representations output. In OT phonology is viewed as a set of universal innate constraints that link input and output structures.

Each language has ordered these constraints in a language particular way. The constraint order evaluates all possible output forms of words and selects one as the most optimal candidate. An important difference with the 'traditional' view of phonology is the focus on output structures. This is also reflected in the principle of 'Richness of the Base', which states that there are no restrictions on input representations. The constraint hierarchy contains different types of constraints. In the simplest model of OT the constraint set is composed of markedness constraints, which ban marked structure, previously captured by rules or morpheme structure conditions, and faithfulness constraints, which formally link input and output structures and demand that output structures equal input structures.

Thus, to evaluate output structures, the input representation still is important, as it determines the satisfaction of faithfulness constraints. In psychology, too, the current view seems to be that representations contain detailed information, based on a growing number of studies showing that both adult and child listeners use detailed and context- sensitive information of spoken words. However, so far, the existence of abstract phonological representations has not been completely denied, as listeners are able to recognize words despite of considerable variation across speakers and environments.

In this chapter I want to show that children's production forms provide evidence for the claim that children 1 build up abstract phono- logical representations of words and 2 make generalizations over their own productive lexicon resulting in phonological constraints which are part of children's developing phonological system.

This view has serious consequences for OT, at least as a theory of acquisition. On the one hand markedness constraints emerge in the course of development instead of being innately present. On the other hand, representations also develop; hence, the interpretation of faithfulness constraints is not stable either. Moreover, I assume that there is a single abstract phonological represen- tation mediating between word recognition and production.