IOS Press Speci city in rehabilitation of word production ...€¦ · Galley Proof 2/11/2011;16:34 File: ben358.tex;BOKCTP/wyn p. 1 Behavioural Neurology 25 (2012) 1J29 1 DOI 10.3233/BEN-2012-0358

Galley Proof 2/11/2011; 16:34 File: ben358.tex; BOKCTP/wyn p. 1

Behavioural Neurology 25 (2012) 1�–29 1DOI 10.3233/BEN-2012-0358IOS Press

Specicity in rehabilitation of wordproduction: A meta-analysis and a case study

Charlotte Jacquemota,b,!, Emmanuel Dupouxa,b, Laura Robothama,b andAnne-Catherine Bachoud-Levib,c,d,e

aLaboratoire de Sciences Cognitives et Psycholinguistique, ENS-EHESS-CNRS, Paris, FrancebDepartement d�’Etudes Cognitives, Ecole Normale Superieure, Paris FrancecINSERM U955 team 1 �“Neuropsychologie Interventionnelle�”, Institut Mondor de Recherche Biomedicale Creteil,FrancedAP/HP, Unite de Neuropsychologie, Service de Neurologie, H�ˆopital Henri Mondor, Creteil, FranceeUniversite Paris 12, Faculte de Medecine, Creteil, France

Abstract. Speech production impairment is a frequent decit observed in aphasic patients and rehabilitation programs have beenextensively developed. Nevertheless, there is still no agreement on the type of rehabilitation that yields the most successfuloutcomes. Here, we ran a detailed meta-analysis of 39 studies of word production rehabilitation involving 124 patients. We useda model-driven approach for analyzing each rehabilitation task by identifying which levels of our model each task tapped into.We found that (1) all rehabilitation tasks are not equally efcient and the most efcient ones involved the activation of the twolevels of the word production system: the phonological output lexicon and the phonological output, and (2) the activation of thespeech perception system as it occurs in many tasks used in rehabilitation is not successful in rehabilitating word production. Inthis meta-analysis, the effect of the activation of the phonological output lexicon and the phonological output cannot be assessedseparately. We further conducted a rehabilitation study with DPI, a patient who suffers from a damage of the phonological outputlexicon. Our results conrm that rehabilitation is more efcient, in terms of time and performance, when specically addressingthe impaired level of word production.

Keywords: Word production, aphasia, rehabilitation, model-driven approach, meta-analysis, case study

1. Introduction

Production difculties are reported to be the mostfrustrating and distressing aspects of aphasia and oneof the rst causes of social drop out and depressionamong patients [1]. After an initial phase of spon-taneous recovery, the speech performance of aphasicpatients usually stabilizes, but sometimes considerablybelow the unimpaired level. Restoring language pro-duction through rehabilitation is therefore of consid-erable clinical interest. Yet, there is still considerable

!Address for Correspondence: Charlotte Jacquemot, LSCP, EcoleNormaleSuperieure, 29 rue d�’Ulm, 75005Paris, France. Tel.: +33 0144 32 26 23; Fax: +33 01 44 32 26 30; E-mail: [email protected].

disagreement regarding the most efcient rehabilitationstrategy, and it is still unclear whether it should targetthe language faculty broadly or focus on the specicdisorder at hand. In fact, several authors have claimedafter years of research that no clear connection betweenthe type of impairment and the most effective thera-py can be established (see for instance [2�–4]). In thispaper, we argue that such questions can be clariedby relying on a single language processing model usedsimultaneously for the diagnosis of the decit and forthe design of rehabilitation tasks. We illustrate thispoint with a statistical meta-analysis of 39 rehabilita-tion studies, and a case study of the rehabilitation of anaphasic patient.

One of the obstacles in comparing the outcome ofrehabilitation studies lies in their different mix of train-

ISSN 0953-4180/12/$27.50 2012 �– IOS Press and the authors. All rights reserved


2 C. Jacquemot et al. / Word production rehabilitation

Fig. 1. Model of language processing including speech input and output systems and written input and output systems.

ing tasks. One can distinguish two broad types of ther-apies: so-called semantic and phonological therapies.Semantic therapy includes tasks such as word-picturematching, naming on denition, and questions on se-mantic features of a picture. Phonological therapy in-cludes tasks such as word reading, word repetition, pic-ture naming with phonological cueing, and tasks thatrequire monitoring word features (number of syllables,rst phoneme or rst syllable) on an implicitly namedpicture or from an auditory input. These two therapieswere proposed to correspond to two broad classes ofproduction decits: respectively, semantic therapy forpatients who make semantic errors in selecting wordsduring production, and phonological therapy for pa-tients who select the right words, but make phonologi-cal errors during planning [5�–12]. However, such a se-lective effect of rehabilitation strategy is challenged bythe observation of rather indiscriminate effects of ther-apy as a function of decit [13�–17]. Marshall and col-laborators [13] observed a positive effect of semanticrehabilitation in patient with no semantic decit whileNickels and Best [16] described a patient suffering fromsemantic impairment who failed to benet from thera-py involving semantic tasks. Additionally, phonologi-cal tasks have also been successful in improving wordproduction abilities in patients with semantic impair-ment [10,11,14,15,18�–20]. Finally, the absence of re-habilitation effects whatever the strategy, whether it besemantic or phonologic, is reported in some but veryfew studies [4,11,21�–23].

A similarly confusing pattern of results is found re-garding the existence of generalization of the rehabil-itation results beyond the set of specically traineditems. Some studies nd generalization [12,14�–16,18,24�–26], whereas some studies failed to observe it [5,7,10,13�–15,17,25�–27]. Hillis [7] proposed that unlikephonological therapy, semantic therapy could inducegeneralization for untrained items that are semanticallyrelated to the trained items. However, generalizationhas also been obtained with phonological therapy [12].

Overall, from these studies, it appears that phono-logical and semantic therapies are both potentially use-ful for aphasic rehabilitation, but that relationshipsbetween impairment and rehabilitation tasks are notstraightforward [6]. One reason could be that thereis no relationship to be found: language rehabilitationrequires the activation of the whole language network,without regard for the particular decit. If this is true,there is no point in trying to devise therapies tailored toa particular language decit. However, before endors-ing such a strong conclusion, one needs to inspect inclose detail the possibility that the tasks used in the dif-ferent kinds of therapies are not very pure and involveoverlapping processing levels. We illustrate this pointwith the model presented in Fig. 1. This is a fairly stan-dard functional architecture for language processing atthe level of individual words adapted from Miceli andcolleagues [28], (see also, [29,30]). It distinguishedtwo input modalities: one for spoken words and one


C. Jacquemot et al. / Word production rehabilitation 3

for written words. Similarly, it recognized two out-put modalities. Within each modality there is a formallevel (orthography or phonology) and a lexical level.The lexical levels are connected to a common a-modalconceptual level, and connected to each other. The for-mal levels are connected to perceptual inputs or motoroutputs and to their respective lexical levels. Finally,some formal levels are connected to each other acrossmodalities (i.e. input phonology to output phonology).Within such a model, any psycholinguistic task can beanalyzed as involving a particular pattern of levels andconnections: for instance, a picture naming task in-volves the following pathway: Picture analysis �– Con-ceptual level �– Output Phonological Lexicon �– OutputPhonology.

With such a model in mind, a quick inspection ofsome of the tasks used in rehabilitation studies revealsthat in many if not in all �‘semantic tasks�’, the form ofthe word is provided as a spoken or written word, oras a feedback response of the experimenter [5,7,13,14,16]. This implies that these semantic tasks are activat-ing phonological and/or orthographic levels. Vice ver-sa, many �‘phonological tasks�’ involve real words pre-sented auditorily or orthographically, which thereforeactivate lexical and probably also semantic levels. AsByng [31] explicitly states, �“most therapies describeddo not represent a single therapeutic process, even ifthey involve a single task. A single task might requirea number of complex processes, but it may not be clearwhich of these processes was the most important foraffecting change�” (p.10). Since the rehabilitation tasksdo not isolate distinct levels of processing, it is there-fore not so surprising that they do not appear to be veryspecic.

In this paper, we propose to revisit the issue of reha-bilitation specicity by relying on the processing mod-el presented in Fig. 1. We rst reanalyzed the tasksused in 39 rehabilitation studies of patients with namingdecits. This analysis was done by identifying for eachtask the levels of the model stimulated and connectionsthey activated. We then derived a measure of therapeu-tic sensitivity for each of these levels and connections,and showed that rehabilitation is much more specicthan the authors of the studies themselves concluded.Secondly, we constructed a rehabilitation protocol with�‘pure tasks�’, that is, tasks that are much more specicto a particular subpart of the language processing sys-tem than usually used, and evaluate the effectiveness offour pure rehabilitation procedures on a single anomicpatient. We conclude by discussing the value of using aprocessing model for devising rehabilitation strategies.

2. Meta-analysis of rehabilitation of naming

2.1. Search strategy and selection criteria

First, we searched articles in the PubMed and Webof Science databases between 1990 and 2009 (Novem-ber) using the following key words: speech produc-tion, therapy, anomia, rehabilitation, and aphasic pa-tient. We expanded this base by looking up the refer-ences cited in the identied articles, arriving at a totalof 85 studies. Second, we selected the papers that metthe following ve criteria:

(1) specic focus on word production rehabilitation,(2) detailed description of material and of the reha-

bilitation tasks allowing us to infer the specicprocesses that were trained,

(3) availability of individual results when the reha-bilitation program was proposed to a group ofpatients,

(4) report of performance baseline before rehabili-tation,

(5) report of the rehabilitation outcome, at least qua-litatively.

According to these inclusion criteria, 39 speech pro-duction rehabilitation studies were selected to be in-cluded in our meta-analysis (see Table 1). The rea-sons of article exclusion were the following: not aword rehabilitation study (N = 31), no individual data(N = 6), no details concerning the therapy used (N =1), patient with high variability of performance duringpretest (N = 2), no access to the articles (N = 6).Two separate investigators did the reviewing phase andthe inclusion phase. Overall, these studies representa total of 124 individual patient rehabilitation results(some studies have several patients, and some patientsare subject to several rehabilitation procedures). De-spite this selection, two difculties remain which ham-per the use of the standard rules of meta-analysis in-volving computing an overall meta-effect size. First,many papers did not report effect sizes, and some ofthem were even lacking numerical results. Second, be-cause the performance under scrutiny is word produc-tion (which implies a difcult to assess �“chance�” levelfor performance), and that the lists of words used fortest vary widely across studies, it is very difcult tocompare the relative strength of rehabilitation betweenstudies (control subjects being at ceiling).

