The Text-to-Picture Semantic Association Task (TP-SAT): standardization, validation and application in people with a brain tumour

(1)

The Text-to-Picture Semantic Association Task (TP-SAT): standardization,

validation and application in people with a brain tumour

Master thesis submitted for the fulfilment of the requirements for the degree of Master of Science in Neurolinguistics

By Cheyenne Svaldi Student number: s3550826 1st_{of July 2020} University of Groningen Academic year: 2019-2020 Supervised by Dr Adrià Rofes

(2)

Abstract

Introduction: There is still a need for tests following the rationale of subcortical language mapping during awake brain surgery, allowing to assess multiple language levels. Also, there

is a need for standardized tests in Dutch looking into semantic processing.

Aims: The present study describes the development, standardization and validation of the Text-to-Picture Semantic Association Task (TP-SAT). This is a multimodal test as it assesses

semantic processing, word reading, lexical retrieval and speech planning and articulation

within the same item. The test is also the first Dutch tool to assess both thematic and

taxonomic semantic relations.

Methods: The TP-SAT was standardized and validated in 54 people without neurological or language impairment. Also, the test was evaluated in two people who had a frontal glioma

removed.

Results and discussion: Results of the standardization study provided expected accuracy scores of people without neurological or language impairment. Participants between 60 and 76 years

scored significantly lower which is believed to be caused by cognitive slowness due to healthy

ageing. Both people with a brain tumour had significantly lower scores than the norm group,

indicating that the TP-SAT is a sensitive tool to detect language impairments in people with a

brain tumour.

Conclusion: The present study indicates that the TP-SAT is a relevant tool in the context of awake brain surgery, but further application in people with a glioma is necessary. Also, the

validation study should be repeated in people with post-stroke aphasia.

Keywords: Awake brain surgery, semantic association, standardization, subcortical language testing

(3)

Acknowledgments

For the development of this thesis want to express my gratitude to my supervisor, Dr Adrià

Rofes, for guiding me through every step of the process and giving me the chance to write

about language in people with a brain tumour, a topic I am very passionate about. Also, I want

to give my biggest thanks to speech and language therapist Erik Robert at AZ Sint Lucas Ghent,

who made me aware about the need for the development of subcortical tests and provided me

with interesting new insights and literature to use for my thesis. I also want to thank Dr Henry

Colle, who provided me with the subject of my thesis and taught me a lot about the surgical

aspects of awake neurosurgery and intraoperative language mapping. I also want to thank

everyone who participated in my study, particularly Elien Hendrickx, who was always open to

(4)

List of Figures

Figure 1 Lateral positioning (park bench position) during awake brain surgery ... 12

Figure 2 A minimal setup of the participants and materials during awake neurosurgery ... 14

Figure 3 Example of intraoperative mapping following Colle ... 16

Figure 4 Example of a test item of the object naming task of the Dutch Linguistic Intraoperative Protocol... 28

Figure 5 Example test item of the Semantic Association Task... 29

Figure 6 Example of a slideshow following the principle of the Quick Mixed Test ... 32

Figure 7 Example of a test item of the Text-to-Picture Semantic Association Task ... 35

Figure 8 Neurocognitive model of single-word processing by Rofes et al. (2019b) ... 38

Figure 9 Three-dimensional representation of the latent semantic spaces of two stimuli of the TP-SAT. ... 55

Figure 10 MRI-scan of the brain of P1 the day before surgery (January 2020) ... 71

(7)

List of Tables

Table 1 Demographic characteristics of the participants in the standardization study ... 61 Table 2 Expected accuracy on the TP-SAT in relation to age and education ... 64 Table 3 Cut-off scores per age group and education level. ... 64 Table 4 Mean accuracy scores on the TP-SAT for taxonomic and thematic conceptual relations

per age group... 65

Table 5 Means and standard deviations of the participant group on the Semantic Association

Task and the digit span task of the WAIS-IV ... 69

Table 6 Accuracy of the participants with a brain tumour on the TP-SAT, SAT and digit span

(8)

1. Chapter one: Introduction

Awake brain surgery is considered the gold standard for the removal of a type of brain tumour,

namely supratentorial gliomas in the left hemisphere (Berger & Ojemann, 1992; De Witt

Hamer, Robles, Zwinderman, Duffau, & Berger, 2012; Duffau et al., 1999; Penfield & Boldrey,

1937; Spena et al., 2010; Taylor & Bernstein, 1999). The main goals of this procedure are to

maximize tumour resection, while minimizing postoperative language/cognitive damage

(Duffau et al., 1999; Taylor & Bernstein, 1999). While several tasks have been developed to

assess these people during surgery and particularly to map language in the cortex (e.g.,

Bastiaanse, Edwards, Mass, & Rispens, 2003; De Witte et al., 2015b; Ohlerth, Valentin,

Vergani, Ashkan, & Bastiaanse, 2020; Połczyńska, 2009), there is still a need for apposite tests

for subcortical language mapping (e.g., Coello et al., 2013; De Witte et al., 2015a).

The current study considers the development, standardization and validation of a new

multimodal task specifically adapted for subcortical language mapping, the Text-to-Picture

Semantic Association Task (TP-SAT), as well as its application in two people with a glioma.

This new task aims to assess multiple language levels within the same item, making it a reliable

tool for the identification of subcortical structures important for language. The TP-SAT also

aims to provide a new tool for the online monitoring of semantic processing and is the first

clinical test in Dutch that distinguishes between taxonomic and thematic semantic relations

(Goldwater, Bainbridge, & Murphy, 2016; Jouravlev & McRae, 2016; Lin & Murphy, 2001;

Simmons & Estes, 2008).

This thesis consists of five chapters. In the first chapter, the details of awake brain surgery are

(9)

of commonly used intraoperative language tasks. In the second chapter, the theoretical

background of the TP-SAT is discussed and its development. In the third chapter, the

standardization and validation of the TP-SAT in a participant group without neurological

impairment of different ages and educational backgrounds (N = 54) are described. In the fourth

chapter, the application of the TP-SAT in two people who successfully had a frontal glioma

removed is outlined. The final chapter encompasses an overall discussion and conclusion.

