Spoken Word Production and Its Breakdown In Aphasia

Lyndsey Nickels

• Publisher: Psychology Press
• Year: 2014
• ISBN:
  9781317775096
• Edition: 1
• Topic: Psychology
• Subtopic: History & Theory in Psychology
• Index: Psychology

This volume combines in-depth reviews of models of spoken word production and cognitive neuropsychological disorders of spoken word production. The first section provides a detailed discussion of the development and structure of current models of language production using data form "normal" subjects. It is these models that form the basis of the study and therefore the text attempts to explain their processing mechanisms and assumptions clearly. The evidence used for the development of these models is described including experimental studies and observation of patterns in naturally occurring speech errors.; The second section focuses on studies of aphasic naming disorders and discusses these disorders in terms of the model described in the first section. The emphasis is on single case studies. These are reviewed in three chapters examining semantic errors and disorders, the range of symptoms attributed to disorders of lexical retriveal and deficits of phonological encoding. The text assumes that the reader has some familiarity with the linguistic and psychological terminology relevant to these areas and therefore is most suited to the graduate student or researcher/lecturer.

PART ONE

Models of spoken word production

Introduction

When we speak we have the problem of “putting our ideas into words”. Models have been developed that aim to explain what is necessarily involved in this process, from the initial conceptualisation of an idea to be expressed to the final articulation of the word, or combination of words, that expresses that idea. The level of conceptualisation of an idea will not be addressed here but rather the emphasis shall be on putting that idea into words. In fact, as has already been made clear, this book explicitly avoids discussion of more than production of a single word at a time. However, as will become evident, even with these restrictions the field is complex and fraught with theoretical conflict!

Thus we begin, in Chapter 1, with the nature of the representation of words in the mental lexicon. This leads on to a discussion of how these representations are retrieved and the problems that are associated with ensuring the correct lexical item is selected. A number of different models are outlined in the course of the chapter, with a primary division being drawn between those models that propose discrete nonoverlapping stages of processing (e.g. Levelt, 1989; Morton, 1970) and those that propose a continuous flow of activation from one level of processing to another (e.g. Dell, 1986).

The phonological information that is retrieved from the stored lexical representation cannot be articulated directly, and in Chapters 2 and 3 we discuss the processes that convert this information into a form that can be used by the articulators for speech. In Chapter 2 we begin with two “slot-and-filler” models, which describe the procedures by which the segments of a Word are inserted into a syllabic frame. The motivation for this mechanism comes from speech error data (both naturally occurring and elicited slips of the tongue) and these are discussed in detail as this is a key concept for the field. Chapter 3 continues by examining further models and integrating this stage of word production with models of lexical retrieval. The chapter concludes with a discussion of theories of motor programming and possible mechanisms for speech monitoring.

CHAPTER ONE

Lexical representation and access

OVERVIEW

Although most authors would agree that a store of knowledge about lexical items is required, there is less agreement regarding the nature, organisation, and mechanism for retrieval of representations. This chapter aims to review some of the literature addressing the issue of lexical access, the first stage in the production of a word.

The chapter begins with a discussion of what size of units are stored in the lexicon (WORDS OR MORPHEMES?). Does it contain an entry for every word we know or are the words broken down into their component parts, which may then be combined as we use them?

This leads on to a discussion of how these stored entries are retrieved (LEXEMES AND LEMMAS). One of the main issues here is whether there is a one- or two-stage process in lexical retrieval. We discuss in some detail an influential experiment, which argues for two stages in lexical retrieval (KEMPEN AND HUIJBERS), and then contrast this with a no less influential model, which incorporates only a single stage of retrieval (MORTON’s LOGOGEN MODEL).

Discussion of the logogen model leads us to a problem that is common to many models of word production (whether they incorporate one or two stages of lexical retrieval)—that of convergence on the right word. In particular, we outline Levelt’s detailed analysis of how to avoid selecting the superordinate for production when aiming for a particular exemplar of that category (e.g. to say “dog” rather than “animal”) (THE HYPERNYM PROBLEM). Although we will not discuss different semantic theories in detail, we briefly describe some semantic theories that may overcome the hypernym problem (NONDECOMPOSITIONAL THEORIES). As their title suggests, these do not break down concepts into sets of semantic features or primitives but rather represent the concept described by the word as an indivisible unit.

