Evaluating ALEA: Testing a New Method for Spontaneous Speech Analysis of Glioma Patients in Spanish

(1)

Testing a New Method for Spontaneous Speech Analysis of Glioma

Patients in Spanish

Caitlin Holme

Student Number: 11312858

MA Thesis: General Linguistics (Clinical Track) Universiteit van Amsterdam

Supervisors: Laura Bos and Carolina Méndez Orellana Second Reader: Jan de Jong

(2)

Abstract

Background: Aphasia in glioma patients is a symptom which has received little attention

in recent years, despite affecting around 53% of patients with left-hemisphere brain tumours (Paratz, 2011). Modern advances in operative techniques and new

chemotherapy agents have led to substantial improvements in morbidity rates among brain tumour patients. In view of these advances, attention has turned to quality of life care, for which maintenance of communicative abilities is a vital consideration.

Differences between post-stroke and glioma-induced aphasia mean that current language tools are not sensitive enough for use with glioma patients.

Aims: The ALEA (Análisis del Lenguaje Espontáneo en Adultos) method was created for

use in Spanish-speaking people (Holme et al., 2017). This pilot study intends to review the method’s sensitivity and reliability by testing the spontaneous speech of healthy controls and glioma patients.

Methods and Procedures: The spontaneous speech of thirty-one healthy controls and

eight pre-operative patients was analysed. To assess test-retest reliability, two patients were also analysed post-operatively, and six controls were tested a second time.

Results: The results suggest that the ALEA has sufficient sensitivity to provide a useful

method in distinguishing between the spontaneous speech of pre-operative patients and healthy controls. Significant differences were found on the measures of MLU (Mean Length of Utterance), Inflection, Incomplete Utterances and Grammaticality. Subsequent case studies of two patients pre- and post-operatively and retests of six controls,

however, found sizeable differences between the first and second tests, which reached significance on the measure of Proportion of Nouns: Verbs.

Conclusions: Adaptations and further developments are suggested to improve

test-retest reliability. However, this pilot study provides promising initial results to support the use of a spontaneous speech analysis in addition to existing speech and language assessments of glioma patients.

(5)

1.

Introduction

The prognosis for people with brain tumours1_{has improved greatly in recent years, thanks}

to improved diagnosis and operative techniques. Patients are living longer and may be able to make a full recovery from what is a devastating diagnosis for the individuals concerned and their families. In accordance with these improvements, clinical considerations have adjusted to take into account not only prolongation of life, but also maintaining quality of life for patients (Miceli et al., 2012).

With this in mind, one aspect important to a good quality of life is effective communication. Language disorders as a result of brain tumours or operations to remove tumours are often overlooked by medical professionals (Paratz, 2011), yet loss of language (medically known as aphasia) can be very alarming for patients. One writer who kept a poignant memoir of his final months living with a brain tumour wrote “the fear of losing language is consuming me” and expressed anguish at how “no sentence is generated without effort” (Lubbock, 2010). It is therefore imperative that discussions of palliative care and treatment for glioma patients also consider day-to-day effects on their linguistic capacities.

Linguistic studies of aphasia and development of language tests have overwhelmingly focused on post-stroke language disorders (Paratz, 2011). Because of differences between post-stroke and tumour-induced aphasic symptoms, use of these same tests often fails to reveal abnormalities in glioma patients’ language (Satoer, 2014). Grammatical abilities are generally well-maintained, and thus patients with brain tumours perform well on standard aphasia assessments. In contrast, the modality in which patients profess most difficulty is taking part in everyday conversations, which requires flawless spontaneous speech (Satoer et al., 2013). This study therefore outlines a new method which has been developed to assess the spontaneous speech of glioma patients. The ALEA (Análisis del Lenguaje

Espontáneo en Adultos, Holme et al., 2017) aims to provide variables which will be sensitive enough to distinguish between glioma patients and controls.

This paper will therefore present an initial study of the ALEA method in practice, and in doing so attempt to answer the following research questions:

1) Is the ALEA method sensitive enough to distinguish deviations in the speech of pre-operative glioma patients when compared to healthy controls?

2) Does the ALEA method have sufficient test-retest reliability to be used to study pre- and post-operative differences in glioma patients?

The paper will be structured as follows: Section 2 is composed of a literature review, which will first present current research on language and the brain, and how this relates to patients with tumours, and then review currently available methods for studying

1_{The terms ‘brain tumour’ and ‘glioma’ are used interchangeably throughout this thesis. Section 2 will explain} the meaning of glioma in more detail.

(6)

spontaneous speech in aphasia patients. In Section 2 the parameters chosen and

instructions for analysis using the ALEA method will also be summarised. Section 3 outlines two hypotheses and makes predictions for the results of the study, and Section 4 describes the method used. Section 5 compares the results of the ALEA analysis on controls and pre-operative patients, and then results of two patients who were tested post-pre-operatively and six controls who were also retested. Finally, the discussion and conclusion in Sections 6 and 7 will deliberate on the implications of these results, and make recommendations for how the ALEA could be adapted in future studies.

(7)

2.

Literature Review

The first part of this paper will examine the current research which was used to form the basis of the ALEA method’s design. This literature review therefore consists of three sections. Section 2.1 will explore the link between brain tumours and language, thereby explaining why the ALEA is necessary. Current theories about language and the brain and the effects of brain tumours will first be explored, then recent developments in techniques for studying and treating aphasia caused by gliomas will be reviewed. Section 2.2 will then describe and review spontaneous speech analysis methods which are currently available for patients with aphasia, and adaptation of these methods for use in different languages and with glioma patients. This section therefore justifies the choices made in the development of the ALEA. The description of the ALEA parameters is in a sense a precursor to the Method in Section 4, however it is included in Section 2.3 of the Literature Review in order to

provide a rationale and motivation behind the choices which were made, in advance of its eventual publication in Spanish. Specific details of problems encountered and changes which were made for this study are also outlined in Section 4.2 of the Method.

2.1. Language and the Brain

Before turning to the effects of brain damage on language capacities, it is important to first explore what we know about the main areas associated with language in a healthy adult brain, and the linguistic role of the two hemispheres.

2.1.1. Language Localization in the Brain

Research into the modular localization of language in the brain dates back as far as the Medieval Cell Doctrine, developed in the fourth and fifth centuries AD (Whitaker, 1998). Since these theories combined scientific observations from autopsies with religious

teachings, the majority of this work has since been disregarded by the scientific community. However, the theory of specialized modules within the brain being responsible for

controlling aspects of human behaviour has endured to the present day, albeit in more refined theories. The researchers commonly cited as responsible for the most prevalent theories in the organization of language in the brain are Broca and Wernicke. Although the two areas of the brain traditionally considered most important for language are named after them, the history is in fact more complex. Broca’s infamous 1861 article was pre-dated by research by Dax, among others, who had already found evidence to support left-hemisphere lateralization of the brain (Whitaker, 1998). Similarly, Wernicke is often recognised for his discovery of the role of the superior temporal gyrus in language comprehension (Murdoch, 2010), although he himself acknowledges the influence of theories by his mentor, Theodor Meynert (Whitaker, 1998).