1.1. Awake brain surgery

Gliomas are one of the most common primary brain tumours (Louis et al., 2016; Ostrom et al.,

2014; Rasmussen et al., 2017). They emerge from the glial cells in the central nervous system

and often infiltrate the brain tissue (Ostrom et al., 2014; Rasmussen et al., 2017). Based on the

histology of the glial cells, a distinction is made between gliomas, which are further graded

according to their morphology and malignancy. A generally accepted classification is provided

by the World Health Organization (WHO) which differentiates between low-grade (i.e., grade

I and II) and high-grade (i.e., grade III and IV) gliomas (Louis et al., 2016). This classification

implies that low-grade gliomas tend to infiltrate the brain tissue due to their slow-growing

nature while high-grade gliomas are more malignant and fast growing, pushing the neural

structures to the sides of the tumoral mass. It is also worth noting that all low-grade gliomas

eventually become anaplastic and will have a higher grade, also causing death (Bello et al.,

2007; Claus et al., 2015; Duffau, 2007). Regardless of this classification, common symptoms

that lead to the discovery of this brain tumour are language, cognitive or motor deficits and

seizures or headaches, but these are often also discovered by accident when looking for

non-related symptoms. For example, during neuroimaging after head trauma (Duffau et al., 1999;

(10)

Gliomas tend to be situated close to or within eloquent areas important for language or motor

functioning. This implies areas in the left but also right hemisphere, typically in the periphery

of the Sylvian fissure and the motor cortex (Duffau et al., 1999; Rasmussen et al., 2017). This

makes tumour resection under general anaesthesia less desirable, since it can result in

permanent postoperative neurological deficits due to interindividual variability in those

eloquent areas (De Witt Hamer et al., 2012; Ojemann, Ojemann, Lettich, & Berger, 1989;

Taylor & Bernstein, 1999). Following this, awake brain surgery, a technique originally used

for people with epilepsy, was introduced for the removal of gliomas (Bulsara, Johnson, &

Villavicencio, 2005; Ojemann & Mateer, 1979; Penfield & Boldrey, 1937; Penfield & Roberts,

1959). This procedure uses direct electrical stimulation (DES) to map eloquent areas

intraoperatively which consists of administering a low-intensity current on cortical or

subcortical areas surrounding the tumour for four seconds at a time (Berger & Ojemann, 1992;

Szelényi et al., 2010).

Nowadays, awake brain surgery with intraoperative language mapping is considered the ‘gold standard’ for the surgical removal of gliomas and has been proven to be a reliable, safe and precise method which maximizes the amount of tumour that can be removed and minimizes

permanent postoperative deficits, leaving the quality of life of the person with a brain tumour

unaffected or even improving it (Berger & Ojemann, 1992; De Witt Hamer et al., 2012; Duffau

et al., 1999; Mandonnet, Winkler, & Duffau, 2010; Spena et al., 2010; Taylor & Bernstein,

1999). At first, only low-grade gliomas were operated on using awake brain surgery, but more

recently high-grade gliomas are also resected with this technique (Bello et al., 2007; Ostrom et

(11)

1.1.1. Asleep-Awake-Asleep procedure

The most commonly used procedure for awake brain surgery in Europe is the

asleep-awake-asleep (AAA) procedure with language mapping during the awake phase (Bello et al., 2007;

Duffau et al., 2002, 2003; Huncke, Van de Wiele, Fried, & Rubinstein, 1998; Spena et al.,

2010). Nonetheless, some centres (e.g., in Japan; Kayama, 2012) are now using a continuous

awake or fully awake procedure, throughout which the person with a brain tumour is lightly

sedated. This has been shown to reduce operation time but could increase stress levels of the

person with a brain tumour (Eseonu et al., 2017; Fiegl et al., 2013; Wahab, Grundy, &

Weidmann, 2011). The medical team during awake brain surgery typically consists of a

neurosurgeon and an assistant, nursing staff, an anaesthesiologist and a linguist and/or

neuropsychologist (Rofes et al., 2017a).

To be admitted for this procedure and to ensure reliable intraoperative testing, the person with

a brain tumour cannot have severe preoperative language deficits. Also, the patient needs to be

in good general health, emotionally stable and cooperative for the awake phase to be successful

(Kayama, 2012; Robert, Visch-Brink, & Beeckman, 2013). Customarily, preoperative

structural neuroimaging is conducted to identify the lesion location. Furthermore, functional

neuroimaging is increasingly used by more and more centres to elucidate the language

dominant hemisphere and based on which intraoperative tasks may also be selected (Freyschlag

et al., 2018; Kayama, 2012; Połczyńska, 2009; Robert et al., 2013). To determine the

associative subcortical pathways within or close to the lesion, brain tractography with diffuse

tensor imaging (DTI) can be conducted serving as a guide during subcortical language mapping

(Bello et al., 2007; Catani & de Schotten, 2008; Duffau, 2007; Duffau et al., 1999; Robert et

(12)

1.1.1.1. First asleep phase

During the first asleep phase, the person with a brain tumour is generally placed in a lateral

position (see Figure 1) (Huncke et al., 1998; Velho, Naik, Bhide, Bhople, & Gade, 2019). This

is considered the most comfortable position for the patient and facilitates endotracheal

intubation at all times which is used to enable breathing (Huncke et al., 1998). When the tumour

is situated close to the supplementary motor area (SMA), patients are often placed in a supine

position which facilitates resection and allows monitoring the movement of both upper limbs

(Robert et al., 2013). Under intravenous general anaesthesia with endotracheal intubation, the

head of the person with a brain tumour is fixated with a three-point Mayfield holder (Huncke

et al., 1998; Taylor & Bernstein, 1999). The contralateral hand and arm are placed on a padded

armrest to enable testing and monitoring of motor functioning (Velho et al., 2019).

Figure 1 Lateral positioning (park bench position) during awake brain surgery. The head of the person with a

brain tumour is fixated using a Mayfield holder and the contralateral arm is placed on a padded armrest to enable mapping and monitoring of motor functions (Velho et al., 2019).

(13)

After this, a craniotomy is performed for the removal of a bone flap comprising the size of the

lesion and adjacent areas (Kayama, 2012; Robert et al., 2013). The person with a brain tumour

is then awoken by the anaesthesiologist before the removal of the dura, to reduce the risk of

complications if the patient coughs during the removal of the tube that assists breathing (i.e.,

endotracheal extubation) (Duffau et al., 1999; Kayama, 2012; Robert et al., 2013).

1.1.1.2. Awake phase: direct electrical stimulation

When the person with a brain tumour is fully awake, DES is performed by the neurosurgeon

with a monopolar or with a bipolar probe with 5mm-spaced tips (Berger & Ojemann, 1992;

Duffau et al., 1999; Ojemann et al., 1989; Taylor & Bernstein, 1999). This probe is directly

applied to the cortex or subcortical associative pathways, producing a monophasic or biphasic

current with a wave pulse of 50 or 60 hertz (Hz) and a maximal duration of four seconds to

avoid epileptic seizures (Berger & Ojemann, 1992; Duffau, 2007; Duffau et al., 1999; Ojemann

et al., 1989). When applied, a transient virtual lesion is induced in those areas eloquent for

motor, cognitive and/or language functioning (Duffau, Moritz-Gasser, & Mandonnet, 2014).

First, a cortical threshold for DES is established, generally ranging from 0.5 to 16 milliamperes

(mA) and defined by a movement of the face or limb, speech arrest or dysarthria when counting

or responding to another type of automated task which requires little to no lexical-semantic

engagement (Bello et al., 2007; Duffau et al., 1999; Kayama, 2012; Szelényi et al., 2010;

Taylor & Bernstein, 1999). A minimal setup of the participants and medical devices during

(14)

Figure 2 A minimal setup of the participants and materials during awake neurosurgery (Fiegl, n.d.): AN =

Anaesthesiologist; AS = Assistant; NES = Neurosurgeon; NS = Neuropsychologist or aphasiologist; SN = Surgical nurse; 1 = Ventilator in standby; 2 = Ultrasonic surgical aspirator to remove large tumours from the inside out; 3 = Electrical stimulator; 4 = Neuronavigation; 5 = Laptop for presenting stimuli to the patient; 6 = Ultrasound to check tumour boundaries; 7 = Surgical Multi-display Unit with surgical instruments. Also, a surgical microscope is used during the procedure which is mounted in the ceiling and for this reason not shown in the figure.