Thus, we resorted to the use of a binary criterionof rehabilitation success (success vs. failure), and thesignal detection theory in order to tease apart the un-



Table 4Word production rehabilitation studies included in the meta-analysis

Paper Patient Treatment task Rehabilitationoutcome

[72] BR Picture naming +RepetitionPicture naming with phonological cues (letter, syllable, word) and distractors wordsElicit rhymes, initial and nal phonemes, number of syllables on picturePicture naming with semantic cuesNaming to auditory denition

JD Repetition +Picture naming with phonological cues (letter, syllable, word) and distractors wordsElicit rhymes, initial and nal phonemes, number of syllables on picturePicture naming with semantic cuesNaming to auditory denitionRepetition

KS Repetition +Picture naming with phonological cues (letter, syllable, word) and distractors wordsElicit rhymes, initial and nal phonemes, number of syllables on picturePicture naming with semantic cuesNaming to auditory denitionRepetition

BB Picture naming +RepetitionPicture naming with phonological cues (letter, syllable, word) and distractors wordsElicit rhymes, initial and nal phonemes, number of syllables on picturePicture naming with semantic cuesNaming to auditory denition

HF Picture naming +RepetitionPicture naming with phonological cues (letter, syllable, word) and distractors wordsElicit rhymes, initial and nal phonemes, number of syllables on picturePicture naming with semantic cuesNaming to auditory denition

WM Picture naming +RepetitionPicture naming with phonological cues (letter, syllable, word) and distractors wordsElicit rhymes, initial and nal phonemes, number of syllables on picturePicture naming with semantic cuesNaming to auditory denition

[42] RF Picture naming with orthographic cues +Repetition "Reading aloud +

MR Picture naming with orthographic cues +Repetition "Reading aloud +

[45] JOW Written naming "Delayed copyAuditory word picture matching +Delayed copying

[73] 02-030 Picture naming +Auditory perception of the correct picture nameRepetition

03-031 Picture naming +Auditory perception of the correct picture nameRepetition




Table 1, continued



03-004 Picture naming -Auditory perception of the correct picture nameRepetition

[47] MB Reading aloud +RepetitionPicture naming with semantic cuesPicture naming with phonological cuesAuditory word picture matching

NO Reading aloud +RepetitionPicture naming with semantic cuesPicture naming with phonological cuesAuditory word picture matching

PU Reading aloud +RepetitionPicture naming with semantic cuesPicture naming with phonological cuesAuditory word picture matching

SK Reading aloud +RepetitionPicture naming with semantic cuesPicture naming with phonological cuesAuditory word picture matching

[48] GM Picture naming with phonological cues +Picture naming with orthographic cuesAuditory sentence completion oral and written response

AS Picture naming with phonological cues +Picture naming with orthographic cuesAuditory sentence completion oral and written response

BM Picture naming with phonological cues +Picture naming with orthographic cuesAuditory sentence completion oral and written response

EL Picture naming with phonological cues +Picture naming with orthographic cuesAuditory sentence completion oral and written response

EG Picture naming with phonological cues +Picture naming with orthographic cuesAuditory sentence completion oral and written response

RH Picture naming with phonological cues +Picture naming with orthographic cuesAuditory sentence completion oral and written response

[74] TV Picture naming +RepetitionPicture naming with phonological cues (letter, syllable, word) and distractors wordsElicit rhymes, initial and nal phonemes, number of syllables on picturePicture naming with semantic cuesNaming to auditory denition

[18] MB Word length judgment on auditory words +Spoken-to-written phoneme matchingPoint to initial/nal letter of auditory wordsPoint to written word that rhymes with auditory wordAuditory word picture matching +Auditory word picture matching + picture naming taskPicture naming with error judgment



Table 1, continued


[75] NS Picture naming with phonological cues (letter, syllable, word) and distractors words "RepetitionPicture naming with semantic cues "Repetition

EG Picture naming with phonological cues (letter, syllable, word) and distractors words +RepetitionPicture naming with semantic cues +Repetition

CH Picture naming with phonological cues (letter, syllable, word) and distractors words +RepetitionPicture naming with semantic cues +Repetition

[5] SS Naming to auditory denition +RepetitionPicture naming with semantic cues +Elicit category, visual features, size from pictureRepetitionRepetition with picture "

MR Naming to auditory denition +RepetitionPicture naming with visual semantic cues +Elicit category, visual features, size from pictureRepetitionRepetition with picture "

[49] HM Picture naming with orthographic cues letter, syllable, word and distractors words +RepetitionPicture naming with phonological cues letter, syllable, word and distractors words

PH Picture naming with orthographic cues letter, syllable, word and distractors words +RepetitionPicture naming with phonological cues letter, syllable, word and distractors words

SC Picture naming with orthographic cues letter, syllable, word and distractors words "RepetitionPicture naming with phonological cues letter, syllable, word and distractors words

DC Picture naming with orthographic cues letter, syllable, word and distractors words +RepetitionPicture naming with phonological cues letter, syllable, word and distractors words

OL Picture naming with orthographic cues letter, syllable, word and distractors words +RepetitionPicture naming with phonological cues letter, syllable, word and distractors words

IK Picture naming with orthographic cues letter, syllable, word and distractors words +RepetitionPicture naming with phonological cues letter, syllable, word and distractors words

NK Picture naming with orthographic cues letter, syllable, word and distractors words +RepetitionPicture naming with phonological cues letter, syllable, word and distractors words

KR Picture naming with orthographic cues letter, syllable, word and distractors words +RepetitionPicture naming with phonological cues letter, syllable, word and distractors words

[50] MH Picture naming with orthographic cues letter, syllable, word and distractors words +RepetitionPicture naming with phonological cues letter, syllable, word and distractors words

HP Picture naming with orthographic cues letter, syllable, word and distractors words +RepetitionPicture naming with phonological cues letter, syllable, word and distractors words

[25] JJ Auditory word picture matching +Reading aloud "



Table 1, continued


HW Auditory word picture matching "Reading aloud +Picture naming +Auditory sentence completion oral responsePicture naming with phonological cuesRepetition

KE Picture naming +Auditory sentence completion oral responsePicture naming with phonological cuesRepetition

[7] HG Written picture naming with anagram, orthographic cues +Writing to dictationDelayed copyPicture naming +Written picture namingReading aloudQuestion about function of object picturesAuditory sentence completion oral responsePicture naming with phonological cues

[40];[3,45,76]

JOW Point to initial letter of auditory wordsPicture naming with phonological cues

+

Picture namingRepetition

[41] PD Repetition "Reading aloud +Auditory sentence completion oral response "

[77] PG written naming +reading

AH written naming +reading

TM written naming +reading

[8] LR Auditory word picture matching +Listening oral denitionPicture naming with orthographic cuesReading aloud

[43] P1 Elicit rhymes, initial and nal phonemes, number of syllables on picture +Picture namingRepetition with picture

P2 Elicit rhymes, initial and nal phonemes, number of syllables on picture +Picture namingRepetition with picture







Table 1, continued


P7 Elicit rhymes, initial and nal phonemes, number of syllables on picture "Picture namingRepetition with picture




[21] BB Written generation of semantic cues of picture +NamingReading aloud

BG Written generation of semantic cues of picture +NamingReading aloud

SB Written generation of semantic cues of picture +NamingReading aloud

[13] RS Written word picture matching with reading aloud +

IS Written word picture matching with reading aloud +

RS Written word picture matching with reading aloud +

[78] CM Association of picture with semantically related written word "Written word picture matching

[10] RBO Reading aloud +Picture namingRepetition +Picture naming

GMA Reading aloud with picture +Picture namingReading aloud +Picture namingRepetition +Picture naming with phonologic cues

[79] HR Written naming with anagram, orthographic cues +Copy

[14] TRC Written word picture matching +Auditory word picture matching "

PA Auditory word picture matching "Reading aloud +Picture naming with orthographic cue and reading aloud +

[80] WT Repetition +Repetition with iconic gesture +

[51] KB Simultaneous auditory and written word-picture matching +Auditory word picture matchingPicture naming with phonological or orthographic cueCopy, anagramWritten picture naming

JI Simultaneous auditory and written word-picture matching +Auditory word picture matchingPicture naming with phonological or orthographic cueCopy, anagramWritten picture naming



Table 1, continued


RI Simultaneous auditory and written word-picture matching +Auditory word picture matchingPicture naming with phonological or orthographic cueCopy, anagramWritten picture naming

[20] WR Picture naming +Elicit initial phoneme and rhyme judgement on auditory wordRepetitionPicture naming +Semantic judgement on auditory wordRepetitionRepetition with picture +Auditory word picture matching +

[15,26] CG Picture naming with phonological cues +Repetition

RJ Picture naming with phonological cues +Repetition

[81] YK Picture naming +RepetitionReading

[82] LV Picture naming +Written word picture matching with reading aloudRepetitionAuditory to written word matching

JP Picture naming +Written word picture matching with reading aloudRepetitionAuditory to written word matching

[83] P1 Elicit category, visual features, size from picture +Picture namingPicture naming with semantic cues

P2 Elicit category, visual features, size from picture +Picture namingPicture naming with semantic cues

P3 Elicit category, visual features, size from picture +Picture namingPicture naming with semantic cues

[12] GF Syllable judgement on picture name supplied with the picture +Initial phoneme judgement/pointing on picture name supplied with the pictureNaming

[84] JB picture naming and iconic gesture of the target +picture naming +iconic gesture +

[52] AB Picture naming with syllable counting and rst phoneme, identication of picture naming errors +Written anagram and readingPicture namingRepetitionPicture naming with syllables counting and iconic gesture of the target +RepetitionPicture naming with syllables counting and iconic gesture of the target +Written anagram and readingRepetition

[85] Anomicpatient

Auditory sentence completion oral responseAuditory sentence completion with semantic cues oral response

+

Auditory sentence completion with phonological cues oral responseRepetition



derlying effects of rehabilitation from the intrinsic bi-ases in relation to the decision threshold for success orfailure left to the experimenter (or the reviewers of thestudy). Although this constitutes a departure from thestandard meta-analysis, it shares with it the same gen-eral methodology and aims. As can be seen in Table 1,the number of case studies in which the rehabilitationbeneted to the patients vastly outnumbers case studieswhere rehabilitation did not (N = 105 vs. N = 19).This suggests that studies mostly selected a successfulrehabilitation, or that any kind of rehabilitation works(placebo effect), or a reporting bias (that is, unsuccess-ful studies tend to have more difculty reaching thepublication stage). In addition, some treatments aremore frequently used for rehabilitating patients, as forinstance naming tasks compared to writing tasks, andtherefore some levels are more frequently involved dur-ing rehabilitation. In order to rule out any frequencyeffect of the task, as well as the imbalance between suc-cess and failure rates in our meta-analysis, we carriedout a signal detection theoretic analysis aimed at dis-tinguishing rehabilitation sensitivity (i.e., the selectiveeffect of rehabilitation) from response bias.

To this end, we used a strategy similar to the onesused by Indefrey and Levelt in his meta-analysis ofthe functional brain imagery of speech production [32].First, we assigned to each rehabilitation task a partic-ular �“activation pattern�” deriving from the frameworkof the model in Fig. 1. The activation pattern consti-tutes all components of the model involved in each re-habilitation task. Second, we correlated the activationpattern across studies with rehabilitation outcome foreach patient. This allowed us to derive, using signaldetection theory, a measure of rehabilitation sensitivityshowing component by component its role in rehabili-tation outcome.

2.2. Method

2.2.1. Task modelingIn order to perform a given task, (picture naming,

word repetition, etc.) certain components of the modelpresented in Fig. 1 are essential, and others not. We ap-plied the standard task decomposition method of cog-nitive neuropsychology [33�–35] for each task to iden-tify its functional components according to the mod-el of Fig. 1: we determined a �“pattern of activation�”by assigning for each task the value �“0�” or �“1�” to theprocessing levels and their connections: an activationvalue of 1 was assigned to all levels and connections

involved in the task, and an activation value of zerootherwise (see Table 2).