Regardless of who was responsible for the discovery of the left-hemisphere lateralisation of language in most people, and the importance of Broca’s and Wernicke’s areas, current research and neuroimaging techniques present a rather more complex picture. In what

(8)

Paratz (2011) calls the ‘textbook brain’, key language functions are located in two main areas in the left hemisphere, surrounding the Fissure of Sylvius. The anterior area, including the inferior frontal gyrus, mainly controls motor-speech language, and damage to this area can cause problems in planning and execution of speech. The posterior area includes the supramarginal and angular gyrus and superior temporal gyrus, and is important in comprehension, recognition and formulation of language (Murdoch, 2010).

Studies of aphasia in patients who have experienced a stroke or cerebrovascular accident attempt to link symptoms to a localized area of the brain, but these ‘pure syndromes’ are rare (Mesulam, 2010) and many forms of aphasia are difficult to isolate to a specific region. Recent years have brought improvements in neuroimaging techniques such as Positron Emission Tomography (PET) and functional magnetic resonance imaging (fMRI). With these developments, theories identifying restricted zones in the brain responsible for language have been questioned. Studies have revealed that almost every cortical region appears to play some role in language (Paratz, 2011). This includes the cerebellum, which has been found to activate during language tasks at the same time as the contralateral area of the dominant hemisphere (Méndez Orellana et al., 2015) and regions of the temporal, parietal and occipital lobes, which help to sequence auditory and visual representations of language into neural word representations (Brownsett & Wise, 2009).

2.1.2. Language Lateralisation

While linguists may still disagree about the exact location of many aspects of language in the brain, or even whether precise locations can be found at all, one fact which is widely accepted is lateralisation of language to the left hemisphere. Language disorders caused by strokes or cerebrovascular accidents have provided substantial evidence to support left-hemisphere lateralisation: Mesulam (2010) notes that in 90% of right-handers and 60% of left-handers, aphasia occurs only after damage to the left hemisphere.

Despite this, it is increasingly clear that brain modularity is rarely straightforward. Firstly, some aspects of language are located in the right hemisphere, notably understanding of prosody (Paratz, 2011) and auditory perception (Stemmer, 2010). Stemmer (2010) notes the new possibilities afforded by neuroimaging techniques to delineate a detailed profile of hemisphere lateralisation in each patient. This is particularly relevant to this study, as the ALEA was created in collaboration with Carolina Méndez Orellana whose work relates to the use of fMRI scanning to determine hemispheric language dominance in glioma patients (Méndez Orellana et al., 2015). She notes the difficulty of using fMRI results to determine lateralisation in glioma patients, since where the tumour is close to language areas, the mass of tissue can lead to false-negative fMRI results. In addition, Stemmer (2010) notes that in all patients with epilepsy, a common symptom of brain tumours, lateralisation tends to be more variable than in healthy people. For example, of nineteen right-handed glioma patients, Méndez Orellana et al. (2015) found that two had atypical lateralisation. One

(9)

Image 1: fMRI scan showing tumour in the left middle frontal gyrus. Reprinted with permission from Méndez Orellana et al. (2015).

possible cause of this is the relocation of language functions due to the presence of a

tumour in eloquent areas (Miceli et al., 2012), which will be discussed in more detail later in this review. Murdoch (2010) argues that regardless of hemisphere, any neuropathological disturbance which causes structural alterations in some portion of the brain is capable of leading to some form of communication deficit. This is very relevant in terms of brain cancers, the focus of the next section.

2.1.3. Gliomas: A Brief Overview

The incidence of brain tumours and other tumours of the Central Nervous System was almost 140,000 males and approximately 115,000 females worldwide in 2012 (CBTRUS, 2016). In 2017, it is predicted that in the USA alone almost 80,000 new cases of primary brain tumours will be diagnosed, of which approximately 32% will be malignant (ABTA, 2017). Since the early 1990s, tumours arising in the brain and central nervous system (CNS) have increased by almost a third in the UK (Cancer Research UK, 2017)

The diagnosis of a brain tumour has a devastating impact on both the patient and their2

family and friends. The disease typically carries a poor life expectancy, with the slowest growing tumours (Grade 1) prompting a prognosis of 5-15 years on average, and the most

malignant (Grade 4) typically leading to a quick deterioration and death within a few months of diagnosis (Espir & Rose, 1970). In recent years this prognosis has improved vastly as a result of better brain scanning technology, new chemotherapy agents and the option of awake

intraoperative mapping surgery. Despite this, as recently as 2014 cancer charities in the UK reported that just 14% of patients have an average survival rate of 10 or more years (Cancer Research UK, 2017).

Tumours are abnormal masses of tissue which can be benign, growing slowly and without invading other parts of the body, or malignant, in which case they infiltrate neighbouring tissues and structures (Espir & Rose, 1970). Image 1 shows an example of an fMRI scan of the brain of a patient who has a tumour in the left middle frontal gyrus. Tumours within the CNS are primary if they grow within the cranial cavity itself or secondary if they have travelled from a tumour in another part of the body to affect the brain (Murdoch, 2010). Secondary brain tumours are most commonly caused by breast, lung and melanoma cancers, and are more frequent than primary brain tumours (Faulkner, 2015). Primary brain tumours could be intracerebral (gliomas) involving cerebral tissues themselves, or extracerebral, arising from areas outside the brain such as the skull or

meninges (Murdoch, 2010). These cases of brain tumour are less common than gliomas,

2_{This study adopts the common practice of using gender-neutral ‘they/their’ in place of the singular form} ‘his/her’ etc. when gender is not identified.

(10)

which form 50% of documented primary brain tumour diagnoses (Faulkner, 2015). Gliomas are therefore the main focus of this study, since intracerebral tumours are the most likely to produce a disturbance in speech and/or language (Murdoch, 2010).

Brain tumours cause a variety of symptoms which may appear at different points in the progression of the disease, and are caused by various factors. For example, intracranial pressure can cause headaches, vomiting and swelling of the optical disc (Faulkner, 2015). Other symptoms are caused by the growing tumour compressing and damaging other areas of the brain, such as motor and cognitive impairment, and behavioural changes. Patients may experience alterations in their language capacities at this point, although there is a low likelihood that symptoms will be severe enough for them to connect this with the existence of a brain tumour. Irritation to the brain causes the final set of symptoms, including tremors, fatigue and epileptic seizures (Faulkner, 2015). Seizures are the symptom which most

commonly lead patients to seek medical assistance, by which time unfortunately the tumour may have grown extensively.