After the threshold is established, intraoperative cortical mapping is performed by the

neurosurgeon and a trained aphasiologist, clinical linguist or neuropsychologist who

administers several linguistic tasks which were rehearsed before surgery (De Witte et al.,

2015b; Kayama, 2012; Robert et al., 2013; Rofes et al., 2017a). An overview of

intraoperatively used language tasks can be found in section 1.3.2.. An area is considered

eloquent for motor or language functioning when a positive response (i.e., a speech, language

or motor disturbance) is observed at least three times while stimulating the same area in a

(15)

results due to peripheral spreading of electrical stimulation as well as the start of an epileptic

seizure, afterdischarges can be monitored on an electrocorticogram (Kayama, 2012; Sanai,

Mirzadeh, & Berger, 2008).

Possible transient speech or language disturbances are, among others, semantic paraphasia (i.e.,

the substitution of the target word by a word with a similar meaning, e.g., ‘orange’ instead of ‘apple’), phonological paraphasia (i.e., one or more phonemes of the target word are omitted or substituted but the target word is still recognizable, e.g., ‘lat’ instead of ‘cat’), dysarthria

(e.g., slow and effortful speech caused by muscle weakness), alexia (i.e., an acquired

disturbance of reading), speech arrest (i.e., the inability to speak while language is intact caused

by seizure) or anomia (i.e., the inability to retrieve a word). These errors indicate that the

stimulated area needs to be preserved and is considered a functional border for tumour resection

(Duffau et al., 1999; Taylor & Bernstein, 1999). Up until recently, eloquent areas were labelled

using sterile numbered paper tags, but the insertion of positive points into the surgical

microscope is starting to be more commonly used (see Figure 3) (Bello et al., 2007; H. Colle,

(16)

Figure 3 Example of intraoperative mapping following Colle in which the positive points (i.e., cortical areas

identified with DES) for language or motor functioning are inserted into the surgical microscope. After three rounds of stimulation, there was a speech arrest in zone 8 and 9 (in red) and dysarthria in zone 2 and 7 (in yellow). In the other stimulation zones (in green) no speech or language disturbances were observed (Robert et al., 2013).

After cortical mapping, the neurosurgical team can decide how to approach the tumour safely.

While tumour resection is performed, some surgical teams monitor patients by asking them to

name objects continuously while moving the right arm up and down and closing the hand to

see how s/he is doing. This is called dual tasking and assesses both language and motor

functioning (Mandonnet al., 2017; Rofes et al., 2017a). Other teams monitor the wellbeing of

the person with a brain tumour by talking with him/her or running a spontaneous speech task,

for example describing a video or re-telling a story (Robert et al., 2013). After some of the

tumour is removed, subcortical mapping with DES is performed in the associative white-matter

pathways (e.g., arcuate fasciculus, inferior fronto-occipital fasciculus) to decide how far

tumoral resection can go while minimizing postoperative deficits (Spena et al., 2010; Szelényi

(17)

between electrical stimulation and tumour resection (Bello et al., 2007; Kayama, 2012).

Generally, the awake phase lasts one to three hours (Robert et al., 2013).

1.1.1.3. Second asleep phase

The second asleep phase is induced when (supra)total tumoral resection has been accomplished

or the functional borders of resection have been reached. The person with a brain tumour is

sedated again under general anaesthesia and reintubated (Huncke et al., 1998; Robert et al.,

2013). In what follows, the common language impairments in people with a glioma are

discussed in detail.

1.2. Common language impairments in people with a glioma

Preoperatively, linguistic functioning of the person with a brain tumour is evaluated to establish

a baseline for intra- and postoperative comparison, to decide if the patient is eligible for awake

brain surgery and to determine which tasks to include for intraoperative testing (De Witte et

al., 2015b; Mandonnet, Nouet, Gatignol, Capelle, & Duffau, 2007; Robert et al., 2013). Only

items that were answered correctly in the preoperative phase are used for intraoperative testing

to ensure that language errors were caused by DES and avoid “false positives” due to errors

that were present preoperatively (De Witte et al., 2015b; Kayama, 2012; Mandonnet et al.,

2017; Robert et al., 2013). An alternative approach is to assess which word types or word

properties are most commonly impaired (e.g., nouns or high-frequency words). When an effect

is detected, all words belonging to a specific category or sharing these properties are deleted.

This approach has been reported in only one study and it was not successful, as the authors

(18)

Talacchi, Santini, Miozzo, & Miceli, 2017b). Preoperative testing generally takes place within

one month or one week before surgery (Rofes et al., 2017a).

People with a glioma may present aphasic symptoms before surgery, but generally, no or mild

language impairments are observed (Bello et al., 2007; Desmurget, Bonnetblanc, & Duffau,

2007; Duffau, 2007; Satoer et al., 2017; Walker & Kaye, 2003). This can be explained by the

low growth rate of on average five millimetres per year of low-grade gliomas which facilitates

functional reorganization of language functions due to neuroplasticity (Bizzi et al., 2012; De

Witte & Mariën, 2013; Duffau, 2014; Satoer et al., 2017). Based on studies using formal testing

(e.g., extensive or comprehensive language batteries) to evaluate language functioning, it was

suggested that several language domains may be impaired, such as lexical-semantics

(Antonsson et al., 2018; Campanella, Skrap, & Vallesi, 2016; Racine, Li, Molinaro, Butowski,

& Berger, 2015; Rofes et al., 2018; Satoer et al., 2012), syntax (Rofes et al., 2018) or spelling

(Tomasino et al., 2015; van Ierschot, Bastiaanse, & Miceli, 2018; Wolthuis, 2019). In

spontaneous speech, linguistic impairments may also be observed (Rofes et al., 2018; Satoer

et al., 2018; Satoer, Vincent, Smits, Dirven, & Visch-Brink, 2013). For example, a longitudinal

study by Satoer and colleagues (2013, 2018) indicated that Dutch-speaking people with a

glioma may produce more incomplete sentences or a higher number of repetitions than people

without brain damage. According to the authors, possible explanations for these observations

are a lexical retrieval deficit or a syntactic impairment.

These preoperative language impairments are a predictor for intra- and postoperative language

functioning (Taylor & Bernstein, 1999; Trinh et al., 2013). When severe language impairments

are observed before surgery, the person with a brain tumour will not be included for awake

(19)

“false positives” (Bello et al., 2007; De Witte & Mariën, 2013). Some teams have advocated for a score of 80% or higher in naming tasks, albeit this measure responds to extensive clinical

experience, it has not been objectively tested (Bello et al., 2006; Robert et al., 2013).