The chapter then moves on to examine the extensive role that speech errors have played in the development of theories of word production (SPEECH ERRORS AS EVIDENCE). In “TIP-OF-THE-TONGUE” STATES words that we are certain are within our vocabulary seem temporarily unavailable. The (partial) information that people can recall about these words has been widely investigated. We review the literature from this area in some detail and examine the theoretical conclusions that have been reached. In particular, it seems likely that there are two stages of lexical retrieval (one semantic and one phonological) and that there is also a stage at which partial phonological information is available. The logogen model cannot explain these findings, and therefore in the next section we look in more detail at models that incorporate both these stages (TWO STAGES OF LEXICAL RETRIEVAL). The first model we discuss (BUTTERWORTH’s SEMANTIC LEXICON MODEL), is similar to the logogen model except that it incorporates an explicit level of lexical-semantic processing (the semantic lexicon). This model also claims to be able to account for the partial (phonological) information that is often available to speakers when in a tip-of-the-tongue state.

However, having two distinct levels of lexical representation does not necessarily mean that these representations have to be accessed strictly sequentially. We next examine a class of models that incorporate both lexical-semantic and phonological representations but do not require processing to be complete at one level before the next level is activated (INTERACTIVE ACTIVATION: LEVELS BUT NOT STAGES). In these models, as soon as there is activation at the first, semantic, level it will be “passed on” to the second, phonological, level. The activation then cycles between the two levels until the point is reached at which the most highly activated item is selected for word production. Recently there has been a great deal of interest in determining how activation is occurring in the production system, to distinguish between the different theories—whether it is in sequential and nonoverlapping stages or continuously flowing through the model. We study in some detail the experimental evidence that has been used to address this issue (THE TIME COURSE OF LEXICAL ACCESS: PAUSES; UTTERANCE INITIATION TIMES; LEXICAL DECISION LATENCY) and the responses of the proponents of different architectures to the challenges these data present (GLOBAL MODULARITY WITH LOCAL INTERACTION).

One issue that is rarely made explicit in the psycholinguistic literature but becomes of great importance when investigating acquired language disorders is whether the same phonological lexicon is used for both comprehension and production (ONE LEXICON OR TWO?). This is a fiercely debated topic and we briefly review some of the experimental evidence.

The ease (speed and accuracy) of word retrieval can be affected by a number of factors associated with words. The final section of this chapter summarises the literature regarding the effects of five of these and discusses the possible loci of their effects within models of spoken word production (VARIABLES AFFECTING LEXICAL ACCESS: IMAGEABILITY AND CONCRETENESS; FREQUENCY AND FAMILIARITY; AGE OF ACQUISITION).

The relationship between the sound of a word and its meaning is largely arbitrary. This relationship has to be learned (stored) for each word so that it may be retrieved for use at a later date. This store, the “mental lexicon”, is the mediator between conceptualisation and preparation for articulation (phonological encoding) and consists of at least four different types of knowledge about lexical items. First, there is a specification of the item’s meaning, that is, a specification of the conceptual conditions that must be fulfilled for the item to be selected. Second, details of syntactic properties including word class (noun, verb, etc.) and requirements for syntactic arguments (subcategorisation rules) are given. Third, each item has a morphological specification, and fourth, a specification of phonological form. The latter includes details of syllable and accent (stress) structure, and the composition in terms of phonological segments (e.g. phonemes). There may also be additional properties associated with each item including those aspects that refer to pragmatic or stylistic variables.

WORDS OR MORPHEMES?

How are the words we know stored in the lexicon? Are they stored as complete indivisible units or as parts (morphemes; stems and affixes) that can be combined. For instance, is “disagreement” one entry or three (dis-; agree; -ment)?

The “Full Listing Hypothesis” states that every word form the speaker knows is explicitly listed in the lexicon, so “disagreement", “disagreements” and “disagreeing” will all be separately listed. This contrasts with the approach traditionally preferred by many linguists, whereby regularities in lexical representation are stated by rules, with only those items not accounted for by rule being listed in the lexicon (see Butterworth, 1983, for a review of this debate).

The Full Listing Hypothesis seems implausible when languages such as Turkish and Finnish are considered. In these agglutinative languages, words consist of strings of morphemes, that is, a root plus affixes that add to the meaning of a word, e.g. in Turkish (Hankamer, 1989) “indirilemiyebilecekler”, meaning “[they] will be able to resist being brought down”, with the root “in” (descend). Hankamer (1989) has calculated that a single Turkish noun can appear in more than 4 million different forms, although most of these forms will never be used by an individual Turkish speaker. Rather than suggest that each form is stored in the lexicon, it is more likely that the Turkish speaker has stored representations of all stems and affixes and a set of procedures or rules to produce new words as output depending on the conceptual input. Hankamer suggests that an account incorporating elements from both theoretical extremes seems the most promising. Thus, in addition to the listing of stems and affixes and the productive use of rules to generate new word forms, some (frequently used) morphologically complex forms are listed.