The pattern of activation for a given task was deter-mined in order to reect the main pathway or pathwaysthat subjects use to achieve correct performance in thattask. More specically, we rst identied the mostdirect processing route for each task, which yielded aset of processing levels and connections. In addition,we added less direct routes, if existing psycholinguis-tic evidence documented that they were generally be-ing used for this task. For instance, the most directpathway for a picture naming task is that from pictureanalysis to conceptual level, phonological output lexi-con and phonological output. Of course, it is theoreti-cally possible to name a picture using a more compli-cated route, for instance by accessing the orthographiclexicon and then orthography to phonology, however,since it is not the most straightforward route, and thispathway has not been documented by psycholinguis-tic investigations, this was not considered. Thereforethe activation pattern for picture naming was set to in-clude all of the said levels as well as their connectionsto each other in the model. For the word repetitiontask, however, we modeled three pathways, the input-to-output phonology pathway, the input lexicon-to out-put lexicon pathway, and the pathways going throughthe conceptual level. This is because there is evidencethat these three pathways are used by participants [29,36�–38]. Two investigators (CJ and ED) independentlycomputed the pattern of activation for each task and athird investigator (ACBL) checked each pattern.

2.2.2. Data analysisAs shown in Table 1, we recoded each rehabilitation

case study in terms of success (+) or failure (!), basedon the data available in the original papers. Successwas attributed to studies in which speech productionperformance improved after treatment, and failure at-tributed to studies in which speech production did notimproved after the treatment,according to the statisticalanalyses, when available, or to the author�’s conclusionswhen not. We then conducted a signal detection theoryanalysis for each component of the processing model,by considering the signal to be the activation of thatcomponent in the rehabilitation task, and the responsebeing whether the rehabilitation was reported to be asuccess or a failure. For each rehabilitation case study,if the processing component had an activation of 1 forone of the rehabilitation tasks, we scored a HIT if therehabilitation was successful and a MISS otherwise. Ifthe processing component was not activated in any of



Tabl

e2

Patte

rnof

activ

atio

nfo

rth

edi

ffer

ent

task

sus

edin

wor

dpr

oduc

tion

reha

bilit

atio

nPA

:Pi

ctur

eA

naly

sis,

C:C

once

pt;

PI:Ph

onol

ogic

alIn

put;

PIL:Ph

onol

ogic

alIn

put

Lex

icon

;PO

L:

Phon

olog

ical

Out

putLex

icon

;PO

:Pho

nolo

gica

lO

utpu

t;O

I:O

rtho

grap

hic

Inpu

t;O

IL:O

rtho

grap

hic

Inpu

tLex

icon

;O

OL:O

rtho

grap

hic

Out

putLex

icon

;O

O:O

rtho

grap

hic

Out

put,

OO

:O

rtho

grap

hic

Out

put

PA-

PI-

PIL-

C-

POL-

OI-

OIL

-C

-O

OL-

OI-

OIL

-O

IL-

OI-

PI-

PI-

PIL-

PIL-

PAC

OO

LO

OPO

LPO

PIPI

LO

IO

ILC

PIL

CPO

LPO

OIL

CO

OL

OO

POPO

LO

OL

OO

POO

OPO

LO

OL

Ass

ocia

tion

ofpi

ctur

ew

ithse

man

tical

lyre

late

dw

rit-

ten

wor

d

10

00

01

10

00

00

00

00

01

10

00

00

01

1

Aud

itory

perc

eptio

nof

the

corr

ectp

ictu

rena

me

11

10

00

00

00

00

00

00

01

10

00

01

10

0

Aud

itory

sent

ence

com

ple-

tion

(ora

lan

dw

ritte

nre

s-po

nse)

01

11

10

01

10

00

00

00

00

11

11

11

10

0

Aud

itory

sent

ence

com

ple-

tion

(ora

lres

pons

e)0

11

11

00

00

00

00

00

00

01

00

11

11

00

Aud

itory

sent

ence

com

pl-

etio

nw

ithph

onol

ogic

alcu

es(o

ralr

espo

nse)

01

11

10

00

00

00

01

00

00

10

01

11

10

0

Aud

itory

sent

ence

com

ple-

tion

with

sem

antic

cues

(ora

lres

pons

e)

01

11

10

00

00

00

00

00

00

10

01

11

10

0

Aud

itory

wor

dpi

ctur

em

atch

ing

11

10

00

00

00

00

00

00

01

10

00

01

10

0

Aud

itory

wor

d�–

writte

nw

ord

mat

chin

g0

11

11

11

00

11

00

10

10

01

00

11

11

11

Aud

itory

wor

dpi

ctur

em

atch

ing

+pi

ctur

ena

min

g

11

11

10

00

00

00

00

00

01

10

01

11

10

0

Cop

y0

00

00

11

11

00

11

00

00

01

11

00

00

11

Cop

y,an

agra

m0

00

00

11

11

00

11

00

00

01

11

00

00

11

Del

ayed

copy

00

00

01

11

10

01

10

00

00

11

10

00

01

1

Elic

itca

tego

ry,vi

sual

fea-

ture

s,si

zefr

ompi

ctur

e1

00

11

00

00

00

00

00

00

11

00

11

00

00

Elic

itin

itial

phon

eme

and

rhym

eju

dgem

enton

audi

-to

ryw

ord

01

10

00

00

00

00

00

00

01

10

00

01

10

0

Elic

itrh

ymes

,ini

tiala

nd-

nal

phon

emes

,nu

mbe

rof

sylla

bles

ofpi

ctur

ena

me

10

01

10

00

00

00

00

00

01

10

01

10

00

0



Tabl

e2,

cont

inue

d

PA-

PI-

PIL-

C-

POL-

OI-

OIL

-C

-O

OL-

OI-

OIL

-O

IL-

OI-

PI-

PI-

PIL-

PIL-

PAC

OO

LO

OPO

LPO

PIPI

LO

IO

ILC

PIL

CPO

LPO

OIL

CO

OL

OO

POPO

LO

OL

OO

POO

OPO

LO

OL

Initi

alph

onem

eju

dgem

ent

/poi

ntin

gon

pict

ure

nam

esu

pplie

dw

ithth

epi

ctur

e

11

10

00

00

00

00

00

00

01

10

00

01

10

0

Lis

teni

ngor

alde

niti

on0

11

00

00

00

00

00

00

00

01

00

00

11

00

Nam

ing

from

den

ition

01

11

10

00

00

00

00

00

00

10

01

11

10

0

Pict

ure

nam

ing

10

01

10

00

00

00

00

00

01

10

01

10

00

0

Pict

ure

nam

ing

with

audi

-to

ryse

man

ticcu

es1

11

11

00

00

00

00

00

00

11

00

11

11

00

Pict

ure

nam

ing

with

erro

rju

dgm

ent

11

11

10

00

00

00

00

00

01

10

01

11

10

0

Pict

ure

nam

ing

with

orth

o-gr

aphi

ccu

es1

00

11

00

00

10

00

00

00

11

00

11

00

10

Pict

ure

nam

ing

with

orth

o-gr

aphi

ccu

es(let

ter,

sylla

-bl

e,w

ord)

and

dist

ract

ors

wor

ds

10

01

11

10

01

10

00

00

01

10

01

10

01

1

Pict

ure

nam

ing

with

orth

o-gr

aphi

ccu

esan

dre

adin

gal

oud

10

01

11

00

01

10

00

00

01

10

01

10

01

1

Pict

ure

nam

ing

with

pho-

nolo

gica

lcue

s1

00

11

00

00

00

00

10

00

11

00

11

10

00

Pict

ure

nam

ing

with

pho-

nolo

gica

lcu

es(let

ter,

syl-

labl

e,w

ord)

anddi

stra

ctor

sw

ords

11

11

10

00

00

00

01

01

01

10

01

11

10

0

Pict

ure

nam

ing

with

sylla

-bl

eco

untin

gan

dr

stph

onem

e,id

enti

catio

nof

pict

ure

nam

ing

erro

rs

10

01

10

00

00

00

00

00

01

10

01

10

00

0

Pict

ure

nam

ing

with

syl-

labl

esco

untin

gan

dic

onic

gest

ure

ofth

eta

rget

10

01

10

00

00

00

00

00

01

10

01

10

00

0

Pict

ure

nam

ing

with

visu

alse

man

ticcu

es1

00

11

00

00

00

00

00

00

11

00

11

00

00

Poin

tto

initi

alle

tterof

au-

dito

ryw

ords

00

00

00

01

10

00

00

00

01

11

10

00

00

0

Poin

tto

initi

al/

nalle

tter

ofau

dito

ryw

ords

00

00

00

01

10

00

00

00

01

11

10

00

00

0



Tabl

e2,

cont

inue

d

PA-

PI-

PIL-

C-

POL-

OI-

OIL

-C

-O

OL-

OI-

OIL

-O

IL-

OI-

PI-

PI-

PIL-

PIL-

PAC

OO

LO

OPO

LPO

PIPI

LO

IO

ILC

PIL

CPO

LPO

OIL

CO

OL

OO

POPO

LO

OL

OO

POO

OPO

LO

OL

Poin

tto

writte

nw

ord

that

rhym

esw

ithau

dito

ryw

ord

00

01

01

11

11

10

00

00

01

11

11

00

01

1

Que

stio

nab

outf

unct

ion

ofob

ject

(pic

ture

s)0

11

11

00

00

00

00

00

00

01

00

11

11

00

Rea

ding

alou

d0

00

11

11

00

11

00

00

00

01

00

11

00

11

Rea

ding

alou

dw

ithpi

ctur

e1

00

11

11

00

11

00

00

00

11

00

11

00

11

Rep

etiti

on0

11

11

00

00

00

00

10

10

01

00

11

11

00

Rep

etiti

onw

ithic

onic

gest

ure

01

11

10

00

00

00

01

01

00

10

01

11

10

0

Rep

etiti

onw

ithpi

ctur

e1

11

11

00

00

00

00

10

10

11

00

11

11

00

Sem

antic

judg

emen

ton

audi

tory

wor

d0

11

00

00

00

00

00

00

00

11

00

00

11

00

Sim

ulta

neou

sau

dito

ryan

dw

ritte

nw

ord-

pict

ure

mat

-ch

ing

11

10

01

10

00

00

00

00

01

10

00

01

11

1

Spok

en-to-

writte

nph

onem

em

atch

ing

00

00

00

00

01

00

01

10

00

00

00

11

01

0

Sylla

ble

judg

emen

ton

wor

dsu

pplie

dw

ithth

epi

ctur

e

11

10

00

00

00

00

00

00

01

10

00

01

10

0

Wor

dle

ngth

judg

men

ton

audi

tory

wor

ds0

11

00

00

00

00

00

00

00

01

00

00

11

00

Writte

nan

agra

man

dre

adin

g0

00

11

11

00

11

00

00

00

01

00

11

00

11

Writte

nge

nera

tion

ofse

-m

antic

cues

ofpi

ctur

e1

00

00

00

11

00

00

00

00

11

11

00

00

00

Writte

npi

ctur

ena

min

g1

00

00

00

11

00

00

00

00

11

11

00

00

00

Writte

npi

ctur

ena

min

gw

ithan

agra

m,o

rtho

grap

h-ic

cues

10

00

01

11

10

00

10

00

01

11

10

00

01

1

Writte

nw

ord

pict

ure

mat

chin

g1

00

00

11

00

00

00

00

00

11

00

00

00

11

Writte

nw

ord

pict

ure

mat

chin

gw

ithre

adin

gal

oud

10

01

11

10

01

10

00

00

01

10

01

10

01

1

Writti

ngto

dict

atio

n0

11

00

00

11

00

00

11

11

01

11

00

11

00



Tabl

e3

Res

ults

ofth

em

eta-

anal

ysis

PA-

PI-

PIL-

C-

POL-

OI-

OIL

-C

-O

OL-

OI-

OIL

-O

IL-

OI-

PI-

PI-

PIL-

PIL-

PAC

OO

LO

OPO

LPO

PIPI

LO

IO

ILC

PIL

CPO

LPO

OIL

CO

OL

OO

POPO

LO

OL

OO

POO

OPO

LO

OL

Hit

9883

8397

9846

5237

2145

267

673

163

196

106

2120

9898

8282

5437

Mis

s12

1414

1212

44

21

21

11

100

100

1217

11

1212

1414

43

Cor

rect

Rej

ectio

n5

33

55

1313

1516

1516

1616

717

717

50

1616

55

33

1314

Fals

eA

larm

924

2410

961

5570

8662

8110

010

134

106

4410

611

186

879

925

2553

70

d�’0,

87!