Most centres conduct a first postoperative linguistic evaluation within three to five days after

surgery (Robert et al., 2013; Rofes et al., 2017a). Currently, there are no bedside tests that are

appropriate to be used in this population, particularly, tests that can be administered by

non-language experts. Nonetheless, different European teams are making efforts to fulfil this need

(A. Rofes, personal communication, March 24, 2020). Immediately after surgery, it is common

that transient and mild language problems are observed which normally resolve within three

months post-surgery. In some rare cases new and permanent postoperative deficits emerge

(Bello et al., 2007; De Witte & Mariën, 2013; Duffau et al., 1999, 2002, 2003; Racine et al.,

2015; Satoer et al., 2012; Spena et al., 2010). When tumour resection caused damage to those

subcortical pathways important for language, however, an increase of novel and long-term

postoperative deficits has been suggested, even when eloquent cortical areas were preserved

(Duffau, 2014; Papagno et al., 2012; Trinh et al., 2013). This stresses the importance of the

preservation of white matter pathways, as these structures are argued to be less prone to benefit

from brain plasticity processes, not to say that they connect multiple brain areas within or

across hemispheres (Catani & de Schotten, 2008; Duffau, 2014; Trinh et al., 2013).

1.3. Intraoperative language testing 1.3.1. Criteria for intraoperative tasks

For a linguistic task to be eligible for intraoperative language mapping, several criteria must

(20)

and relevant for awake brain surgery (Rofes, de Aguiar, & Miceli, 2015a; Rofes & Miceli,

2014; Rofes, Spena, Miozzo, Fontanella, & Miceli, 2015b). In what follows, every criterion

will be discussed in detail:

First, the task needs to be based on the neurocognitive models of language processing. These

typically box-and-arrow models encompass hypotheses based on the functional architecture of

the cognitive system (e.g., Coltheart, 2001; Whitworth, Webster, & Howard, 2014). Test

developers should describe in detail which levels of language processing the task assesses to

enable identification of the origin of language errors (Rofes et al., 2015b). Also, the main

language component the task targets to assess needs to be extensively described in terms of the

current neurolinguistic theories (Rofes & Miceli, 2014). It should also be indicated how deficits

of the language component would manifest themselves as can be seen in people with aphasia.

For example, object naming assesses, among many processes, semantics or conceptual

representations. It is important to assess this level, because impairments in semantics or

conceptual representations affect both language production and comprehension, hence, having

a big impact on overall language abilities and quality of life (Goodglass & Wingfield, 1997;

Jefferies & Lambon Ralph, 2006; Warrington & Shallice, 1984; Whitworth et al., 2014). This

is deemed relevant since lexical-semantic deficits have been indicated in people with a glioma

(Antonsson et al., 2018; Campanella et al., 2016; Racine et al., 2015).

Besides, the task needs to be easy to administer within the context of awake brain surgery

(Rofes & Miceli, 2014). To be used during intraoperative language mapping, item presentation

and response cannot last longer than four seconds which is the time of DES (De Witte et al.,

2015b; Kayama, 2012; Rofes & Miceli, 2014). Nonetheless, some teams are currently using

(21)

translation (e.g., Motomura et al., 2014). Other than that, the test should allow online

monitoring of the assessed language functions and should be computer-based. The fixed

position of the head of the patient and the restricted movement of the contralateral hand and

arm need to be taken into account since these limit the possibilities of tasks requiring head

and/or hand movements, such as writing and dual tasks (e.g., indicating the correct answer on

a touch screen) (De Witte & Mariën, 2013; Mandonnet et al., 2017).

Furthermore, the linguistic test needs to be standardized in a population without neurological,

speech or language impairments and in the targeted language to reliably interpret results in the

clinical population (Rofes et al., 2015a; Rofes et al., 2017a; Rofes & Miceli, 2014; Van Borsel,

2009). Here, it needs to be taken into account that older people, particularly older than 70, may

perform worse on language tasks, especially when involving lexical retrieval (e.g., object

naming) (Goral, 2004; Peelle & Peelle, 2019). Several cognitive and neurobiological theories

have been proposed to explain this, such as cognitive slowness (Salthouse, 1996) and lighter

and smaller brains due to healthy ageing (Peelle & Peelle, 2019). Every test item needs to be

carefully selected and has to elicit the target word or response and should be controlled for

relevant psycholinguistic variables or word properties (e.g., word frequency, imageability, age

of acquisition). It is only in this manner that results can be compared and generalized across

studies and that intraoperative errors can be carefully diagnosed (Rofes & Miceli, 2014). The

test should be administered consistently in terms of instructions and scoring. Also, naming

tasks should consist of pictures that are easy to respond to, for example, black and white line

drawings commonly used in several other tasks (e.g., De Witte et al., 2015b; Howard &

Patterson, 1992; Snodgrass & Vanderwart, 1980) or videos since these are believed to be a

more natural representation of actions than pictures or drawings and require fewer inferences

(22)

Finally, the task should be relevant for intraoperative language mapping, meaning it should

identify those areas that are important for language functioning, while allowing extensive

tumour resection (Rofes & Miceli, 2014). For this purpose, the task is more considered as a

surgical tool than an experimental tool. This implies that, together with the correct

administration of DES by the neurosurgeon, the test should accurately identify those cortical

and subcortical structures eloquent for linguistic functioning without inducing new

postoperative deficits other than those that were observed preoperatively (Rofes & Miceli,

2014).

Apart from its use during intraoperative testing, the linguistic test could be used in other

contexts. For example, the test can contribute to the evaluation of linguistic functioning in pre-

and postoperative phases, or in other populations (e.g., post-stroke aphasia). In this manner,

the clinical linguist or neuropsychologist can obtain an in-depth analysis of the linguistic

functioning of the person with a brain tumour and can give possible indications for therapy.

Differently, standardized tasks can be used on an experimental level, providing new insights

into the neural organization of language.

1.3.2. Criteria for subcortical tests

On top of these general criteria for the development of intraoperative language tests, rationale

can be formulated for tests aiming to identify eloquent subcortical structures (Bello et al., 2007;

De Witte et al., 2015a; Duffau, 2015; Duffau et al., 2014; Połczyńska, 2009). Over the years,

a shift took place from a ‘topological’ (i.e., static) view of the neural organization of language

(23)

model; Lichtheim, 1885) towards an hodotopical (i.e., dynamic) view (from the Greek “hodos”

[pathway]), which heavily relies on the relation between language functions and white-matter

brain structures (e.g., Duffau et al., 2014). According to this account, language is organized in

several dynamic subnetworks functioning in parallel and subserved by both cortical and

subcortical structures.

An example of a hodotopical model is the visual processing model of Duffau and colleagues

(2014) in an adaptation of the cognitive model of Hickok and Poeppel (2007) which

emphasizes a dual-stream architecture of language processing with a dorsal or ‘phonological’

stream and a ventral or ‘semantic’ stream. This model implies that axonal connectivity needs

to be taken into account when performing intraoperative language mapping to reduce

postoperative deficits, even when eloquent cortical areas are preserved (Bello et al., 2007;

Chang, Raygor, & Berger, 2015; Duffau, 2015; Duffau et al., 2003; Duffau, Gatignol,

Mandonnet, Capelle, & Taillandier, 2008; Papagno et al., 2012; Połczyńska, 2009). This

implies that subcortical language mapping needs to consider specific rationale to accurately

identify associative tracts relevant for language functioning in a time-efficient manner (Coello

et al., 2013; De Witte et al., 2015a; Duffau, 2014; Satoer et al., 2017).