2.1.4. Gliomas and Aphasia

The term aphasia means an “acquired disorder of language caused by brain damage” (Mesulam, 2010, p49). In this sense, aphasia could be caused by any disease of the brain, including among others cerebrovascular accidents, strokes, brain haemorrhages and tumours. Despite this, aphasia as a term has traditionally been linked with post-stroke patients, who are known to often exhibit clear language difficulties. In contrast, aphasia caused by neurological cancers, such as gliomas, are an understudied etiology (Shafi & Carozza, 2012). It is reported that around 53% of patients with dominant hemispheric primary brain tumours suffer from some form of aphasia (Paratz, 2011).

As explained in Section 2.1.1, gliomas often have a poor prognosis in terms of life

expectancy, thus most research has concentrated on operative possibilities and end of life care, rather than the patient’s communication difficulties, which are often seen as a secondary problem. However, in recent years attention has been increasingly focused on language abilities, for two key reasons. Firstly, where a patient receives a poor prognosis and medical professionals have few options with regard to removing the tumour or

prolonging life expectancy, effective communication is a vital aspect of good palliative care. Paratz (2011) highlights the frustration which can be caused by a language deficit when trying to communicate with family and friends in final weeks. For this reason, Ford et al. (2012) suggest that a speech and language therapist is an important addition to a palliative care team. Among patients with a short life expectancy, it is still possible that better communication could extend prognosis. For example, a study on a group of patients with high-grade gliomas found that those with accompanying speech deficits had a median survival of 6 months, in comparison with 10.5 months among those without speech problems (Thomas, O’Connor & Ashley, 1995).

(11)

The second reason for increased interest in glioma-related aphasia is the overall improving trend in prognosis for brain tumours. This trend is attributed to a variety of developments, including new surgical interventions such as intraoperative awake mapping techniques, which will be explained in more detail in Section 2.1.6. This technique means that more of the tumour can now be resected, with less post-operative effect on language capacities for the patient. Also, new chemotherapy agents have permitted more effective destruction of cancerous cells (Faulkner, 2015). This has led to a new focus on quality of life for the patient after recovery, including the importance of preserving language function (Miceli et al., 2012).

Glioma-related aphasia can be caused by various factors. The most commonly recognised cause is infiltration of the tumour mass into other cerebral tissues, producing focal

destruction of language areas in the brain (Murdoch, 2010). Aphasia could also be caused by the operation to resect the tumour, in which surgical removal may require destruction of both grey and white matter which has been infiltrated by the tumour (Murdoch, 2010). Glioma-related aphasia is mild in the majority of cases (Paratz, 2011), and the most common type is anomia (word-finding difficulties) (Davie et al., 2009). Although categorised by

medical professionals as less severe than many other forms of aphasia, this does not constitute a small effect on the patient’s life. Any form of communication difficulty can be enormously frustrating. This is particularly true at a time when communication skills are necessary in understanding complex medical terms and explanations given by doctors (Paratz, 2011). Even patients who appear to have very mild aphasia often report problems in their spontaneous everyday conversation, something which is crucial in daily life. As Satoer et al. (2013) highlight, the production of spontaneous speech requires impeccable

performance at many levels of language comprehension and production. If just one level of this is disrupted due to brain damage, the patient’s spontaneous speech will also suffer in some way. It is thus vital that we understand deficits in spontaneous speech if we are to understand aphasia caused by brain tumours. In particular, it is necessary to distinguish between language difficulties which are provoked by strokes and by gliomas, which as we will see in the next section can lead to very different outcomes.

2.1.5. Post-Stroke Aphasia and Glioma-Related Aphasia: Similarities and Differences

Section 2.1.1. highlighted the main areas of the brain usually denoted to control language function. Many of these discoveries come from language disorders observed in patients who suffered a stroke, leading to a corresponding aphasia which is directly related to the specific area of the brain affected. Modern imaging techniques have meant post-stroke lesions and their related aphasias can be “meticulously delineated” (Paratz, 2011, p16). In contrast, in both pre- and post-operative glioma patients, the relationship between tumour location and type of aphasia is less straightforward. Miceli et al. (2012) report that patients with large tumours in areas considered critical for language may still have normal or minimally

(12)

impaired language function, and vice versa: for example, there are cases of tumours in non-language areas of the brain causing linguistic difficulties.

Two potential causes for this deviation in the conventional lesion-aphasia relationship have been suggested by researchers. Firstly, Miceli et al. (2012) highlight that vascular damage through a stroke or cerebrovascular accident has a brutal onset. In these cases, the suddenness of brain damage leads to language difficulties directly correlated with the language area which has been affected. In contrast, gliomas have a more gradual

progression and slower growth. Thanks to the brain’s neuroplasticity, this means that in some cases it is possible for language functions to be reorganised, so that destruction of brain tissues important to language does not lead to permanent aphasia in the patient. Paratz (2011) notes that for this reason, patients who have a stroke in the fronto-parietal operculum will suffer from aphasia, while a glioma patient with a tumour in the same area is unlikely to report language problems.

The second reason that glioma patients do not exhibit aphasias which can be so precisely localized as in stroke patients is that one of the symptoms of tumour growth is compression and distortion of other cerebral tissue, which can occur throughout the brain (Murdoch, 2010). Due to this, it is possible that a tumour occurring in a non-language area could still affect a patient’s linguistic abilities, as the sections of the brain relevant to language are compressed by the growth. Paratz (2011) further notes that location of the tumour is unimportant in aphasia diagnosis, as cancers in the brain produce a generalised hypometabolic state even in areas which are distant from the glioma.

Taking this information into account, Faulkner’s (2015) conclusion that the linguistic profiles of brain tumour patients are distinctly different from those of post-stroke aphasia seems logical. This point is relevant to discussion of the current techniques which are used in studying the language deficits of patients with gliomas, which will be examined further in the following two sections.

2.1.6. Current Techniques: Intraoperative Awake Mapping

As has been mentioned earlier in this chapter, in recent years the most significant

development in terms of improving both levels of patient morbidity and retaining language abilities is the use of awake craniotomy with intraoperative language mapping. The

operation involves use of local anaesthesia during the operation, so that the patient can retain consciousness and participate in language tests. During surgery, areas of the brain to be resected are electronically stimulated while language tests are performed. Where

stimulation causes a reproducible error at least three times, it is assumed to be a subcortical language site and thus this section is not removed (Santini et al., 2012). An error could be an anomic episode, where the patient forgets the word entirely, semantic paraphasia or total speech arrest. Intraoperative mapping has been proven to have a high success rate. In a

(13)

study by Bello et al. (2007), 94% of their patients had mild or no language deficits three months after surgery. Although 2% did show a moderate or severe impairment, the reason given for this is that subcortical areas identified as language sites still had to be removed during the operation in order to control bleeding.