0,16

!0,

160,

800,

870,

470,

590,

700,

680,

860,

810,

06!

0,03

0,22

0,00

0,73

0,68

0,65

0,87

0,87

!0,

19!

0,19

0,63

0,47

beta

0,50

1,20

1,20

0,52

0,50

0,58

0,50

0,41

0,40

0,33

0,33

0,93

1,03

0,79

1,00

0,53

0,40

0,42

0,50

0,50

1,24

1,24

0,48

0,57

mea

n0,

72!

0,10

!0,

080,

901,

250,

550,

490,

620,

621,

100,

73!

0,18

!0,

230,

420,

000,

730,

530,

501,

221,

24!

0,16

!0,

180,

580,

53

sd0,

410,

370,

410,

430,

490,

360,

370,

350,

340,

330,

300,

420,

410,

370,

350,

430,

350,

330,

470,

470,

380,

390,

360,

37

2.5%

quan

tile

!0,

11!

0,81

!0,

860,

080,

34!

0,11

!0,

16!

0,11

!0,

050,

470,

04!

1,12

!1,

07!

0,34

!0,

66!

0,11

!0,

11!

0,21

0,26

0,35

!0,

88!

0,95

!0,

11!

0,13

97.5

%qu

antil

e1,

490,

630,

711,

782,

301,

261,

191,

201,

261,

751,

230,

510,

441,

130,

721,

571,

171,

142,

212,

200,

570,

561,

251,

25

0.5%

quan

tile

!0,

48!

0,98

!0,

95!

0,20

0,03

!0,

35!

0,38

!0,

28!

0,44

0,29

!0,

28!

1,20

!1,

34!

0,49

!0,

79!

0,40

!0,

45!

0,43

0,05

0,08

!0,

95!

1,07

!0,

31!

0,29

0.05

%qu

antil

e!

0,60

!1,

07!

1,01

!0,

39!

0,17

!0,

49!

0,53

!0,

37!

0,60

0,21

!0,

40!

1,67

!1,

49!

0,77

!0,

95!

0,55

!0,

58!

0,61

!0,

16!

0,22

!1,

01!

1,16

!0,

51!

0,42

!0.

05;

!!0.

01;

!!!0.

001

!!!

!!!

!!!

!!



Fig. 2. D�’ values of the effect of rehabilitation reported on the model of language processing.

the rehabilitation tasks, we scored a FALSE ALARM ifthe rehabilitation came out positive, and a CORRECTREJECTION otherwise. We then computed d�’ and be-ta for each processing component, by using the averageHIT and FALSE ALARM rates across all of the casestudies [39]. The value of d�’ can be interpreted as abias-free measure of the sensitivity of the reeducationstrategy, i.e. a measure of rehabilitation effectiveness,and beta as a measure of bias (i.e. the bias to only reportor publish rehabilitation case studies that work, or theunspecic effect of rehabilitation). The effect of gener-alization is not included in our meta-analysis becauseof the small number of studies in which it was reported.Note that in the meta-analysis, the activation of each ofthe processing levels and connections were consideredindividually for computing the effect of therapy. Wedid not analyse the different combination of process-ing levels or connections that could jointly affect theeffectiveness of therapy. Indeed our model is alreadyquite complex (10 nodes, 17 links), and the number ofpossible simple combinations between these elementsbecomes too large (351 combinations) compared to thenumber of data points available (124 outcomes), espe-cially since none of the published data attempt to reha-bilitate the processing levels in systematic or a factorialdesign.

2.3. Results and discussion

The values of d�’ and beta are shown in Table 3. Theimbalance between the number of case studies in whichthe rehabilitation beneted to the patients and thosewith no results (N = 105 vs. N = 19) result in a largeresponse bias (from 0.33 to 1.24, with an average 0.67)which is rather homogeneous across processing com-ponents. This might be interpreted either as a genericeffect of rehabilitation (placebo effect) or, alternatively,as a reporting bias. The d�’ values, which range between!0.19 and 0.87 are projected onto the processingmodelusing gray levels in Fig. 2. Interestingly, the level thatgave the highest value of d�’ are the phonological outputlexicon (d�’ = 0.87, p < 0.01), the output phonology(d�’ = 0.87, p < 0.01) and the links between these twolevels (d�’ = 0.87, p < 0.01). Given that the conceptuallevel is always trained in all rehabilitation studies, itwas impossible to derive a d�’ score for this level. Yet,the link between the conceptual level and the phono-logical output lexicon yielded a d�’ of 0.80 (p < 0.05).Noteworthy is the role of the links between input (lex-ical and sublexical) orthography to the output (lexicaland sublexical) phonology (d�’ = 0.81, p < 0.05 andd�’ = 0.86, p < 0.001, respectively). The other levelshad a lower d�’ which did not reach signicance. In



particular, the phonological input and the phonologicalinput lexicon had a low d�’ of !0.19.

This meta-analysis shows that all the tasks are notequally efcient in producing a positive effect [5,14,25,40�–42]. Indeed, rehabilitation of the phonological out-put lexicon and of the phonological output of the wordproduction levels are more prone to successful rehabil-itation than, say, the phonological input components orthe components of the orthographicmodality. The rela-tive inefciency of the phonological input componentsis interesting to consider given the widespread use ofsyllable counting, monitoring techniques and phono-logical cueing techniques in rehabilitation studies [5,10,12,18,20,25,26,43�–52]. Surprisingly, word produc-tion can also benet from rehabilitation triggering thepathways linking orthographic input to phonologicaloutput and orthographic input lexicon to phonologi-cal output lexicon. This may result from an alterna-tive strategy induced by reading aloud words and pseu-dowords using grapheme-phoneme direct conversion;patients that are trained in reading tasks may producea word by accessing the orthographic form of the wordto be named, and producing it aloud.

This set of results shows that the use of a process-ing model can help to extract more information fromrehabilitation studies than can be done in a cumula-tive survey, even when the rehabilitation tasks have notbeen specically designed to be specic to a particularprocessing component. This conclusion is similar tothat reached by Indefrey and Levelt [53] with a similarmethodology applied to functional imagery.

Yet, we should be aware that this kind of meta-analysis is inherently limited in four ways: it is depen-dant on the quality of the patient�’s diagnostic, on theadequacy of the processing model, on the adequacy ofthe modeling of each of the tasks, and on the correla-tions within the data set itself. First, given that manyof the published rehabilitation studies we used in thismeta-analysis did not report any detailed diagnostic testto the specic locus of patient impairment, (no attemptto specify brain localization), this information couldnot be taken into account in our meta-analysis. Yet,it is likely that rehabilitation outcomes may vary withthe specic pattern of production impairment. Second,the processing model does not distinguish between dif-ferent subcomponents of lexical processing, yet thesecomponents could be differentially targeted by rehabil-itation strategies. Third, we included in the analysisonly the standard pathways for performinga given task,yet, different patients may use differing strategies byemploying alternate pathways to perform a given task.

Fourth, the published studies did not independently tar-get each possible rehabilitation foci (or combination offoci). While the signal detection analysis can take intoaccount the relative frequencies of these rehabilitationstrategies, nothing can be done when a particular focusor combination is missing. For instance, the role ofthe semantic/conceptual component cannot be evalu-ated statistically because all of the studies included atask that involved this component; hence there cannotbe any observation in the FALSE ALARMS and COR-RECT REJECTION cells. This can only be alleviatedby conducting a rehabilitation study that manipulatesthis component. Another example is given by the factthat in virtually all of the studies, the phonological out-put lexicon, the phonological output and the connectionbetween these two levels are either all simultaneouslyactive or inactive. It is therefore impossible to disen-tangle the role of these various components in the meta-analysis alone. These limitations make it impossible todissociate the effects of so-called semantic therapy andphonological therapy within the existing dataset. Thesame limit would apply if we were to study the effect ofthe combination of rehabilitation factors: not all suchcombinations have been tried in our database.

Despite these limitations, our meta-analysis resultedin a picture that was considerably clearer than what arapid browse of Table 1 suggests. In addition, it is pos-sible to address some of the outstanding issues througha targeted rehabilitation study. In the second part ofthe paperwe therefore establish a specic rehabilitationprotocol for an anomic patient in order to assess in amore controlled way the impact of the activation of thedifferent levels (conceptual knowledge, phonologicaloutput lexicon, phonological output and phonologicalinput) of word production.

3. Case study: DPI

DPI is a 68-year-old, right-handed, retired medicaldoctor. Five years before the rehabilitation program,he had a stroke leading to Wernicke�’s aphasia and aright hemiparesis. A CT scan (at admission) and a MRI(one year later) conrmed a left middle cerebral arterystroke extended to the junction territory with the pos-terior cerebral artery. The lesion encompassed the lefttemporal artery territory including the superior, medial,and inferior temporal gyri the anterior temporal lobe.A previous detailed study of the patient showed thathe suffered from a phonological output lexicon decitand had preserved conceptual abilities [54]. Briey,



DPI was awless in all non verbal tasks that tap con-ceptual processes using ve tasks constructed along thelines of Caramazza and Shelton [55] (anomaly detec-tion, picture completion, intruder detection, functionalmatching task and object color matching task). On apicture naming task, DPI producedmany errors (47.1%of errors, N = 70). Because this decit occurred in thepresence of intact conceptual knowledge, and good per-formance in word reading (10% of errors, N = 100),this suggests a decit at phonological output lexicon.Speech perception was awless. After the completionof this assessment, the patient suffered a trafc accidentresulting in a dramatic deterioration of his word pro-duction. DPI received out-clinic speech therapy twicea week. After 12 months, he still complained that hedid not recover his former speech production level. Forthis reason, we proposed to enroll him in a rehabilita-tion study. We rst assessed his performance in nam-ing production, reading and conceptual knowledge inorder to ensure that DPI still suffered from a phonolog-ical output lexicon decit. His performance in nam-ing was lower than before (68% of errors, circumlocu-tion or non-responses, N = 109) whereas his concep-tual knowledge was still intact (errorless performanceon the ve previous conceptual knowledge tasks), hisword comprehension as well (errorless performance onan auditory word �– picture matching task, N = 48) andhis word reading performance remain unchanged (10%of errors). For the duration of our study, the out-clinicspeech therapy was suspended.