While cortical areas can be clearly defined with neuroimaging, this is not the case for white

matter tracts, as the latter run through the whole three-dimensional space beneath the cortex

and can connect distant cortical areas (Catani & de Schotten, 2008; Duffau, 2014; Kinoshita et

al., 2015; Martino, Brogna, Robles, Vergani, & Duffau, 2010; Sarubbo, De Benedictis,

Maldonado, Basso, & Duffau, 2013). For example, the inferior fronto-occipital fasciculus has

terminations in the prefrontal and occipital cortex (Martino et al., 2010; Sarubbo et al., 2013).

(24)

Sakas, & Stranjalis, 2015). The frontal aslant and frontostriatal tract which both run through

the frontal lobe, for example, cross fibres with each other when running towards the

supplementary motor area (SMA), making it difficult to distinguish them (Kinoshita et al.,

2015). A commonly used technique for the visualisation of the subcortical tracts is diffusion

tensor imaging (DTI).

While DTI can serve as a guide for subcortical mapping, it cannot always distinguish these

crossing fibres (Koutsarnakis et al., 2015; Szelényi et al., 2010). Also, this technique can be

influenced by interindividual variability and tracts invaded by a tumour are difficult to visualise

(Chang et al., 2015; Leclercq et al., 2010; Raffa et al., 2016). To provide a more reliable

visualization of the cortico-subcortical connections before surgery, the combination of

navigated transcranial magnetic stimulation (nTMS) and DTI has been proposed (Negwer et

al., 2017; Raffa et al., 2016; Sollmann et al., 2016). This technique, however, is time-intensive

and expensive making subcortical mapping still the most reliable technique to identify

associative pathways important for language functioning (Chang et al., 2015; Leclercq et al.,

2010; Sollmann et al., 2016).

Apart from an anatomical overlap, associative white matter tracts also show functional overlap

and the specific functions of every pathway are still under debate (Bello et al., 2007; Duffau,

2014; Herbet, Zemmoura, & Duffau, 2018; Kinoshita et al., 2015; Mandonnet, Gatignol, &

Duffau, 2009; Sali et al., 2018). A wide range of functions can be attributed to one pathway

due to its connection with different cortical areas or, the other way around, one function has

been attributed to several pathways (Bello et al., 2007; Duffau et al., 2014). The inferior

longitudinal fasciculus which connects the temporo-occipital areas with the anterior temporal

(25)

such as semantic processing, reading and object recognition (Duffau et al., 2014; Herbet et al.,

2018; Mandonnet et al., 2009; Sali et al., 2018). To come up with this functional understanding

of subcortical pathways, teams have used different tasks such as object naming (Bello et al.,

2007; Duffau et al., 2008; Han et al., 2013; Mandonnet et al., 2009), semantic association (Han

et al., 2013; Herbet, Moritz-Gasser, & Duffau, 2017) and reading (Motomura et al., 2014). In

what follows, a review of commonly used intraoperative language tests is given.

1.3.3. Commonly used intraoperative language tests

Since awake neurosurgery is used for the resection of brain tumours, object naming (e.g., the

picture of a duck is presented and the participant has to say “this is a duck” or “duck”) and

automated speech tasks such as counting have been considered the ‘gold standard’ for language

mapping (for a review see De Witte & Mariën, 2013; Dragoy, Chrabaszcz, Tolkacheva,

Buklina, 2016; Duffau, 2014; Duffau et al., 1999, 2002; Kayama, 2012; Rofes et al., 2015b;

Taylor & Bernstein, 1999). While originally this was often the only paradigm used

intraoperatively (e.g., Duffau et al., 2002, 2003; Ojemann & Mateer, 1979; Penfield & Roberts,

1959), over the years novel tests and assessment protocols were developed for intra- and

perioperative (i.e., pre- and postoperative) testing (e.g., Bastiaanse et al., 2003; De Witte et al.,

2015a-b; Ohlerth et al., 2020; Połczyńska, 2009; Robert et al., 2013; Rofes et al., 2017a). These

novel tests focus on language levels that are not assessed by object naming, such as syntactical

processing (e.g., Bastiaanse et al., 2003; Połczyńska, 2009).

A survey conducted in several European centres that perform awake brain surgery indicated

that the most commonly used tasks in addition to object naming are repetition, reading, action

(26)

spontaneous speech has received more attention, since it integrates several levels of language

processing and is a very naturalistic reflection of daily communication (Połczyńska, 2009;

Rofes et al., 2017a; Satoer et al., 2013, 2018). The use of spontaneous speech is mostly relevant

for monitoring speech and language functioning and the general well-being of the patient

during tumour resection and periods without DES (Kayama, 2012; Rofes et al., 2017a).

In Flanders and the Netherlands, some teams use the Dutch Linguistic Intraoperative Protocol

(DuLIP, De Witte et al., 2015b). This protocol provides standardized tests assessing different

levels of language processing and recommendations based on the current literature to select the

tests based on lesion location and to individually tailor them to the person with a brain tumour

(De Witte et al., 2015b). A shorter version of the DuLIP is also underway (A. Rofes, personal

communication, March 24, 2020). In what follows, used intra- and perioperative tests for the

assessment of semantic processing are discussed.

1.3.3.1. Semantic processing and language mapping

In several neurocognitive models of single-word processing, the semantic system containing

the meaning of words and concepts is considered the central component (Ellis & Young, 1988;

Rofes et al., 2015a; Whitworth et al., 2014). This is because, when impaired, both language

production and comprehension are affected, having a substantial impact on the quality of life

of a person with aphasia. This makes the assessment of the semantic system during awake brain

surgery necessary to avoid postoperative deficits (Campanella, D’Agostini, Skrap, & Shallice,

2010; Campanella et al., 2016; Goodglass & Wingfield, 1997; Hart & Gordon, 1990; Jefferies

& Lambon Ralph, 2006; Racine et al., 2015; Warrington & Shallice, 1984; Weitzner & Meyers,

(27)

were specifically designed to tap into semantics (De Witte et al., 2015b; Deloche & Hannequin,

1997; Howard & Patterson, 1992; Połczyńska, 2009; Visch-Brink, Stronks, & Denes, 2005).

These are, for example, object naming tasks that also tap into other language levels, such as

lexical retrieval and articulation (De Witte et al., 2015b; Deloche & Hannequin, 1997).