De Benedictis, Mortiz-Gasser and Duffau (2010) studied a group of patients who were undergoing their second operation on a brain tumour which had previously been resected. In each case, the patient’s first operation was completed without and the second with intraoperative awake mapping. The results show that after the first operation (without awake mapping), three patients were unable to return to work, two of them due to subsequent aphasia, while after their second operations all patients returned to work. Previously, sections of the tumour found in eloquent areas were avoided as being crucial for language function, while with intraoperative mapping it is possible to remove more sections of the tumour even within so-called critical areas.

In order to identify appropriate candidates for awake craniotomy, pre-operative language assessment must be effective (Faulkner, 2015), since the findings of these tests can identify which tasks and stimuli will be most appropriate for use during intraoperative testing (Miceli et al., 2012). In addition, post-operative testing is essential in detecting any language

changes which may have been caused by surgery (Miceli et al., 2012). It is therefore vital that alongside gains produced by intraoperative mapping, language assessments are used to provide an extensive picture of pre- and post-operative language abilities. The following section will review current techniques used in completing these assessments.

2.1.7. Language Testing for Glioma Patients

Faulkner (2015) reports that there is substantial variation in the protocols used to assess the language of glioma patients, including self-report measures and formal aphasia assessments such as the AAT (Aachen Aphasia Test: Huber, Poeck & Wilmes, 1984). This is relevant since figures stating the number of glioma patients suffering from aphasia could be over- or underestimated, depending on which test is used. Estimates of the proportion of language deficits in this population indeed vary between 37% and 63% (Faulkner, 2015). In their study, Bello et al. (2007) did not use any standardized language examination as they found that normal batteries were unable to detect the mild aphasic deficits presented by patients with tumours. While traditional protocols are useful in assessment of post-stroke patients with more severe aphasia, they may not have enough sensitivity to detect differences in the speech of glioma patients (Satoer et al., 2013). It is therefore important that newer, more sensitive methods be developed which will be useful not only for patients suffering from more ‘classical’ forms of aphasia but also those with milder forms.

One such attempt has been made in a study by Faulkner (2015) and Faulkner et al. (2017). The authors developed a new method called the BLAST (‘Brief Language Assessment for

(14)

Surgical Tumours), designed to test the language and cognition of brain tumour patients. It is based on key language behaviours as identified by cognitive theories of language, for example articulatory-motor programming and sentence-level planning. By assessing

cognitive skills considered crucial for language production and comprehension, the goal is to maximise the likelihood of detecting even very specific linguistic impairments. It is intended to be a brief language assessment, taking around 45 minutes to complete, and thus

applicable in the clinical setting. Faulkner’s (2015) results show that 94% of pre-operative and 90% of post-operative patients were significantly different to healthy controls in at least one task. It therefore appears to be an applicable measure in assessing differences between patients and controls. However, it is also important to note that the BLAST includes tasks which are not necessarily specific to language function, for example the Stroop task, which tests the impact of information interference on reaction time. The inclusion of such tests may dilute the BLAST’s usefulness as an assessment of language function alone. In addition, the BLAST lacks any stimuli which include whole sentences. This is an omission if the

assessment intends to test the linguistic capabilities of patients, particularly in cases where the degree of aphasia is mild but still has an impact on effective communication. Therefore, while the recently developed BLAST method is useful, it seems that additional language testing is necessary to observe the full linguistic profile of a patient.

Although assessment of spontaneous speech is already done in most intraoperative

mapping sessions (Bello et al., 2007), inclusion of a spontaneous speech analysis in pre- and post-operative assessments is rare (Satoer et al., 2013). Edwards (1995) highlights the usefulness of studying patients’ connected speech, in determining a baseline from which future changes can be assessed. With this in mind, the current study develops and expands on the new ALEA method for spontaneous speech analysis (SSA) in glioma patients, which it is hoped can be used to provide more detailed linguistic profiles in future assessments. The next section will review current methods of SSA and explain the rationale behind the choices made in developing the ALEA.

2.2. Spontaneous Speech Analysis

As a method for measuring language ability, SSA is a relatively recent phenomenon, gaining popularity in the 1980s with a new focus on the daily communicative difficulties of people with aphasia rather than their specific impairments (Prins & Bastiaanse, 2004). It is useful for two key reasons: first, it represents the most accurate picture of what a patient is

capable of in their daily communication. Outside the testing room, a patient’s language may be very different to their language scores. A deficit in spontaneous speech is often the most frustrating aspect of aphasia for patients, given that spoken language defines “an

individual’s intellect, social life and personality”, allowing communication and sharing of knowledge (Lopez-de-Ipiña et al., 2015, p46). Secondly, spontaneous speech analysis has the potential to highlight problems which might be missed by other language tests, given that

(15)

during connected speech linguistic processing occurs at more than one level (Bastiaanse & Jonkers, 1998). Spontaneous speech analysis is capable of detecting even very subtle deficits in aphasia and other neurological disorders (Jaecks, Hielscher-Fastabend & Stenneken, 2012). SSA can also be used to provide baseline measures against which possible recovery post-stroke or tumour resection can be measured (Brookshire & Nicholas, 1994). Although SSA is a technique which has been widely “underrepresented” (Bastiaanse, 2011) by many speech and language professionals, there are still a variety of methods available. Aspects of these have been used and adapted in the creation of the ALEA. The following section will review the advantages and disadvantages of SSA methods which have been developed for English, Dutch and German. Their adaptability and current research into SSA use with glioma patients will then be analysed. This will be followed by some comments on the use of SSA across different languages, and the importance of testing psychometric properties.

2.2.1. SSA Methods: A Review

Various methods for Spontaneous Speech Analysis have been developed, most often for use in English and Dutch. This loose term encompasses methods which use rating scales,

conversation analysis and quantitative methods (Prins & Bastiaanse, 2004). Subjective ranking scales, for example the Functional Communication Profile, involve rating different communication behaviours of the patient (Prins & Bastiaanse, 2004). However, the subjectivity of these measures leads to a lack of consistency and poor levels of reliability (Edwards, Garman & Knott, 1993). Conversation analysis was developed in the 1980s and analyses speech in terms of pragmatics, for example repair in talk, turn taking and topic bias, rather than the language itself per se (Wilkinson, 2008). Though providing potentially interesting results about how people with communication disorders deal with interaction constraints, this is not the objective of this study. Therefore, the ALEA method is based on quantitative methods for SSA. A number of these methods are reviewed in this section.

Quantitative Production Analysis

One of the most influential methods for analysing spontaneous speech is the Quantitative Production Analysis (QPA) method by Saffran et al. (1989). The goal of their analysis was to create a clear, objective and detailed procedure which could be repeated across multiple studies with reliable results (Kong, 2016).