3.1. Method

We designed the rehabilitation program in order toinvestigate three issues concerning the rehabilitationof an anomic patient. Firstly, we wanted to furtherthe results of the meta-analysis by determining if theactivation of the conceptual knowledge could induceany improvement in word production. Secondly, wewanted to assess whether the activation of the impairedlevel could be more efcient compared to the activa-tion of non-impaired level within the two levels of theword production system �– phonological output lexiconor phonological output. Finally, tasks involving the ac-tivation of the phonological input have been extensive-ly utilized in rehabilitation programs, but the results ofthe meta-analysis suggest that they do not impact pos-itively the word production performance. Therefore,we wanted to prospectively verify whether they couldbe successful in word production rehabilitation. Thus,DPI was submitted to 4 phases of rehabilitation, each

of them involving the activation of a specic speechlevel: conceptual knowledge, phonological output lex-icon and phonological output, and phonological input(Fig. 3). We followed the methodology proposed byNickels [4]:

Use of a pre-therapy baseline that should be assessedon more than one occasion to establish degree of spon-taneous recovery and/or variability,

Use of identical sets in terms of naming difculty forthe different rehabilitation phases,

Use of the same task for assessing patient�’s perfor-mance before and after therapy,

Use of a task that contains enough items to allowchange to be demonstrated,

Use of statistical comparisons such as McNemar�’sTest to identify whether any change in performance isgreater than might be expected by chance.

3.1.1. Baseline of performanceTo demonstrate the efcacy of a rehabilitation pro-

gram, it is important to assess whether trained itemsare better named after the rehabilitation than they werebefore, thus comparing the magnitude of change at-tributable to spontaneous recovery and the magnitudeof change attributable to therapy. Whereas comparingperformance of different groups of patients ismarred bythe great variability in individual patients�’ proles [56,57], a possible alternative to override the heterogene-ity between groups is to conduct a single case study.It consists in comparing performance of a single pa-tient after and before the rehabilitation. Performancebaseline is dened before beginning the rehabilitationprogram and constitutes the control performance as op-posed to performance after intervention. This base-line is generally evaluated by testing the patient sev-eral times on the same set of items and by quantify-ing his spontaneous performance variability [4,7,10].Here, we established the performance baseline by ask-ing DPI to name the same set of pictures three differenttimes. 156 pictures were tested and no feedback wasprovided. The three successive sessions were separatedby two weeks. These sessions were used to quantifyDPI�’s performance variability and constitute the threepretests.

3.1.2. Selection of the picture stimuliAmong the 156 pictures presented in pretests, we se-

lected the pictures never correctly named. We obtained106 pictures that were divided in four experimental setsof equal difculty. The sets were composed of 27 pic-tures (Set 1) or 26 pictures (sets 2, 3 and 4) matched for



Fig. 3. (a) Simplied model of word processing from Figure 1. This model includes word comprehension and word production. (b) Levelsactivated through each rehabilitation phase.

word frequency, word length, word gender and seman-tic categories (approximately 50% of pictures depictedartifacts, 19% vegetables, 27% animals, and 4% bodyparts).

3.1.3. Rehabilitation procedureThe four following levels of word processing were

trained during four separate phases (see Fig. 3): con-ceptual knowledge, phonological input, phonologicaloutput and phonological output lexicon. For eachphase, we used one of the 4 experimental sets of pic-tures to construct the training material. The 4 sets wererandomly assigned to the 4 rehabilitation phases. Feed-

back and correct responses were provided to the patientduring the rehabilitation phases.

Phase 1: Conceptual knowledge rehabilitationConceptual knowledge corresponds to a non-verbal

general knowledge of the world. Several non-auditorytasks involving picture comprehension and conceptualknowledge processing were proposed: anomalous/nonanomalous picture categorization, picture completion,categorical intruder detection, functional matching,correct color detection, picture categorization in sev-eral semantic categories, and knowledge assessment(see [55]). Details of the procedure are provided in



Fig. 4. (a) Example of anomalous and correct pictures, (b) example of picture completion task, (c) example of intruder detection task, (d) exampleof functional matching task.

the Supplementary Material section. In all these tasks,only yes/no responses or pointing were expected (ex-amples are provided in Fig. 4). DPI was instructed tonever name the picture, neither did the examiner.

Phase 2: Phonological input rehabilitationTo avoid access of semantic information related to

the picture names of the Set 2, pseudowords were con-structed using the speech-sound of the picture namesof the Set 2. Three lists of pseudowords were createdin order to prevent boredom and effective pseudowordlearning. The rst list was constructed by intra-wordsyllables cross-placing in multisyllabicwords (e.g. �“ar-tichaut�” /aRti"o/ (artichoke) yielded to /"otiaR/) andinter-word phonemes cross-placing in monosyllablewords. Therefore, the number of syllables was similarfor pseudoword and real word. The second list wasconstructed by syllable cross-placing inter-word andby preserving the rime of the source words (e.g; �“ar-tichaut�” and �“cerise�” /s"Riz/ (cherry) yielded to /s!"o/and /aRtiRiz/). The third list was constructed by inter-word phonemes cross-placing. This list contained ex-actly the same phonemes and preserves the distributionof mono-, bi-, tri- and quadrisyllables from the origi-nal list (e.g.: /"!soRiz/ and /tiaR/ for �“artichaud�” and�“cerise�”). The examiner pronounced the pseudoword atrandom and DPI was asked to perform several tasks in-volving speech-sound analysis such as syllable, rhyme,phoneme discrimination and detection tasks, auditory-written syllable matching task on syllable and rhyme,and pseudoword and syllable counting. Details are pro-vided in the Appendix section. Again, only yes-noresponses and nger pointing were required.

Phase 3: Phonological output rehabilitationThis phase was based on speech-sound production.

In order to reduce activation of both conceptual knowl-edge and phonological output lexicon levels we used atype of rebus. The rebus we used consisted of pictorialsymbols that represent syllabic or phonemic sounds.Each picture name of Set 3 was decomposed in sylla-bles (except for mono-syllables) and each syllable orphoneme was represented by a pictorial symbol thatwould elicit the corresponding syllable or phonemesound, for instance the sound [ki:] was elicited with apicture of a key and so on. The patient was asked toproduce the speech-sound corresponding to the picto-rial symbol (see Fig. 5). The pictorial symbols werepresented either in isolation to make the patient pro-duce monosyllables or diversely associated to make thepatient produce pseudowords of increasing complexi-ty. Details are provided in the Appendix section. Forthe monosyllabic words, the picture used was not thetarget picture but another picture that elicited the samespeech-sound. For instance, the word �“po�ˆele�” /pwal/frying pan was instantiated by a picture of a �“poil�” /pw-al/ hair. For the 54 syllables issued from the picturenames of the Set 3, we used a total of 22 pictorial sym-bols. Our aim was to induce an automatic link betweena pictorial symbol and a speech-sound and then to trig-ger the activation of phonological output via the mostdirect means, preventing the activation of the other lev-els involved in word production. Of course, the useof abstract symbol might have been methodologicallypurer, but would have been required DPI to learn toassociate a given abstract symbol and a speech-sound,



Fig. 5. Example of rebus for syllables (Rat in French is pronounced /Ra/ and step /pa/).

Fig. 6. Example of the naming task with the picture and the rebus for syllables. Vacuum cleaner is /as/ /pi/ /ra/ /t�œr/ in French.

which was deemed to be too difcult. The patient wasalways helped when he could not nd the associatedsound to a given pictorial symbol and rapidly becamefamiliar with them.

Phase 4: Phonological output lexicon rehabilitationFour different strategies could be used to activate the

phonological output lexicon: through the repetition of aperceived word, through the reading of a written word,through the production of speech output by the use ofa verbal uency task or a naming task. In this rehabil-itation phase, activation of the speech perception sys-tem and of the reading system was avoided. Moreoverwe wanted the patient to activate specic phonologicalword-forms, the ones that composed the picture namesof the Set 4. It is in theory hardly conceivable to acti-vate the phonological output lexicon where the phono-logical word-forms are retrieved without activating the

conceptual knowledge and the phonological output as-sociated to these word-forms. Therefore, we decidedto use a naming task that would activate the conceptuallevel, the phonological output lexicon and the phono-logical output and to subtract the effect of the activationof the conceptual level obtained during the Phase 1 andthe effect of the activation of the phonological outputobtained during the Phase 3. Thus, DPI was instruct-ed to name the pictures of the Set 4. Together withthe picture, the pictorial symbols corresponding to thesyllables of the picture name were presented to DPI.As in the previous phase, each syllable was symbolizedby a pictorial symbol, and the pictorial symbols werepresented below the target picture (Fig. 6).

3.1.4. Testing sessionAt the end of each rehabilitation phase, we ran a test-

ing session. We used a cross-over design for assessing



Fig. 7. Flow chart or rehabilitation program. In the pretest, grey rectangles represent the set of items correctly named and white rectanglesrepresent the four sets of unsuccessfully-name items.

DPI�’s performance. After each rehabilitation phase,DPI was asked to name the all 156 pictures without anyfeedback. This included the pictures of the set used ina given rehabilitation phase i.e. the trained stimuli, plusthe pictures of the three other experimental sets i.e. theuntrained stimuli, plus the pictures that the patient hadsuccessfully named during the pretests and that werenot included rehabilitationmaterial. These last pictureswere used to check for spontaneous variability.

The examiner who evaluated the patient�’s perfor-mance in the test sessions was blind to the assignmentof the pictures in the 4 experimental sets. Finally,patient and relatives were interviewed in an informalmanner about their subjective feelings regarding therehabilitation program.

3.1.5. Rehabilitation ow-chartBecause Hillis (1998, see also [58] has shown that

short-term but intensive training program (5 days perweek during 2 weeks) induces better improvement thanlong term non-intensive training program (2 days perweek during 5 weeks), our patient was trained dailyaccordingly. The duration of each rehabilitation phasewas of two weeks (Fig. 7). Training was provided onehour per day except on weekends. Each rehabilitationphase was followed by one to two weeks of rest.

3.2. Results

Statistical analysis was performed with the McNe-mar test [4,10,12] to compare DPI�’s results before andafter the different rehabilitation phases. Three types ofanalyses were conducted:

Performance after each pretest and after each of thefour rehabilitation phases was assessed in all picturesin order to assess whether the rehabilitation programis successful and determine which phase is the mostsuccessful;

Performance in each trained set (material set of arehabilitation phase) before and after the rehabilita-tion phase was assessed in order to investigate whethertrained items were successfully named at the end of thephase. Note that each trained set was different for eachrehabilitation phase, ie Set 1 was the trained set for thePhase 1, Set 2 was the trained set of the Phase 2 and soon.

Performance for untrained sets (that is, for eachphase, the 3 sets that were not used as trained set duringthe rehabilitation phase) before and after the rehabilita-tion phase was assessed in order to investigate whetherthe successful effect of a phase is generalized to un-trained picture names. Note that untrained sets were



Fig. 8. Naming performance for each subset after the four rehabilitation phases. Arrows indicate the trained set for each rehabilitation phase.

different for each rehabilitation phase, ie sets 2,3,4 werethe untrained sets for the Phase 1, sets 1,3,4 were theuntrained sets of the Phase 2 and so on.