Object naming has for a long time been indicated as the ‘gold standard’ for language mapping

(De Witte & Mariën, 2013; Dragoy et al., 2016; Duffau et al., 2008; Kayama, 2012; Mandonnet

al., 2017; Rofes et al., 2017a). Furthermore, object naming speed has been indicated to correlate

positively to the quality of life of the person with a brain tumour and the ability to return to

work (Moritz-Gasser, Herbet, Maldonado, & Duffau, 2012). While this test includes the

assessment of semantic processing, other levels of language processing are also assessed, such

as picture recognition (i.e., identifying line strokes as an object) and lexical retrieval (i.e.,

retrieving the correct word form). Furthermore, this task does not allow to differentiate between

lexical and conceptual deficits based on the errors that are made (Rofes et al., 2019b;

Visch-Brink et al., 2005; Whitworth et al., 2014). This is because when an error is elicited during

DES, it can emerge in picture recognition, the semantic system, the output lexicon or

articulation processes. In Figure 4, an example of the object naming task included in the DuLIP

(De Witte et al., 2015b) is shown. The person with a brain tumour needs to say the lead-in sentence ‘dit is’ (this is) followed by the name of the object that is represented (here ‘dog’). The lead-in sentence or carrier phrase is used to distinguish speech arrest (i.e., the inability to

speak caused by seizure) from anomia (i.e., a word retrieval deficit) (Kayama, 2012; Ojemann

(28)

Figure 4 Example of a test item of the object naming task of the Dutch Linguistic Intraoperative Protocol (DuLIP;

De Witte et al., 2015b). The participant needs to say the lead-in sentence ‘dit is’ (this is) followed by the name of the presented concept (here ‘hond’ [dog]).

Several tasks have been developed that allow to specifically assess semantic processing (e.g.,

Howard & Patterson, 1992; Visch-Brink et al., 2005). The most widespread task to assess

semantic association is the Pyramids and Palm Trees Test (PPTT; Howard & Patterson, 1992).

In this task, a triad of words and/or pictures is presented and the person with aphasia needs to

point out the picture that is more closely related with the top picture out of two options. The

PPTT allows to mix words and pictures, but no psychometric data is available for this version

(Howard & Patterson, 1992). Also, this test has no normative data for Dutch, which is the target

population in the present study. An adapted Dutch version of the PPTT, the Semantic

Association Test (SAT), was developed by Visch-Brink and colleagues (2005) and consists of

a verbal and non-verbal version. This test allows to differentiate between a verbal and a visual

semantic disorder and gives indications for language therapy (see Figure 5 for an example of a

test item of the SAT). By virtue of providing four possible answers to which the person can

(29)

75%), while the PPTT can be responded correctly 50% of the time, even when there is a

language impairment.

Figure 5 Example of a test item of the Semantic Association Task (SAT; Visch-Brink et al., 2005) in which the

participant needs to point out the picture that matches with the object ‘glasses’ (i.e., ‘eye’). Two less related semantic distractors (i.e., ‘ear’ and ‘lips’) and an unrelated item (i.e., ‘elephant’) are provided.

While these tests allow assessing semantic processing, some remarks can be formulated for

their use in people with a glioma. First, both the PPTT and SAT were originally developed for

people with post-stroke aphasia. This is relevant as people with post-stroke aphasia tend to

show more severe language deficits than people with a brain tumour (e.g., Desmurget et al.,

2007; Papagno et al., 2012; Walker & Kaye, 2003). Following this, these tests are possibly not

sensitive enough to detect the mild language impairments in people with a glioma during intra-

or perioperative testing. If the SAT (Visch-Brink et al., 2005) or PPTT (Howard & Patterson,

1992) fails to identify a language impairment intraoperatively, there is a risk that important

structures for semantic processing will be missed resulting in possible new postoperative

(30)

two options for answering can encourage guessing and, therefore, a correct answer is possible

50% of the time. Furthermore, the degree of semantic similarity between the word pairs in the

PPTT and the SAT was not controlled for using objective measures. Especially for

cross-categorical word pairs (e.g., ‘cat’ and ‘basket’), the degree of association might vary

significantly across individuals. Finally, the SAT does not distinguish between different types

of semantic relations, such as taxonomic (i.e., based on shared features between concepts, e.g., ‘dog’ and ‘cat’) and thematic (i.e., based on the co-occurrence of concepts in the same event or scenario, e.g., ‘dog’ and ‘leash’) conceptual relations (see chapter 2.1.3. for more

information).

The consideration of all of these parameters is paramount, as studies of healthy individuals and

lesion studies have indicated a neural dissociation between thematic and taxonomic relations

(Chen et al., 2013; Kalénine et al., 2009; Schwarz et al., 2011). This dissociation relates to the

fact that thematic relations rely more on the temporo-parietal areas in the left hemisphere, while

taxonomic relations rely more on the anterior temporal areas in the left hemisphere (Chen et

al., 2013; Kalénine et al., 2009; Mirman, Landrigan, & Britt, 2017; Schwarz et al., 2011). This

differential pattern of activation was shown, for example, in an fMRI study by Kalénine and

colleagues (2009) in people without brain damage and was confirmed in a study with people

with post-stroke aphasia using voxel-based lesion-symptom mapping in which taxonomic (e.g., ‘pear’ when presented with the picture of an apple) and thematic (e.g., ‘worm’ when presented with the picture of an apple) object naming errors were related to these neural regions (Schwarz

et al., 2011). Other studies, however, did not find this differential pattern of activation (e.g.,

Peelen & Caramazza, 2012; Semenza, Bisiacchi, & Romani, 1992). For example, in an fMRI

study by Peelen and Caramazza (2012) activation in the anterior temporal lobe was associated

(31)

processing of the two types of semantic relations necessary for which DES could be a reliable

tool.

If there is indeed a neural dissociation between thematic and taxonomic processing, it is

important to take this into account during intraoperative language mapping. If a semantic test

only focuses on thematic relations, as is the case in the SAT, there is a risk of missing those

structures important for taxonomic processing. In sum, there is a need for a Dutch

intraoperative test differentiating between the two types of conceptual relations while taking

into account the rationale of intraoperative language tests.

1.3.3.2. Tests for subcortical mapping: the case of multimodal tests

If the SAT or PPTT is used for subcortical testing, a wide range of tasks would have to be

selected since these fail to assess several language levels, such as word retrieval and

articulation. This could significantly increase the duration of the awake phase, reducing patient

cooperation and increasing the risk of infection. This was shown in a study by Bello and

colleagues (2007) in which patient fatigue increased with longer mapping times. On the other

hand, when using a limited number of tests, not all language levels are assessed, which can

increase postoperative deficits (De Witte et al., 2015a; Rofes et al., 2015a). To solve these

issues, the use of multimodal or mixed tests has been proposed which in essence aim to map

multiple linguistic and non-linguistic functions within a short period (e.g., Coello et al., 2013;

De Witte et al., 2015a).