The QPA method uses semi-spontaneous speech elicited through the telling of a famous fairytale. The sample is transcribed and split into segments according to grammatical utterances, which are then analysed quantitatively on measures such as proportion of closed-class words, speech rate and sentence elaboration. It was originally developed for analysis of agrammatic patients, although Bird and Franklin (1996) found that the method was also useful in distinguishing people with fluent aphasia from healthy controls. Edwards (1995) describes the method as one of the most detailed grammatical analyses of connected

(16)

speech, and Rochon et al. (2000) report that its comprehensiveness has led to its use in various studies.

However, one problem with the QPA is the fact that it only logs occurrence or

non-occurrence of features, rather than their appropriateness. For example, if an inflection on a verb is present it is counted as inflected, even if it is not the correct usage. This is only useful in languages like English, for which omission of grammatical morphemes is a common error in aphasia (Abuom & Bastiaanse, 2012). In Spanish, it is more common for an inflection to be substituted, not omitted. (Martínez-Ferreiro, 2003) This error would not be registered through use of the QPA method. Also, the QPA works with what the patient intended to say only, and therefore discards repetitions, false starts and fillers. Edwards (1995) notes the importance of highlighting errors in aphasic speech as well as correct usage. In line with this, for the ALEA it was considered that errors can be just as informative as correct speech, and exclusion of large amounts of text may not be a useful approach.

Dutch SSA Methods

Many techniques for SSA originated in the Netherlands, where methods are widely used by clinicians for patients with aphasia (Prins and Bastiaanse, 2004). Prins, Snow and Wagenaar (1978) developed an SSA method for Dutch intended to be useful in analysing the

spontaneous speech of all types of aphasia, and particularly in differentiating between fluent and non-fluent speakers. Like a ranking system, their method scores spontaneous speech on 28 variables, although each of these is numerically calculated, rather than being subjectively rated. Categories include, among others, speech tempo (number of words produced in 6 minutes), phonemic paraphasias (as % of total number of content words) and function word substitutions and deletions. The original method used a speech sample of 6 minutes. This was later adapted by Vermeulen, Bastiaanse and Van Wageningen (1989) to use a fixed corpus of 300 words. This is a more reliable technique, since six minutes of aphasic speech could give vastly different numbers of words for a speech analysis

depending on the quantity of pauses and speed of speech of the patient. Brookshire and Nicholas (1994) also conclude that 300 words is a reliable length for study of spontaneous speech, hence this was the amount of words chosen for the ALEA method.

A method drawn up by a group of Dutch clinical linguists, VKL (‘Vereniging voor Klinische Lingui ̈stiek) is the ASTA (‘Analyse voor Spontane Taal bij Afasie’) (Boxum et al., 2013) This method is used in practice by clinical linguists and speech and language therapists in the Netherlands. It shares some aspects of analysis with the QPA and Prins et al.’s method, for example inclusion of number of paraphasias and neologisms, and a count of the number of finite verbs divided by total verbs. Additional measures include MLU (Mean Length of Utterance) and a count of modal and copula verbs. Like Vermeulen et al. (1989) the ASTA is based on a fixed corpus of 300 words. The manual reports normative values derived from 41 healthy Dutch controls (Boxum et al., 2013), to allow for comparison with the scores of

(17)

people with aphasia. The ASTA method provides the main basis for the measures used in the ALEA, since it is comparatively quick to administer and sets out clear instructions. The

description of the ALEA in the methods section will outline the key adaptations which were made.

Computer-Assisted Methods for SSA

It is worth noting that the drawback of all the methods outlined above, and including the one used in this study, is the time-consuming nature of spontaneous speech analysis. Even methods which reduce the number of measures calculated to a minimum still require the recording, transcription and segmentation of lengthy transcripts before analysis can begin. In addition, the methods tend to require working knowledge of linguistic terms. Given that assessments are completed by speech and language therapists and other medical

professionals who are not necessarily familiar with linguistics, improvements in computer-assisted technology are a welcome development.

Some computer-assisted methods have already been developed to analyse children’s speech, for example SALT (Miller & Iglesias, 2012). Once a transcript is prepared for SALT, the software automatically analyses the sample and provides output data, including tables comparing scores to standard measures of normally developing children. Holland et al. (1985) used an adapted version of the software to analyse a stroke patient with global aphasia, and were able to detect changes in the first weeks post-onset.

One method developed specifically for adult language is the ASPA (Aachner Sprachanalyse) (Grande et al., 2008). This method, tested on German adults with aphasia, automatically analyses transcripts on a number of categories including open and closed-class words, complete/incomplete clauses and simple/complex clauses. It should be noted that even with the technological assistance of this method, it is still necessary to transcribe and segment the utterances, and to mark them as complete, incomplete or elliptic.

A final method for computer-assisted SSA worth noting is used for patients with Alzheimer’s disease. Blanken et al. (1987) note that people with dementia have language problems which resemble fluent aphasia, including impoverishment of vocabulary and word-finding problems. A new method developed by Lopez-de-Ipiña et al. (2015) uses the speech analysis programme Praat to complete automatic speech processing. The results of this compare the speech of people with Alzheimer’s to that of healthy controls, and find a number of

differences on acoustic measures such as fluency and percentage of voiced and voiceless sounds. The authors of this study hope that automatic speech processing could allow for earlier diagnosis of Alzheimer’s patients.

Each of these methods carries certain advantages over traditional SSA in saving time and possible problems with interrater reliability through using computer-assisted programmes. However, in the case of this study, the methods’ lack of commercial availability and the

(18)

limited time and money available for developing the ALEA meant that conventional SSA methods were more appropriate as the basis for analysis.

2.2.2. SSA and Glioma Patients

The methods described in Section 2.2.1 were developed for use with patients suffering from aphasia caused by a stroke, or in the final case by Alzheimer’s disease. As explored in the first section of this literature review, glioma patients have more subtle language deficits which may not be apparent in normal language tests, including traditional SSA methods. In developing the ALEA it was therefore important to explore measures sensitive enough to distinguish glioma patients too.

There is so far very little research into the use of SSA for patients with brain tumours. Satoer et al. (2013) completed one of the first studies after noting a lack of available data

concerning characteristics of spontaneous speech in glioma patients, despite patients complaining about problems in everyday language. Their study attempted to find sensitive parameters for glioma patients pre- and post-operatively. Like the ASTA method, they used samples of 300 words which were elicited through discussion of topics like medical status, work and hobbies. The study revealed that even where spontaneous speech appears to be fluent, closer analysis shows deviations on several measures, including incomplete

sentences, self-corrections and repetitions (Satoer et al., 2013). These deviations reached significance pre-operatively for incomplete sentences and post-operatively for MLU and incomplete sentences. Tumour resection was not found to induce a substantial linguistic change in these results. The authors suggest that deficits found may point to a more general lexical problem due to word-finding difficulties. They conclude that a spontaneous speech analysis is a “valuable additional task in the domain of language” (Satoer et al., 2013, p690), shedding light on various different linguistic levels. Satoer et al.’s (2013) findings correlate well with the ALEA results, as will become clear in Section 5.