3.2.1. Global performance: All pictures (N = 156)The analyses were conducted with all the 156 pic-

tures. This includes the pictures that were successful-ly named during the pretests (N = 52), the picturesthat composed the trained set for each rehabilitationphase (N = 26�–27) and the pictures that composedthe untrained set for each rehabilitation phase (N =79�–80). In the pretests, performance was remarkablystable (Pretest 3: 42/156 correct vs Pretest 1: 37/156correct; McNemar !2 = 0.9, p = 0.33). Comparisonof results before and after the rehabilitation programshows a signicant effect of the rehabilitation (Phase 4of rehabilitation: 64/156 correct vs Pretest 3: 42/156;McNemar !2 = 15.8, p < 0.001). When comparingthe results before and after each phase of rehabilita-tion, performance does not improve after the rst threephases (Phase 1: 39/156 correct, Phase 2: 37/156 cor-rect, Phase 3: 39/156 correct; McNemar !2 < 1, p >0.1) but improves after the Phase 4 (Phase 4: 64/156correct; McNemar !2 = 17.36, p < 0.001).

Performance for the successfully named pictures(N = 52)was quite accurate and constant along the dif-ferent phase (Pretest 1: 37/52 correct, Pretest 2: 39/52correct, Pretest 3: 42/52 correct, Phase 1: 35/52 cor-rect, Phase 2: 33/52 correct, Phase 3: 35/52 correct,Phase 4: 39/52 correct).

Trained set performanceThe analyses were conducted considering the 26 or

27 pictures of each trained set. Detailed results areprovided in Fig. 8. Performance remains stable afterthe rst three rehabilitation phases (McNemar !2 <

1, p > 0.1) whereas it dramatically improves after thePhase 4 (Set 4: 16/26 correct vs 2/26 correct; McNemar!2 = 11, p < 0.001).

3.2.2. Untrained sets and generalizationPerformance for the untrained sets does not improve

after the rst three phases (McNemar !2 < 1, p > 0.1)whereas it improves after the Phase 4 (Set 1 + 2 +3: 13.9% correct vs. 2.5%; McNemar !2 = 7.11 p =0.007).

A post-hoc analysis of performance in the untrainedsets after the Phase 4 was performed according to thesemantic category. Performance improved after thePhase 4 in the artifact category (see Table 4, McNemar!2 = 5; p = 0.025) but not for the other semantic cate-gories (animals, vegetables and body parts, McNemar!2 < 1, p > 0.1).

Finally, DPI reported that he was more condent inspeaking with others. This informal evaluation sug-gests a positive impact of the rehabilitation program oneveryday life.

3.3. Discussion

We conducted a short and intensive rehabilitationprogram with a stable long standing aphasic patient,DPI. His decit has been detailed in a previous studyand is localized at the phonological output lexicon [54].Four different phases of rehabilitation that successive-ly activate conceptual knowledge, phonological input,phonological output, and the phonological output lexi-con were tested.

DPI�’s global production performance improved atthe end of the rehabilitation program. This conrmsthat an intensive rehabilitation program can have a sig-nicantly positive effect on patient�’s production perfor-



Table 4Distribution of correct responses of untraineditems according to their semantic categorybefore and after the Phase 4

Before AfterArtifact (N = 40) 1 (2.5%) 6 (25%)Animal (N = 20) 0 (0%) 1 (5%)Vegetable (N = 14) 0 (0%) 2 (14%)Body-part (N = 5) 1 (20%) 2 (40%)

mance long after his brain lesion [4,7,59]. Althoughthe benets of rehabilitation appeared only at the lastphase, it is unlikely that they appear solely by practiceeffects. Practice effects would have induced a gradualincrease of patient performance along the four phasesrather than an effect restricted to the phase 4. Moreover,before being enrolled in the rehabilitation program, thepatient was seeing a speech therapist two times a weekwith no effect of his speech production performance.For these two reasons, it seems reasonable to attributethe positive outcome after the forth phase essentially tothe treatment occurring during this phase.

This experimental data conrm that all tasks are notequivalent in therapy. Specically, the Phase 1 thatinduces the activation of the conceptual knowledgedid not positively affect the patient production perfor-mance, suggesting that the activation of the conceptuallevel per se is not efcient in terms of speech reha-bilitation. Within the word production pathway, twotypes of rehabilitation were tested. The rst one se-lectively activated the phonological output procedures(phonological planning, articulation), in the absence oflexical processes. The result showed no improvementin the patient�’s performance. The second one activatedthe phonological output lexicon which is the impairedlevel in DPI. Results showed signicant improvementof DPI�’s production performance, meaning that for re-habilitating the production of lexical forms, the moreefcient way is to train the patient to produce the wordsand not only the components of these words. Finally,the focused rehabilitation of input phonology did notyield any improvementconrming that the activation ofunimpairedprocesses, namely input and output phonol-ogy did not induce any positive effect of DPI�’s speechproduction performance. These experimental resultsvalidated and claried the meta-analysis results show-ing that rehabilitation tasks that specically tap into thedamaged level, i.e., word production, could make thepatient performance in production improve. Of course,the rehabilitation phase that worked was also a phaseinvolving multiple levels, i.e., concept activation, lexi-cal activation and phonological output. Could it be thatthis aspect of the task was responsible, at least in part,

for the results? It is difcult to completely discard thishypothesis, but is it worthwhile mentioning that all ofthe rehabilitation phases included multiple levels: forinstance, the conceptual phase required both pictureanalysis and conceptual systems, as well as workingmemory and executive functions. Involving multiplelevels can therefore not be the only explanation to thepresent results.

Rehabilitation is proposed to improve the patient�’severyday life, and an important issue with rehabilita-tion programs is to assess whether performance im-provement generalizes to untrained items or whether itis restricted to trained items during the rehabilitationprogram. Here, the positive effect of Phase 4 is notrestricted to the trained set of items but spreads to en-compass some other untrained pictures. In previouspapers, generalization is inconsistently observed andthe factors that could explain this variability are stillunknown. It has been proposed that the generalizationmay be driven by semantic factors. Indeed, Miceli andcollaborators [10] described two patients with phono-logical output lexicon decit who take signicantly ad-vantage from speech rehabilitation but only to nametrained items. They proposed that generalization mayoccur when untrained items are semantically related tothe trained ones. In our study, pictures were select-ed from four semantic categories: animal, body-part,artifact and vegetable and these four categories wereequally distributed within the four sets. If any effect ofsemantic priming is expected, one could predict that itwould be greater for items belonging to a homogenouscategory, as for instance, the body-part category. Con-sidering the four categories in our study, the artifactcategory is the largest one and the most diversied. Itranges from wall to paperclip, from cigarette to skirt.Thereby, it�’s unlikely that items from artifact categorywould prime the other items of this category becauseof their semantic content distance. Contrary to thisprediction, after the Phase 4, our results show that forthe untrained sets, the performance improvement is on-ly signicant for items from the artifact category (Ta-ble 4). This suggests that generalization does not resultfrom an effect of facilitation for semantically relateditems. What other factors could be proposed to explainthe generalization of performance to untrained items?In addition to the semantic properties of the items, gen-eralization may also depend on the mechanisms thatpromote word recovery. Indeed, there are at least twomechanisms that could offer plausible explanations forthe positive outcome of a rehabilitation program. Theyare (i) the restoration of the damaged level and (ii) the



development of compensatory strategies that allow thedamaged level to be bypassed. Generalization, accord-ingly, may be differently affected. Finally, it could bethe case that generalization depends on the type of im-pairment. DPI�’s impairment involves the phonologicaloutput lexicon, but the damage he sustained could im-pact the access to this level or this level itself. Rehabil-itation may differently affect the connection betweenlevels as well as the levels themselves. Hence, factorsthat favored generalization still need to be explored.

4. General discussion

In order to address the specicity of rehabilitationstrategies for aphasic patients, our approachwas torst-ly provide a comprehensive meta-analysis of the rel-evant peer-reviewed literature on word production re-habilitation. We rst introduced a functional model oflanguage processing (Fig. 1). Based on this model, weconducted a functional reconstruction meta-analysis of39 studies involving a total of 124 rehabilitation cases.Our technique, inspired by model-based meta-analysesof fMRI data [32], consists of dissecting the tasks usedin rehabilitation in terms of the processing componentsof the model, and subsequently in reconstructing thecontribution of each component to the rehabilitationoutcome. The enterprise of relating the effect of speechrehabilitation to the activated components of spokenand written word processing requires a detailed, ex-plicit theory of the process of word processing. Themeta-analysis presented here is based on a consensu-al model of language processing [28,60]. This modelexplicates the successive computational stages of spo-ken and written word perception and spoken and writ-ten word production. The componential analysis ofthe tasks involved in the different rehabilitation studiesprovided the processes and pathways involved in eachof the tasks. The results of the meta-analysis, how-ever, do not hinge on this particular choice of theory,since differences between this model and other mod-els [30,32,61�–65] do not concern the assumed process-ing levels but the exact nature of the information owbetween them. The meta-analysis provides a clear-cutpicture, wherein only the phonological output process-ing components (phonological output lexicon and out-put phonology) signicantly contribute to rehabilitationsuccess.

Secondly, we experimentally tested this outcomewith a case rehabilitation study on a patient specicallyimpaired in the phonological lexicon. We used four sets

of �’pure�’ tasks that specically target one of the follow-ing: conceptual knowledge, input phonology, outputphonology, or the phonological output lexicon. Onlythe latter rehabilitation tasks yielded any improvementfor the patient, conrming the above conclusion regard-ing specicity of rehabilitation procedures. Further-more, this successfully trained processing componentwas able to generate signicant improvement on un-trained items, thereby displaying generalization. Thisresult supports the claim that the specic componentyielding rehabilitation success was precisely the com-ponent that was impaired in this patient (the phonologi-cal output lexicon). Yet, because the phonological out-put lexicon is deeply embedded within the global lan-guage model, it is impossible to train this componentalone, that is, without simultaneously involving othercomponents that function as inputs or outputs. Ourconclusions rest on the fact that independent trainingof these other components did not improve patient�’sperformance (see Fig. 3). However, we recognize thatthe successful rehabilitation task was also the only onethat encompassed multiple components of speech pro-duction processing. We therefore cannot discard thatthe positive effect of the phonological output lexiconactivation is not specically due to the phonologicaloutput lexicon component per se, but due to the factthat the training task involved a processing chain link-ing the phonological output lexicon to its normal inputand output in the speech production pathway.

Overall, the outcome of this study conrms that alltherapies are not equally effective and that a rehabili-tation focused on his decit could partially reactivatethe impaired process. Even if this idea motivated manypast studies [2,4,14,16,49,66], the previous literature ofrehabilitation studies failed to reach such conclusion.Furthermore, our data reinforce the importance of amodel-based approach for specifying the componentsimpaired in the patient, as well as the tasks used forrehabilitation (see [4]. As many rehabilitation studiesdid not report the specic locus of patient impairment,we were not able to take into account this informationin our meta-analysis. Thus, the next step of this type ofanalysis would be to include the locus of the decit andto correlate it with the rehabilitation outcome as we didfor the case study, using the same methodology.

From a clinical point of view, this data could helptherapists in developing rehabilitation tools for aphasicpatients. Speech production decit typically could in-volve one or several components. If our results gener-alize to these other components, a patient-specic re-habilitation strategy, focussing on the impaired compo-



nents and their connections with the rest of the systemcould prove more time-effective than generic rehabil-itation. Of course, further work is needed in order toassess whether our conclusions hold up quantitativelywith more patients, and across different kinds of apha-sic decits.