In a first attempt to fill up the gap for tests that would be specifically used on a subcortical

(32)

test follows the rationale of subcortical language mapping by rapidly screening several

linguistic and non-linguistic functions. In this test, items of several intraoperatively used tests

are mixed and displayed in an alternating manner. In the presented example of a QMT

paradigm in Figure 6, items of an object naming, calculation, repetition and line bisection task

are mixed. For example, the person with a brain tumour has to name an object in the first item,

in the next item, s/he has to complete a line bisection task. This test avoids increasing

intraoperative mapping time and missing subcortical structures important for linguistic

functioning. In the presented paradigm in Figure 6, the arcuate fasciculus can be identified with

the repetition task, the superior longitudinal fasciculus with the line bisection task, the angular

cortex with the calculation task and the object naming task can identify structures involved in

articulation, object recognition and semantic processing (e.g., inferior fronto-occipital

fasciculus and inferior longitudinal fasciculus). The rapid alternation between tasks, however,

requires a high level of cognitive flexibility and can result in “false positives”. Hence, this task

cannot be used in people with preoperative cognitive problems, as they would produce too

many errors without stimulation (i.e., false positives) (e.g., Giovagnoli, 2012; Satoer et al.,

2012; Taphoorn & Klein, 2004).

Figure 6 Example of a slideshow following the principle of the Quick Mixed Test (QMT; De Witte et al., 2015a)

(33)

‘crimineel’ [criminal]), object naming (e.g., ‘konijn’ [rabbit]), a calculation task (e.g., addition of 2 and 11) and a line bisection task in which the person with a brain tumour needs to point to the middle of a line.

While the QMT is a valuable first attempt in the development of subcortical language tasks,

the test requires a high level of cognitive functioning which can rapidly be exhausting for the

person with a brain tumour and excludes those who have a preoperative cognitive slowness

(e.g., Giovagnoli, 2012; Satoer et al., 2012; Taphoorn & Klein, 2004). Furthermore, the SAT

(Visch-Brink et al., 2005) which is currently used to evaluate semantic processing in people

with post-stroke aphasia is probably not sensitive enough to be used in people with a glioma

(Desmurget et al., 2007; Papagno et al., 2012; Walker & Kaye, 2003).

In an attempt to make a better and more sensitive tool following the rationale of subcortical

language mapping, the Text-to-Picture Semantic Association Task (TP-SAT) was developed,

assessing both thematic and taxonomic relations. Also, the semantically related word pairs

were controlled for using an objective measure. In the next chapter, the theoretical background

(34)

2. Chapter 2: The Text-to-Picture Semantic Association Task

(TP-SAT)

In the previous chapter, the details of intraoperative language mapping during awake brain

surgery were described as well as a review of commonly used intraoperative tasks. It was

emphasized that there is still a lack of standardized tests in Dutch looking into semantics as

well as of multimodal tests for subcortical language mapping. In an attempt to resolve these

needs, the present study considers the development, standardization and validation of the

TP-SAT.

The TP-SAT is a noun-task developed for native speakers of Dutch and consists of 50 items

and two practice items. The task is based on the design of the SAT (Visch-Brink et al., 2005)

and the PPTT (Howard & Patterson, 1992). For every item, a written noun is presented in the

middle (e.g., Oerwoud), surrounded by four black and white drawings (see Figure 7 for an

example of a test item). One of the drawings is closely semantically related to the written noun

and is the target word (e.g., ‘aap’ [monkey]). For each target word, a semantic distractor

semantically related to the written noun in a lesser degree (e.g., ‘varken’ [pig]), a phonologic

distractor of the written noun (e.g., ‘boer’ [farmer]) and a non-related distractor (e.g., ‘stoel’

[chair]) are presented (Howard & Gatehouse, 2006; Whitworth et al., 2014). The person with

a brain tumour is instructed to read the word in the middle out loud and name the drawing that

is closest semantically related to it. Following intraoperative language mapping, there is a time

(35)

Figure 7 Example of a test item of the Text-to-Picture Semantic Association Task with stimulus ‘Oerwoud’

(Jungle), target word ‘aap’ (monkey), semantic distractor ‘varken’ (pig), phonological distractor ‘boer’ (farmer) and non-related distractor ‘stoel’ (chair).

The purpose of the TP-SAT is three-folded. First, the task adds to the need for standardized

tests in Dutch (Rofes & Miceli, 2014). The test is based on the current neurocognitive theories

of language and semantic processing (see section 2.1.), is computer-based, standardized in a

population without neurological or language impairments (see Chapter 3) and controlled for

several word properties (i.e., concreteness, age of acquisition and word frequency) which can

influence semantic processing and lexical retrieval (Shallice, 1988; Whitworth et al., 2014).

The second purpose of the TP-SAT is to fill the gap for multimodal tests for subcortical

language mapping. Apart from semantic processing, the TP-SAT assesses several language

(36)

the four surrounding drawings), lexical retrieval (i.e., retrieving the target word) and speech

planning and articulation (i.e., naming the object). In contrast to the QMT, all these language

levels are assessed in the same item allowing to identify a language impairment within the

four-second timeframe without the high cognitive demands of constantly switching between tasks.

Finally, the TP-SAT differs in several aspects from the commonly used tests to evaluate

semantic association in a clinical setting. In contrast to the SAT (Visch-Brink et al., 2005) and

PPTT (Howard & Patterson, 1992), the word pairs in the TP-SAT are constructed with an

objective statistical tool to calculate semantic similarity using latent semantic analysis (LSA;

Landauer & Dumais, 1997). In this manner, interindividual variability in responses can be

reduced. Also, the TP-SAT assesses two types of conceptual relations, more specifically

taxonomic or categorical relations (e.g., ‘moon’ and ‘sun’ are both celestial bodies) and

thematic relations (e.g., ‘journalist’ and ‘newspaper’). These have been indicated to be neurally

dissociated, implying that only assessing one relation during testing could result in missing

eloquent structures for semantic processing (e.g., Chen et al., 2013; Kalénine et al., 2009;

Schwartz et al., 2011). Finally, the TP-SAT has one semantic and one phonological distractor

while the SAT has two semantic distractors. This allows participants to make different kinds

of errors (i.e., semantic and phonological) and provides room for mistakes (i.e., 75% of the

time). Since the TP-SAT was normed in a participant group without brain damage, it can be

more sensitive than the SAT (Visch-Brink et al., 2005) and PPTT (Howard and Patterson,

1992) to identify the subtle language disturbances in people with a brain tumour (Desmurget

et al., 2007; Papagno et al., 2012; Walker & Kaye, 2003). In what follows, the theoretical

(37)

2.1. Theoretical framework of the TP-SAT

2.1.1. Neurocognitive models of language processing

As stated in the previous chapter (see section 1.3.1.1.), an intraoperative linguistic test needs

to be theory-based (Rofes & Miceli, 2014). The design of the TP-SAT is based on the

single-word processing model of Rofes and colleagues (2019b) which is an adapted version of the

model of Whitworth and colleagues (2014) (see Figure 8). This model, in which every box

represents an independent cognitive function (i.e., a mental process needed to perform a task),

encompasses the production and comprehension of written and spoken words as well as visual

object and picture recognition. Central in the model is a shared lexical-semantic system in

which the conceptual representations of words are situated. The impairment of the

lexical-semantic system during DES can induce errors for both input and output levels of processing.

The model also contains several sublexical or non-semantic routes indicated with dashed lines.