In contrast to Turkish, English has relatively little spontaneous generation of “new” words, with most inflected and derived forms being those used before. However, this does not mean that the English speaker cannot produce new words (or understand them), merely that it is rare to do so. For example, “unreformattable” is a perfectly possible word but not one that most speakers will have uttered (or heard) before.

What then of inflections (e.g. -s; -ing; -ed, etc.) and derivations (un-; -ment) in English? Do they differ in the way they are stored? Garrett (1980) argues for lexical representation of stems alone, with affixes being added by other processes after access of the stem. In contrast, Butterworth (1983) claims that the evidence for this theory is inconclusive and by no means decisive in rejecting the Full Listing Hypothesis. Butterworth notes, however, that supporters of the Full Listing Hypothesis are not forced to deny that speakers have knowledge of affixing rules but merely that they are not routinely used. Other authors have variants of these theories of the nature of lexical storage. For example, Aitchison (1994) concludes that inflectional suffixes (-s; -ing, etc.) are added as needed with stems alone stored but that derivational forms (un-; -ment, etc.) are stored as whole-word units. Stemberger and MacWhinney (1986) also suggest that high frequency inflected forms are stored whereas low frequency forms are computed on-line. Marslen-Wilson, Tyler, Waksler, and Older (1994) draw a distinction (at least for comprehension) between the representation of semantically opaque derived forms (where the meaning cannot be determined from the component parts, e.g. release, department), which are stored as undecomposed wholes, and semantically transparent derived forms (where their meaning can be determined from the stem and affixes, e.g. happiness, rebuild), which are stored in a decomposed form as stems and affixes.

LEXEMES AND LEMMAS

Assuming that (at least for English) most word forms are stored in the mental lexicon, while remaining neutral as to whether inflected and derived forms are fully listed or constructed by rule, how are these word forms retrieved? In particular, is all the stored information retrieved simultaneously or is retrieval broken down into two or more stages?

Kempen and Huijbers

Kempen and Huijbers (1983) introduced terminology distinguishing two parts of a lexical entry that are considered to be independent. Thus, the lemma is the part of a lexical entry that refers to syntactic (grammatical) and semantic (meaning) properties, relevant for grammatical encoding (sentence planning) and utterance planning. This contrasts with the lexeme, which consists of the morphological and phonological properties of an entry essential for phonological encoding (assembling the sounds of a word for production) but largely irrelevant for grammatical encoding. Each lemma maps on to its corresponding lexeme, that is it “points” to where in the store of lexemes (the phonological lexicon or form lexicon) the morpho-phonological information for that item is stored.

Kempen and Huijbers (1983) argue for this distinction on the basis of the results of a number of experiments that measured the length of time it took subjects to describe a picture after seeing it. Dutch subjects were required to describe pictures of actions in terms of either a single word (subject noun, e.g. girl; verb, e.g. kick), sequences of words (noun + verb, e.g. girl-kick; verb + noun, e.g. kick-girl) or sentences (subject + verb, e.g. the girl kicks; or verb + subject, which is a possible construction in Dutch). In their first experiment they found that subjects were faster to describe the pictures using a subject noun alone (S) than using a verb alone (V) or SV and VS sentences, whose averages were close together.

To account for these data, Kempen and Huijbers propose that the naming processes for actor (the person doing the action, who in these sentences is the subject) and action start simultaneously and proceed largely in parallel. “Lexicalisation” (retrieval of word information corresponding to a concept) is postponed until the “to-be-expressed” concept has been fully identified. In other words, in the SV condition the actor’s name is retrieved only when both the actor and action have been recognised. (Identification of an action takes longer than identification of an actor as evidenced by the latencies in the S condition being faster than latencies in the V condition.) However, data from a second experiment necessitated the addition of the “double look-up hypothesis” to the model. Thus, initially an “abstract” (nonphonological) lexical item is retrieved. Only when this abstract lexical item (lemma) corresponding to both actor and action has been retrieved will the phonological form (lexeme) be retrieved for the first item to be produced, followed by that for the second.

In this second experiment, Kempen and Huijbers studied the extent to which practice effects found over successive trial blocks were disrupted by the introduction of a new set of verbs for the same actions. Thus, half-way through the experiment subjects were instructed to change the verb they were using for a particular action. For example, if they had been using the verb “meppen” (slap) they would be required to change to describing the same action using “slaan” (beat). They proposed that if the slower speed of SV responses compared to S responses is caused by lexical processing of the verb (i.e. finding the appropriate lemma to correspond to the picture), then introducing a new verb will slow the responses even more (by eliminating the effects of practice o...

Table of contents

• Cover
• Title Page
• Copyright Page
• Table of Contents
• Acknowledgements
• Preface
• Part One 1: Models of spoken word production
• Part Two: The breakdown of spoken word production in aphasia
• Postscript
• References
• Author index
• Subject index