Two directions in particular would be worth explor-ing. First, as presented in the introduction, word pro-duction problems can surface with two distinct proles:(1) patients with predominantly semantic paraphasia,who can be described as having an impaired link be-tween the conceptual component and the phonologicallexicon, (2) patients with predominantly phonologicalparaphasia, who have a decit located at the phono-logical lexicon (or in the link with the phonologicalrepresentation) [61,63]. It would be very interesting touse our approach to rene the so-called �‘semantic�’ and�‘phonological�’ rehabilitation but using purer tasks, andto test whether the most successful rehabilitation strate-gies are indeed the ones linked to the impaired compo-nents. A second direction of research is inspired by thefunctional model itself. Such a model contains manyparallel and partially redundant routes. A given taskcan therefore be performed using several more or lessefcient strategies. For instance, to perform a picturenaming task, instead of using the phonological outputlexicon, one could covertly recover the spelling of theword from the orthographic output lexicon, and thenuse spelling-to-sound conversion to generate a phono-logical output. In our meta-analysis (see Fig. 2), we seethat, indeed, the spelling-to-sound route has a positiveimpact on rehabilitation outcome, suggesting that suchbackup strategies could be useful to incorporate into acomplete rehabilitation procedure. Of course, the con-tribution of these alternate strategies would have to beassessed independently of the direct rehabilitation ofthe impaired component.

Finally, most of the rehabilitation studies containedvery few details regarding the anatomy of the lesions,so it was not possible to integrate anatomical informa-tion into the meta-analysis. However, such an approachcould benet from the analysis of the brain regionsinvolved in the decit and/or the recovery. It wouldbe, in principle, possible to apply our signal detectionapproach using intact versus lesioned brain regions asinput to the analysis [67,68]. Additionally, function-al imaging data, coupled with an anatomo-functionalprocessing model could enable to study the effect ofdifferent rehabilitation strategies (see [69,70]. In or-der to enable this kind of study, much more effort to-wards normalization and systematic archiving of pa-

tient�’s 3D anatomical and functional imaging data isneeded (see [71] for an initiative for functional imag-ing).

5. Appendix: Rehabilitation tasks

5.1. Conceptual knowledge

1. Anomalous/non anomalous picture categorizationFor each picture a corresponding anomalous plau-

sible picture was drawn (Fig. 4, a). DPI was askedto categorize the anomalous and the normal picturesseparately and to indicate the anomalous part of thepicture.

2. Picture completionApart of each picturewas printed and the subjectwas

asked to complete it with one part among four piecesof drawing belonging to different pictures. To sensitizethe task and make it non-perceptual but only semantic,the picture parts were not perceptually complementaryand their orientation and size were modied (Fig. 4, b).

3. Categorical intruder detectionA intruder detection taskswas constructed using each

of the target picture. Each picture was presented, atrandom location, on a sheet of papers with 3 otherpictures belonging to a another semantic category. DPIwas asked to detect the intruder picture (Fig. 4, c).

4. Functional matching (for artifacts)A multiple-choice task was constructed containing

the target picture, a functional related picture and twodistracters. DPI was asked to point out the functionalrelated picture. Because functionality refers mostlyto artifacts, this experiment was conducted only withartifacts (Fig. 4, d).

5. Correct color detection (for vegetables)A multiple-choice task was constructed using each

picture of vegetables. Each vegetable was presentedin four exemplars: one with its correct color, and 3exemplars of incorrect color in random position. DPIwas instructed to point to the correct colored picture.

6. Picture categorizationDPI was instructed to categorize the pictures in four

semantic categories.



7. Knowledge assessmentThirteen type of questions requiring a yes/no re-

sponse were asked about the items displayed in the pic-tures: for example �“Is it eatable?�”, �“Can it be put ina shoe box?�”, �“Does it live in France?�”, �“Does it haveseeds? �” (for vegetables), �“How to use it?�” (for ar-tifacts), �“How many legs does it have?�” (for animals)etc.

5.2. Phonological input

1. Rhyme judgementThe examiner pronounced two pseudowords which

rhymed or did not rhyme and DPI was asked to say ifthe two pseudowords rhymed or not (e.g. banoume andpanoume, expected response: yes).

2. Discrimination taskThe examiner pronounced two pseudowords that

could be the same or not (if not they differed by asingle phoneme) and DPI had to decide if the twopseudowords were or not identical (e.g. banoume andpanoume, expected response: no). Another version ofthis task consisted in repeating three or four times therst pseudoword before pronouncing the second one.

3. Phoneme detectionA target vowel or consonant was presented to DPI

(auditory and written modality) and he had to indicateif it was contained in the pseudoword pronounced bythe examiner (e.g. Is the sound �‘v�’ in fanre, expectedresponse: no).

4. Syllable detectionA target syllable was presented to DPI (auditory and

written modality) and he had to indicate if it was con-tained in the pseudoword pronounced by the examin-er (e.g. Is the sound �‘vo�’ in volire, expected response:yes).

5. Syllable number identicationDPI had to indicate the number of syllables compos-

ing the pseudowords pronounced by the examiner. DPIcould respond orally or by pointing the correct numberwritten on a sheet of paper.

6. Auditory written syllable matchingDPI had to point thewritten syllable corresponding to

the auditory syllable pronounced by the examiner. Thetarget written syllable was presented among a choiceof 3, 6 or 12 syllables. (e.g. The sound �‘vo�’ has to bematched with his written form among for instance 3possibilities vo, fo and ka).

7. Auditory written pseudoword matchingDPI had to point the written pseudoword correspond-

ing to the auditory syllable pronounced by the exam-iner. The target written pseudoword was presentedamong a choice of 3, 6 or 12 syllables.

8. Auditory written rhyme matchingDPI had to point the written rhyme corresponding to

the auditory rhyme pronounced by the examiner. Thetargetwritten rhymewas presented among a choice of 3,6 or 12 syllables (e.g. the item �‘volire�’ has to bematchedwith the written item that would rhyme with �‘volire�’ ifpronounced among for instance 3 possibilities: ire, are,and ile).

5.3. Phonological output

1. Production of single syllablesDPI had to pronounce the sounds represented by the

rebus. 22 rebus were used to illustrate the all set ofsyllables. As the aim of this session was to make DPIproduce speech sounds, if DPI had difculties inndingthe speech sound corresponding to a given rebus, theexaminer pronounced the name of the rebus.

2. Production of syllabic sequences with increasingcomplexity

To make DPI produce several syllables, more thanone rebus was presented to him. We began to presenta repetition of the same syllable and then complicat-ing the task, we presented different syllables graduallyincreasing the number of different syllables (Fig. 5).

3. Production of syllable with various rhythmsRebuses were presented with some indications of

rhythm. Under rebus a white circle indicated that DPIhad to make a �“long�” syllable, a black circle a normalsyllable and an hyphen stood for silence.

Acknowledgements

The authors thank J. Belmihoub, M. Chevalier, O.Decouche and M. Magne for their help with thiswork. The current study was supported by DirectionRegionale de la Recherche Clinique (AP/HP) through aPRA grant and by a contract interface INSERM AP/HPawarded to AC Bachoud-Levi.



References

[1] A. Guidotti, S. Zanasi and M. Guidotti, Current interpretationof aphasia and medico-social problems of recovery and reha-bilitation of the aphasic patient, Minerva Medica 73(28�–29)(1982), 1997�–2004.

[2] A. Hillis, The role of models of language processing in reha-bilitation of language impairments, Aphasiology 7(1) (1993),5�–26.

[3] W. Best and L. Nickels, From theory to therapy in aphasia:where are we now and where to next? NeuropsychologicalRehabilitation 10(3) (2000), 231�–247.

[4] L. Nickels, Therapy for naming disorders: Revisiting, revisingand reviewing, Aphasiology 16(10�–11) (2002), 935�–980.

[5] M.L. Greenwald et al., Contrasting treatments from severe im-pairments of picture naming, Neuropsychological Rehabilita-tion 5(1�–2) (1995), 17�–49.

[6] A. Caramazza and A.E. Hillis, For a theory of remediationof cognitive decits, Neuropsychological Rehabilitation 3(3)(1993), 217�–234.

[7] A.E. Hillis, Treatment of naming disorders: new issues re-garding old therapies, J Int Neuropsychol Soc 4(6) (1998),648�–660.

[8] G. Le Dorze and C. Pitts, A case study evaluation of theeffects of different techniques for the treatment of anomia,Neuropsychological Rehabilitation 5(1�–2) (1995), 51�–65.

[9] R. Lesser, Cognitive neuropsychological inuences on aphasiarehabilitation, Aphasiology 1(3) (1987), 3�–54.

[10] G. Miceli et al., The treatment of anomia resulting from outputlexical damage: Analysis of two cases, Brain and Language52(1) (1996), 150�–174.

[11] J. Nettlelon and R. Lesser, Therapy for naming difcultiesin Aphasia: Application of a cognitive neuropsychologicalmodel, Journal of Neurolinguistics 6(2) (1991), 139�–157.

[12] J. Robson et al., Phonological naming rehabilitation in jargonaphasia: Positive but paradoxical effects, Journal of the Inter-national Neuropsychological Society 4(6) (1998), 675�–686.

[13] J. Marshall et al., The use of picture/word matching tasksto assist word retrieval in aphasic patients, Aphasiology 4(2)(1990), 167�–184.

[14] L. Nickels and W. Best, Therapy for naming disorders (partII): Specics, surprises and suggestions, Aphasiology 10(2)(1996), 109�–136.

[15] A.M. Raymer et al., Phonological treatment of naming decitsin aphasia: Model-based generalization analysis, Aphasiology7(1) (1993), 27�–53.

[16] L. Nickels and W. Best, Therapy for naming disorders (PartI): Principles, puzzles and progress nresult, Aphasiology 10(1)(1996), 21�–47.

[17] T. Pring et al., Picture/word matching tasks and word retrieval:some follow up data and second thoughts, Aphasiology 4(5)(1990), 479�–483.

[18] S. Franklin, F. Buerk and D. Howard, Generalised improve-ment in speech production for a subject with reproductionconduction aphasia, Aphasiology 16(10) (2002), 1087�–1114.

[19] J. Marshall, T. Pring and S. Chiat, Verb retrieval and sentenceproduction in aphasia, Brain Lang 63(2) (1998), 159�–183.

[20] A.M. Raymer and T. Ellsworth, Response to contrasting verbretrieval treatments: A case study, Aphasiology 16(10�–11)(2002), 1031�–1045.

[21] S. Lowell, P. Beeson and A. Holland, The efcacy of a se-mantic cueing procedure on naming performance of adultswith aphasia, American Journal of Speech-Language Pathol-ogy 4(4) (1995), 109�–114.

[22] R.L. Drew and C.K. Thompson, Model based semantic treat-ment for naming decits in aphasia, Journal of Speech, Lan-guage and Hearing Research 42(4) (1999), 972�–989.

[23] D. Howard et al., Treatment of word retrieval decits in apha-sia. A comparison of two therapy methods, Brain 108(Pt 4)(1985), 817�–829.

[24] R. Cubelli, A. Foresti and T. Consolini, Reeducation strategiesin conduction aphasia, Journal of Communication Disorders21(3) (1988), 239�–249.