2.2.3. SSA Across Languages

Considering that the ALEA is developed for use in Spanish, it is worth making some

observations about the use of SSA in different languages. Virtually all aphasia research has concentrated on the aphasic deficits found in English-speaking patients (Benedet et al., 1998), a bias which affects the observations that can be made. It has already been noted that most SSA methods were developed for English, with some in Dutch and at least one (the ASPA) in German. Languages differ greatly in typology of grammar as well as lexicon, which is certain to lead to differences in how aphasia symptoms manifest (Benson & Ardila, 1996). For example, languages which allow null subjects such as Italian and Spanish have complex verb inflection paradigms, which means that bare stems of verbs are

ungrammatical. For this reason, it is extremely rare for people with aphasia in these languages to fully omit an inflection, as this would be grammatically impossible in their language. In contrast, it is common for English people with aphasia to omit inflections

(19)

entirely (Benedet et al., 1998). Bastiaanse, Edwards and Kiss (1996) note that Italians with aphasia use a greater number of relative clauses than English or German aphasia patients. However, this information is useless if we do not know whether this is a patient-specific phenomenon or if Italian people in general use more relative clauses. This highlights the necessity of comparing patients to controls who are native speakers of their language. Abuom and Bastiaanse (2012) propose that SSA could be used in languages for which standardised aphasia tests are not available, given that it is not necessary to translate any test items. They demonstrated the usefulness of this method in their comparison of agrammatism in English and Swahili. Rossi and Bastiaanse (2008) also used aspects of Saffran et al.’s QPA method (1989) in Italian, adapting aspects such as utterance

segmentation according to the typological specificities of the language. In line with this, although the ALEA method was developed for use in Spanish, it is hoped that the typological neutrality of its measures mean it could be used for other languages as well. Normative control data is only available for Spanish controls so any future study would need to obtain this data for each individual language.

A literature review found no standardised methods for SSA in Spanish, although analyses have been completed by Martínez-Ferreiro et al. (2017). Their study was focused on quantifying mixed aphasias which had no clear diagnosis, and concentrated principally on verb use. Therefore, the measures included were MLU, grammaticality, number of

embeddings and number of finite clauses. It is relevant to note here that Martínez-Ferreiro et al. (2017) found that both type-token ratio for nouns and verbs and the counting of copulas and modals was uninformative for Spanish, as patients were not notably different from controls. Based on these findings, it was decided that these categories would be omitted in the pilot study of ALEA.

Developing the ALEA method thus had two motives: the first was to create a protocol which would be sensitive enough to distinguish deficits in glioma patients. A secondary, but also important, reason for creating it in Spanish was the lack of standardised SSA methods available in this language.

2.2.4. SSA Methods: Standardization and Psychometric Properties

The ALEA method will be outlined in detail in the final section of this literature review, with each decision influenced or justified by some aspect of an SSA method explored in this chapter. However, it is first necessary to make a final brief observation about the importance of validity and reliability in linguistic tests.

When developing a test to be used in clinical practice there are various considerations which need to be taken into account, aside from the content of the test itself. One of these is the importance of test standardization. In line with the Dutch ASTA method (Boxum et al., 2013), the ALEA is standardized by analyses of healthy controls, which provide normative

(20)

averages and ranges against which patients can be compared. This means the ALEA is a norm-referenced test, which in aphasia studies signifies testing in comparison to a healthy population (Law et al., 1998). The other most common form of assessment is criterion-referenced tests, which concern how the participant performs against a specific criterion of mastery, for example on a scale or checklist (Haug, 2005). For aphasia, this would mean assuming a control score of 100% and giving the patients a score out of this total. The possible implications of choosing a norm-referenced test are discussed in reference to the results in Section 6.

Another important consideration in developing linguistic tests is assessing the psychometric properties of the assessment – specifically its validity and reliability. Validity means the extent to which the test assesses what it intends to. It can be further subdivided into sensitivity (rate of identifying true cases of language being affected) and specificity (rate of identifying true cases where language is not affected). The validity of the ALEA relies on its being able to distinguish between patient and control speech. This pilot study will test this by comparing the scores of healthy control participants and glioma patients. It should be noted that a lack of distinction between a patient and control does not automatically mean the ALEA is not valid. Whereas in most aphasia tools, we assume that all patients should exhibit some form of affectedness, and controls should not, here it will not be surprising if some of the patients act like controls. It was noted in Section 2.1 that around half of glioma patients have language difficulties, therefore the scores of many tumour patients could fall within normal ranges. This is not, therefore, an expression of the test’s lack of specificity. For this reason, the ALEA should ideally be used alongside results from more traditional tests such as the Boston Naming Test (Kaplan, Goodglass & Wintraub, 1983) as well as proper assessment of tumour grade and location, in order to determine the patients who are most likely to exhibit aphasia. Without this information, only general statements can be made about the ALEA’s validity.

The reliability of a test is its stability when used under different conditions or by different observers (Haug, 2005). It is usually assessed with test-retest reliability (giving the test to the same participants on two occasions and comparing the scores) or inter-rater reliability (two different raters scoring the same participant). For other SSA methods, Prins and Bastiaanse (2004) note that quantitative analysis is usually more reliable than instruments measuring communicative and pragmatic abilities, at least in terms of inter-rater reliability. They do however point out that “hardly anything is known about test-retest reliability” (p1087) for any of the SSA methods described in the previous section. Tommerdahl and Kilpatrick (2014, p291) argued that in many analyses there is a “tacit assumption that a given sample of language is a true representation of a person’s typical production”.

Therefore, they checked the test-retest reliability of spontaneous speech samples in normal adult speech on a variety of morphosyntactic variables. They discovered a lack of significant

(21)

correlations between the tests and retests across several items. Although many of these items were used with high frequency, these frequencies were not reliably maintained across the two tests. This suggests there may be some problems with reliability in spontaneous speech analysis methods.

In the case of the ASTA, which provided the main basis for the ALEA method, Boxum et al. (2013) give interrater reliability scores but not test-retest reliability values. A separate study by Wolthuis et al. (2014) did check the ASTA’s stability across two tests for patients with mild aphasia. They found that scores on the linguistic variables remained stable, and even a change in conversation topic did not provoke a significant change in language use. However, the emotional effect of the topic discussed did influence language, and therefore

recommend that the emotional charge of the topic remain stable throughout the interview (how this can be ensured in real-life assessments is less clear).