That is, ways to produce or comprehend words which do not pass through the lexical-semantic

system. These routes are necessary for non-word repetition, non-word writing, and arguably

for naming without the aid of the semantic system. The latter is a theoretical question which is

unresolved and was revived some years ago with DES data (Gatignol, Capelle, Le Bihan, &

Duffau, 2004; Kremin, 1986). The added lexical-syntax component refers to agreement

processes, such as when the lead-in sentence in object naming ‘this is…’ results in an

agreement in the determiner and target noun. It is important to notice that a task is not

necessarily restricted to one function. Accordingly, the TP-SAT taps into multiple levels of

(38)

Figure 8 Neurocognitive model of single-word processing by Rofes et al. (2019b) based on the model of

Whitworth et al. (2014) with the addition of a direct connection between object recognition and the phonological output lexicon as has been shown with intraoperative DES.

(1) Oral word reading & comprehension: In the TP-SAT, the written word in the middle

has to be read out loud. Several components of processing can be distinguished. First,

discrimination of graphemes will occur during visual orthographic analysis. Second,

the orthographic input lexicon is activated for recognition of the visual word form as

familiar. Third, the meaning of the word will be retrieved from the lexical-semantic

system which is the storage of word meanings necessary to define a word. While it is

(39)

with a brain tumour needs access to perform the task correctly by semantically

associating the word in the middle to the target word. Finally, the word will be read out

loud by activating the phonological output lexicon which contains spoken word forms,

followed by the phonological buffer which contains sequences of graphemes. After

speech programming which converses the phonemes into neuromuscular commands,

the written word will be articulated.

(2) Object recognition and object concepts: The four surrounding pictures have to be recognized through several processes analysing their visual perceptual features (e.g.,

shape, colour) and binding these together for recognition in the structural description

system. This system provides access to the object concepts of the pictures. Here, object

concepts are defined as the non-verbal conceptual representations going beyond the

aspects needed to define a word and its retrieval from the lexical output lexicon in

contrast to the lexical-semantic system (Rofes et al., 2019b; Whitworth et al., 2014).

(3) Semantic association: For the semantic association aspect of the task, multimodal

semantic processing is necessary. The object concepts of the drawings have to be

matched to the written word in the middle through their connection with the

lexical-semantic component in which the word meanings of the nouns representing the

drawings are stored.

(4) Spoken word production: The object that is semantically related to the written

stimulus will be named using the same processing components as discussed in reading

out loud, more specifically the phonological output lexicon, phonological buffer,

articulatory programming and articulation.

Within this model, several word properties can influence task performance situated on several

(40)

controlled for to avoid “false positives” due to an increased difficulty of certain items. In what

follows, the word properties that were controlled for during the development of the TP-SAT

are discussed, more specifically concreteness, age of acquisition (AoA) and word frequency.

2.1.2. Relevant word properties

In the previous section, the language levels the TP-SAT taps into were described following the

single-word processing model of Rofes and colleagues (2019b). Within this model, several

word properties can affect task performance and naming latencies. According to the critical

variable approach of Shallice (1988), it is important to describe these word properties to avoid

errors that were caused by increased difficulty of certain items and to situate errors at a specific

level of language processing. When applied to awake brain surgery, this implies that it is important to control for these word properties to avoid “false positives” during intraoperative language mapping (Whitworth et al., 2014). It should be noted that interindividual variability

is possible since not every individual shows the same pattern of influence of psycholinguistic

variables on language processing (Nickels & Howard, 1995). In what follows, the three word

properties that are relevant in aphasiological studies (i.e., mostly in people with post-stroke

aphasia and neurodegeneration) and that have been considered in the standardization of tests

for awake surgery are discussed (e.g., Gisbert-Muñoz et al., 2020; Ohlerth et al., 2020; Rofes

et al., 2015a). These are word frequency, AoA and concreteness.

2.1.2.1. Word frequency

Word frequency or the number of times a word appears in a corpus can be situated at the

(41)

Grande, Meffert, Huber, Amunts, & Heim, 2011; Meschyan & Hernandez, 2002; Rofes et al.,

2019a-b). In people without brain damage, it has been indicated that high-frequency words are

retrieved faster and evoke fewer errors during object naming tasks (Barry et al., 1997;

Meschyan & Hernandez, 2002). These frequency effects have also been shown to affect

reading speed and accuracy both in people with and without dyslexia with longer reading times

for low-frequency words (Grande et al., 2011). This indicates that it should be avoided that a

target word of the TP-SAT has a lower frequency than its distractors.

2.1.2.2. Age of acquisition

Another relevant variable to control for is AoA or the age at which a word is learned in the

written or spoken form (Hernandez & Li, 2007; Rofes et al., 2019a). No consensus has been

reached about its place in the language processing model with both indications for the

phonological output lexicon and the semantic system (Alyahya, Halai, Conroy, & Lambon

Ralph, 2020; Cuetos, Herrera, & Ellis, 2010; Hernandez & Li, 2007; Whitworth et al., 2014).

When situated at the phonological output lexicon, AoA can influence lexical retrieval and is

negatively correlated with word frequency meaning that words with a higher AoA are more

difficult to retrieve (Johnston & Barry, 2006; Meschyan & Hernandez, 2002; Rofes et al.,

2019a; Whitworth et al., 2014). AoA has also been indicated to be situated at the semantic level

(Brysbaert, Van Wijnendaele, & De Deyne, 2000; Cuetos et al., 2010; Hernandez & Li, 2007;

Howard & Gatehouse, 2006; Johnston & Barry, 2006). In this case, later acquired words would

have less dense semantic representations, making them more vulnerable to a semantic

impairment (Cuetos et al., 2010; Hernandez & Li, 2007; Howard & Gatehouse, 2006). Also,

AoA has been indicated to affect object naming with faster and more accurate retrieval of words

(42)

Hernandez, 2002; Nickels & Howard, 1995). This makes it a relevant word property to control

for when developing the TP-SAT since the task assesses both lexical retrieval and semantic

processing.

2.1.2.3. Concreteness

Concreteness is the degree to which a certain concept gives rise to a perceptible entity (e.g., ‘book’ is a highly concrete word while ‘joy’ is not) and has been indicated to be situated at the lexical-semantic level (Howard & Gatehouse, 2006; Rofes et al., 2019a; Sandberg & Kiran,

2014; West & Holcomb, 2000; Whitworth et al., 2014). People with and without brain damage

have been suggested to perform better on tasks with concrete words (Sandberg & Kiran, 2014).

Also, the number of semantic errors has been indicated to be correlated to concreteness in

aphasic people with a semantic deficit (Howard & Gatehouse, 2006). This indicates that more

concrete items of the TP-SAT will be easier for a person with a brain tumour.

In sum, several studies suggested that word frequency, AoA and concreteness can influence

performance on the TP-SAT. Following this, when designing the TP-SAT, the target word and

its distractors were matched for these word properties. While the model of single-word

processing (Rofes et al., 2019b) provides valuable insights into the language levels the

TP-SAT assesses, it provides little information about the organization of semantic representations

which is the main focus of the TP-SAT. In what follows, two current neurolinguistic theories

The Text-to-Picture Semantic Association Task (TP-SAT): standardization, validation and application in people with a brain tumour