[25] A. Hillis and A. Caramazza, Theories of lexical processingand rehabilitation of lexical decits, in Cognitive Neuropsy-chology and Cognitive Rehabilitation, M.J. Riddoch and G.E.Humphreys, eds, Lawrence Erlbaum Associates: Hillsdale,1994.

[26] C.K. Thompson, A.M. Raymer and H.R. Le Grand, Effects ofphonologically based treatment on aphasic naming decits: Amodel-driven approach, in Clinical Aphasiology, T. Prescott,ed., Pro-Ed: Austin, 1991, pp. 239�–261.

[27] T. Pring, A. Davis and J. Marshall, Therapy for word ndingdecits: Can experimental ndings inform clinical work?, inProceedings of the European Concerted Action on the Eval-uation of the Efcacy of Technology in the Assessment andRehabilitation of Brain Damaged Patients, F.J. Stachowiak,ed., Gunter Narr Verlag: Tubingen, 1993.

[28] G. Miceli, R. Capasso and A. Caramazza, The interactionof lexical and sublexical processes in reading, writing andrepetition, Neuropsychologia 32(3) (1994), 317�–333.

[29] C. Jacquemot, E. Dupoux and A.C. Bachoud-Levi, Breakingthe mirror: Asymmetrical disconnection between the phono-logical input and output codes, Cognitive Neuropsychology24(1) (2007), 3�–22.

[30] F. Ramus et al., A psycholinguistic perspective on the ac-quisition of phonology, in Laboratory Phonology 10: Varia-tion, Phonetic Detail and Phonological Representation. Berlin:Mouton de Gruyter., K. Fougeron, D�’Imperio, Vallee, ed., DeGruyter Mouton: Berlin, 2010, pp. 311�–340.

[31] S. Byng, What is aphasia therapy? The treatment of apha-sia from theory to pratice, ed. C.C.D. Muller. 1995, London:Whurr.

[32] P. Indefrey and W.J. Levelt, The spatial and temporal sig-natures of word production components, Cognition 92(1�–2)(2004), 101�–144.

[33] A.W. Ellis and A.W. Young, Human Cognitive Neuropsychol-ogy: A Textbook with Readings 1996, Hove, East SussexLawrence Erlbaum.

[34] R.A. McCarthy and E.K. Warrington, Cognitive Neuropsy-chology: A Clinical Introduction, 1990, San Diego AcademicPress Inc.

[35] T. Shallice, N. Venable and R. Rumiati, From neuropsychol-ogy to mental structure, Cambridge: Cambridge UniversityPress, 1988.

[36] R. Martin, J.R. Shelton and L.S. Yaffee, Language processingand working memory: neuropsychological evidence for sepa-rate phonological and semantic capacities, Journal of Memoryand Language 33 (1994), 83�–111.

[37] R.A. McCarthy and E.K. Warrington, Repeating without se-mantics: surface dysphasia? Neurocase 7(1) (2001), 77�–87.

[38] C. Jacquemot, E. Dupoux and A.C. Bachoud-Levi, Is theword-length effect linked to subvocal rehearsal? Cortex 47(4)(2011), 484�–493.

[39] W.P. Tanner and J.A. Swets, A decision-making theory ofvisual detection, Psychological Review 61(6) (1954), 401�–409.



[40] D. Howard, The treatment of acquired aphasia, Philos TransR Soc Lond B Biol Sci 346(1315) (1994), 113�–120.

[41] R. Jokel and E. Rochon, Treatment for an aphasic namingimpairment: When phonology met orthography, Brain andCognition 32(2) (1996), 299�–301.

[42] A. Basso et al., Acquisition of new �“words�” in normal subjects:a suggestion for the treatment of anomia, Brain Lang 77(1)(2001), 45�–59.

[43] C. Leonard, E. Rochon and L. Laird, Treating naming impair-ments in aphasia: Findings from a phonological componentsanalysis treatment, Aphasiology 22(9) (2008), 23�–947.

[44] J.M. Annoni et al., Advantage of semantic language therapy inchronic aphasia: A study of three cases, Aphasiology 12(12)(1998), 1093�–1105.

[45] W. Best et al., Cueing the words: A single case study of treat-ments for anomia, Neuropsychological Rehabilitation 7(2)(1997), 05�–141.

[46] S. Carlomagno et al., Recovery from moderate aphasia in therst year post-stroke: Effect of type of therapy, Archives ofPhysical Medecine Rehabilitation 82(8) (2001), 1073�–1080.

[47] C. Eales andT. Pring, Using individual and group therapy to re-mediate word nding difculties, Aphasiology 12(10) (1998),913�–918.

[48] R.B. Fink et al., A computer-implemented protocol for treat-ment of naming disorders: Evaluation of clinician-guided andpartially self-guided instruction, Aphasiology 16(10) (2002),1061�–1086.

[49] J. Hickin et al., Phonological therapy for word-nding dif-culties: A re-evaluation, Aphasiology 16(10�–11) (2002), 981�–999.

[50] J. Hickin et al., Efcacy of treatment: Effect on word retrievaland conversation, in: Aphasia Therapy File, S.C. Byng, K.Swinburn and C. Pound, eds, Psychology Press: Hove, 2002.

[51] P.M. Pedersen, K. Vinter and T.S. Olsen, Improvement of oralnaming by unsupervised computerised rehabilitation, Aphasi-ology 15(2) (2001), 151�–169.

[52] M. Rose, J. Douglas andT. Matyas, The comparative effective-ness of gesture and verval treatments for a specic phonologicnaming impairment, Aphasiology 15(10�–11) (2002), 977�–990.

[53] P. Indefrey et al., Neural responses to the production and com-prehension of syntax in identical utterances, Brain Lang 89(2)(2004), 312�–319.

[54] A.C. Bachoud-Levi and E. Dupoux, An Inuence of Syntacticand Semantic Variables on Word Form Retrieval CognitiveNeuropsychology, 2003. 20(2): pp. 163�–188.

[55] A. Caramazza and J.R. Shelton, Domain-specic knowledgesystems in the brain: the animate-inanimate distinction, Jour-nal of Cognitive Neuroscience 10(1) (1998), 1�–34.

[56] R. Schoonen, The internal validity of efcacy studies: de-sign and statistical power in studies of language therapy foraphasics, Brain and Language 41(3) (1991), 446�–464.

[57] R.R. Robey, The efcacy of treatment for aphasic persons: Ameta-analysis, Brain and Language 47(4) (1994), 582�–608.

[58] F. Pulvermuller and M.L. Berthier, Aphasia therapy on a neu-roscience basis, Aphasiology 22(6) (2008).

[59] S.K. Bhogal, R. Teasell and M. Speechley, Intensity of aphasiatherapy, impact on recovery, Stroke 34(4) (2003), 987�–993.

[60] L.A. Nickels, A sketch of the cognitive processes involvedin the comprehension and production of single words, inhttp://www.maccs.mq.edu.au/#lyndsey/model.doc. 2000.

[61] W.J. Levelt, Models of word production, Trends Cogn Sci 3(6)(1999), 223�–232.

[62] G.S. Dell, F. Chang and Z.M. Grifn, Connectionist Modelsof Language Production: Lexical Access and GrammaticalEncoding, Cognitive Science 23(4) (1999), 517�–542.

[63] A. Caramazza, How many levels of processing are there in lex-ical access? Cognitive Neuropsychology 14(1) (1997), 177�–208.

[64] S. Franklin, Dissociations in auditory word comprehension;evidence from nine uent aphasic patients, Aphasiology 3(3)(1989), 189�–207.

[65] D. Norris, Shortlist: A connectionist model of continuousspeech recognition, Cognition 52(3) (1994), 189�–234.

[66] A.M. Raymer et al., Translational research in aphasia: fromneuroscience to neurorehabilitation, J Speech Lang Hear Res51(1) (2008), S259�–S275.

[67] N.F. Dronkers, A new brain region for coordinating speecharticulation, Nature 384(6605) (1996), 159�–161.

[68] A.E. Hillis et al., Re-examining the brain regions crucial for or-chestrating speech articulation, Brain 127(Pt 7) (2004), 1479�–1487.

[69] M. Meinzer et al., Functional re-recruitment of dysfunction-al brain areas predicts language recovery in chronic aphasia,Neuroimage 39(4) (2008), 2038�–2046.

[70] M. Meinzer and C. Breitenstein, Functional imaging studies oftreatment-induced recovery in chronic aphasia, Aphasiology22(12) (2008), 1251�–1268.

[71] Brainmap, http://www.brainmap.org.[72] S. Abel, K. Willmes and W. Huber, Model-oriented naming

therapy: Testing predictions of a connectionist model, Apha-siology 21(5) (2007), 411�–447.

[73] B. Crosson et al., Regional changes in word-production later-ality after a naming treatment designed to produce a rightwardshift in frontal activity, Brain Lang 111(2) (2009), 73�–85.

[74] C.A. Fisher, C.E. Wilshire and J.L. Ponsford, Word discrimi-nation therapy: A new technique for the treatment of a phono-logically based word-nding impairment, Aphasiology 23(6)(2009), 676�–693.

[75] J. Fridriksson et al., Neural correlates of phonological andsemantic-based anomia treatment in aphasia, Neuropsycholo-gia 45(8) (2007), 1812�–1822.

[76] W. Best et al., A treatment for anomia combining semantics,phonology and orthography, in Language disorders in childrenand adults: Psycholinguistic approaches to therapy, S. Chiat,J. Law and J. Marshall, eds, Whurr Publishers Ltd: London,1997, pp. 102�–129.

[77] M. Laganaro and F.X. Alario, On the locus of syllable fre-quency effect in speech production, Journal of Memory andLanguage 55(2) (2006), 178�–196.

[78] J. Marshall et al., Why does monitoring fail in jargon aphasia?Comprehension, judgment, and therapy evidence, Brain Lang63(1) (1998), 79�–107.

[79] A. Murray and D.M. Jones, Articulatory complexity at itemboundaries in serial recall: the case of Welsh and English digitspan, J Exp Psychol Learn Mem Cogn 28(3) (2002), 594�–598.

[80] G. Pashek, Gestural facilitation of noun and verb retrieval inaphasia: A case study, Brain and Language 65(1) (1998),177�–180.

[81] K. Renvall et al., Anomia treatment with contextual priming:A case study, Aphasiology 17(3) (2003), 305�–328.

[82] K. Renvall, M. Laine and N. Martin, Treatment of anomiawith contextual priming: Exploration of a modied procedurewith additional semantic and phonological tasks, Aphasiology21(5) (2007), 499�–527.



[83] J.D. Rider et al., Using semantic feature analysis to improvecontextual discourse in adults with aphasia, American Journalof Speech-Language Pathology 17(2) (2008), 161�–172.

[84] M. Rose and J. Douglas, Treating a semantic word productiondecit in aphasia with verbal and gesture methods, Aphasiol-ogy 22(1) (2008), 20�–41.

[85] C.K. Thompson, K.Kearns andL.A. Edmonds, An experimen-tal analysis of acquisition, generalisation, and maintenanceof naming behaviour in a patient with anomia, Aphasiology20(12) (2006), 1226�–1244.

Documents

IOS Press Speci city in rehabilitation of word production ...€¦ · Galley Proof 2/11/2011;16:34 File: ben358.tex;BOKCTP/wyn p. 1 Behavioural Neurology 25 (2012) 1J29 1 DOI 10.3233/BEN-2012-0358