In the case of the ALEA, it is very important that test-retest reliability is stable if the ALEA is to be used to study pre- and post-operative patients, so that any changes can be reliably assumed to be due to the tumour resection and not random retest variation. As well as testing post-operative patients, this study will therefore also retest a selection of control participants to check reliability. Given the previous results on SSA, variation in scores will not be surprising, however it is hoped that the scores will give additional information about useful measures which can be used for future studies. It was not possible on this occasion to test inter-rater reliability, which will be assessed before final publication of the ALEA.

2.3. ALEA: Design

The final section of the literature review details the decisions made in designing the ALEA. The ‘Análisis del Lenguaje Espontáneo en Adultos’ (ALEA) method for assessing spontaneous speech in Spanish speakers was developed at the Speech and Language Therapy department of the Pontificia Universidad Católica de Chile, in a collaboration between Carolina Méndez Orellana and the author of the present study (Holme et al., 2017)3_.

The ALEA assessment method consists of four sections: 1) obtaining the speech sample, 2) transcribing the sample, 3) analysing the sample and 4) quantitative analysis, the results of which provide the final ALEA scores. All examples given in the description below are extracted from control and patient interviews recorded for this study.

2.3.1. Speech Sample

Speech samples should be elicited through a recorded interview of around 4-5 minutes for fluent patients and controls. The length of the interview is likely to exceed this for non-fluent patients who have a slower rate of speech, for example patients with Broca’s aphasia. However, although Santini et al. (2012) and Satoer et al. (2013) note difficulties in word

3_{The reference for the ALEA manual is cited here. It is so far unpublished, but the protocol will appear on the} Pontificia Universidad Católica de Chile’s website later this year.

(22)

fluency among glioma patients (naming words in a category against a time limit), the type of aphasia exhibited is usually mild in terms of general fluency (Paratz, 2011), and therefore 4-5 minutes should be sufficient.

The ALEA states that samples should consist of at least 300 words. The examiner should intervene as little as possible, only prompting with further comments where the participant is hesitant or the sample is too short. The interview questions are the same as in the ASTA (Boxum et al., 2013), with an additional fourth question in case a longer sample is needed:

1. (To controls) Can you tell me about your last visit to the doctor? (To patients) Can you explain what happened to you?

2. Talk to me about your family (example prompt questions: who makes up your family, what do they do, are they married, do they have children, etc.

3. What do you like to do in your spare time? Do you have a favourite hobby? 4. Tell me about what you do (or did) for work.

2.3.2. Transcription

Samples should be transcribed orthographically in Spanish, starting a new line each time the examiner or participant speaks, denoted by ‘EX’ for examiner and ‘PT’ for participant. The following markers are used for transcription:

• brackets for any repetitions or self-corrections. E.g.: PT: <vivo con mi> vivo con mis papas4

‘<I live with my> I live with my parents’ • XXX for unintelligible sections of phrases. E.g.:

PT: XXX se arregla todo

‘XXX everything is sorted out’

• & for false starts and other incomplete words (where at least 50% of the word is missing). E.g.:

PT: mi abuela tiene <&och> setenta y nueve años ‘my grandmother is <&eig-> seventy-nine’

• precedes filler words and other exclamations which do not add information to the sample. E.g.:

PT: *ya *eh al médico hace tiempo que no voy

‘*OK *eh I haven’t been to the doctor for some time’

4_{Translations are provided without glosses for Spanish phrases unless clarification is necessary due to a} typological difference.

(23)

2.3.3. Analysis

Once the samples have been transcribed, analysis words are extracted. The analysis itself is based on a sample of 300 words, within which neologisms, paraphasias and stereotypical speech are included. The ASTA method (Boxum et al., 2013) also includes repetitions, minimal responses, self-corrections and filler words within the 300-word samples. This was adapted for the ALEA, which does not include them in the 300, however they are still transcribed and analysed, as they are used to provide a measure of fluency. These categories are later added together to make up the Approximation Index, denoting the proportion of a person’s speech that is composed of meaningless circumlocutions and interjections.

Another important aspect of preparing the transcriptions is segmentation of utterances. Both the QPA (Saffran et al., 1989) and ASTA (Boxum et al., 2013) methods give limited guidelines for separating utterances, although this process can be complex. The ALEA method therefore gives a deliberately lengthy list of instructions, including clarification of what a simple, coordinate and subordinate clause is. As in the ASTA, the general rules to be followed when segmenting utterances are:

1) Following syntactical criteria, by identifying the main verbs of the utterance 2) Following prosodic criteria when syntax is not clear

3) Following the pattern of pauses. This third option is not preferable given that pauses can be common in language disordered speech, but the option may be necessary in the absence of other determining factors.

The rules of segmentation differ from the ASTA on some measures. For example, the ASTA states that phrases joined by the conjunction ‘and’ should be split into two utterances, while those joined by ‘but’ or ‘as’ should be kept as one. All three of these conjunctions mark coordinate clauses, and therefore these rules are quite unclear. Since the ALEA method is meant for use in clinical practice, where professionals may not have extensive linguistic knowledge, the manual attempts to use clear and simple terminology. Therefore, simple clauses with one verb and subordinate clauses are always counted as one utterance, while coordinate clauses with two verbs linked by conjunctions like ‘and, or, nor, so, therefore, but, however’, among others, are split into two utterances. For example:

• tuve que llevarles los exámenes ‘I had to take the exams to them’ Simple clause = 1 utterance

• fue horrible porque odio sacarme sangre

‘it was horrible because I hate having blood taken’

Subordinate clause = 1 utterance

(24)

‘and I have to do that every month/so it’s horrible’ Coordinate clause = 2 utterances

Once the utterances are separated, the transcripts should be analysed using an Excel spreadsheet which will be available with the ALEA manual. To produce the lexical analysis, the following must be counted:

1. The number of words transcribed and the number of words for analysis These categories are calculated automatically, as the number of words transcribed is simply the total number plus the ‘non-words’: repetitions, false starts and approximations. Both numbers are needed as they provide

information for different measures.

2. The number of repetitions and self-corrections, false starts and filler words per utterance.

3. The number of nouns per utterance – days of the month, proper names and paraphasias are counted, numerals and repetitions are not.

4. The number of verbs per utterance – all verb forms, conjugated or unconjugated, were counted. Combined tenses and verbal periphrases were counted as one verb. E.g.:

• quizás esperé mucho para ir al medico

‘maybe I waited a long time to go to the doctor’ = 2 verbs • eso puede ocurrir

‘that can happen’ = 1 verb (verbal periphrase) • me ha ocurrido

‘it has happened to me’ = 1 verb (combined tense) 5. The number of correct and incorrect verbs per utterance. 6. The number of paraphasias and/or neologisms

7. Non-phrasal/phrasal utterance – the utterances is marked 1 if it is phrasal, meaning it includes a verb, or 0 if it is non-phrasal, where no verb is present, for example phrases like:

• ¿y qué más? ‘and what else?’ • ‘difícil la pregunta’

‘difficult question’

8. Grammaticality – each phrasal utterance is marked for whether it is grammatical or not. This could be due to being incomplete or for general ungrammaticality, e.g. agreement errors.

9. Completeness – ungrammatical phrases are then marked for whether they are complete (all elements of the utterance are present) or incomplete (some part of the utterance is missing). E.g.:

(25)

• después te mandan para la casa pero

‘afterwards they send you home but’ = ungrammatical (incomplete) • pude pasarlo juntos

‘I could spend it together’ = ungrammatical (complete)

10. Subordination – finally, the utterances are marked for whether they include subordination. Utterances with subordination are marked as ‘1’ regardless of the number of individual subordinate clauses.

2.3.4. Quantitative Analysis

Table 1 lists the final measures which provide the ALEA scores, the calculation used to work out this figure, the SSA method which influenced this measure and the rationale behind its inclusion in the ALEA.

Table 1

Description of ALEA Linguistic Variables with SSA method and rationale as justification.

NB: MLU = Mean Length of Utterance; SSA = Spontaneous Speech Analysis; ASTA = Analyse voor Spontane Taal bij Afasie; QPA = Quantitative Production Analysis.

Linguistic Variable Calculation SSA Method Rationale

1. MLU No. analysis words/total no. utterances ASTA (Boxum et al., 2013)

Useful in distinguishing aphasia from control speech

2. Approximation Index

No. repetitions + false starts + filler

words/no. words transcribed Satoer et al. (2013)

Glioma patients deviated from controls on repetitions

3. Proportion Nouns:Verbs No. nouns/no. verbs

QPA (1989); Bird and Franklin (1996)

Useful for distinguishing people with fluent aphasia from controls

4. Inflection Index No. correct verbs/no. verbs

ASTA (Boxum et al., 2013), Martinez Ferreiro et al. (2017)

Distinguished between people with aphasia & controls; also relevant for Spanish patients. 5. No.

paraphasias/neologisms No. paraphasias + neologisms ASTA (Boxum et al., 2013)

Useful in distinguishing aphasia from control speech

6. % Non-phrasal Utterances

No. non-phrasal utterances/total no.

utterances N/A

No known previous study, but found to be a useful distinction during analysis

7. % Incomplete Utterances No. incomplete/total no. utterances

Satoer et al. (2013); Grande et al. (2008)

Glioma patients deviated from controls on this measure

8. % Grammaticality No. grammatical/total no. utterances ASTA (Boxum et al., 2013)

Useful to distinguish both people with aphasia and glioma patients from controls

9. Subordination Index

No. utterances with subordination/total

no. utterances ASTA (Boxum et al., 2013)

Adapted (index rather than total no. of

subordinate clauses); disordered speech usually has less subordination

(26)

3.

Hypothesis & Predictions

Section 2 highlighted the relevance of a spontaneous speech analysis for glioma patients. Compared with traditional language tests for aphasia, SSA is more effective in capturing very subtle deviations in speech. We can therefore draw a first hypothesis and accompanying prediction regarding the expected results:

Hypothesis 1: The ALEA method has sufficient sensitivity to highlight deviations in the

spontaneous speech of glioma patients, as compared to healthy controls.

Prediction 1: Comparing the ALEA results of healthy controls and pre-operative

glioma patients will reveal differences in patients who have some form of language deficit.

The wording of this prediction acknowledges the fact that not all glioma patients have a language deficit. As explained before, this factor makes it difficult to assess the specificity of the ALEA. It may be counterproductive to base evidence for language problems on standard aphasia assessment, since it has been noted that this lacks sensitivity for glioma patients. These tests do however provide evidence for any specific linguistic tendencies, for example problems with word-finding. Tumour grade and location are not appropriate predictors of language problems either, given the possibility of neuroplasticity and language

re-localization. However, a combination of this data and language scores would be enough to make an educated assessment of a patient’s language deficit, and thereby assess whether the ALEA is sufficiently valid to uncover deviation in individuals. In this study, language scores and detailed tumour information were not available. For this reason, the prediction is deliberately general. In order to answer the research question, it can only be assumed that any deviation found within the ALEA scores represents a language deficit, although the limitations of this approach are acknowledged.

The second objective of this study is to test the ALEA’s reliability as a measure to distinguish between the same patient pre- and post-operatively. In this context, it will be useful in determining any possible language changes which are caused by the tumour resection. The ALEA can be used to monitor changes over a long period of time in order to determine the overall improvement or deterioration of language post-operatively. To be useful in this context however, the ALEA method would need to have sufficient test-retest reliability. We can therefore draw a second hypothesis and prediction:

Hypothesis 2: The ALEA method will have sufficient test-retest reliability to be useful

in examining differences in the same patient pre- and post-operatively

Prediction 2: The control tests and retests will be sufficiently similar to ensure the

ALEA’s test-retest reliability. Therefore, deviations in the pre- and post-operative case studies can be interpreted in relation to patients’ resections.

(27)

Etiology Gender Age Education Hemisphere Location CDP F 24 14 L Frontal RJA M 51 12 R Fronto-parietal GCHF F 44 12 R Temporal CVV F 31 16 L Parietal CRJ M 24 14 L Parietal SEM M 31 12 R Frontal SSP M 48 14 L Frontal JBF M 58 8 L Parieto-occipital

4. Method

4.1. Participants

In this study, the spontaneous speech of eight pre-operative glioma patients and thirty-one healthy controls matched for age, gender and educational level5_{was analysed. The}

summary demographics of the patients and controls are shown in Table 2:

Table 2

Table showing demographics of participants’ gender, age and years of education

The inclusion criteria for patients and controls was to be aged between 18 and 85 and be monolingual in Spanish. A summary of the lateralisation and location of the patients’ tumours is shown in Table 3: three patients had gliomas located in the frontal area, two parietal, one fronto-parietal, one temporal and one parieto-occipital. It is important to note that three of the eight patients included in the study have right-hemisphere tumours (see Table 3). Although traditional aphasias are associated with damage to the left hemisphere, it was noted in the literature review that for glioma patients there is greater variation, and tumour location may have little influence on the type of aphasia exhibited. For this reason, it was decided that patients with right-hemisphere tumours would also be included in the study, although additional analyses were completed to see whether the tumour

lateralisation had any effect on results. All patients included were right-handed, as determined by the Edinburgh Handedness Inventory (Oldfield, 1971).

Table 3

Table showing gender, age, education and hemisphere and location of patients’ tumours

5_{Eleven patients were initially matched, but due to three having arteriovenous malformations rather than brain tumours,} they were later removed from analysis. Due to this, the groups are less closely matched on gender and education.

Group N Male Female Mean SD Mean2 SD2

Patient 8 5 3 38.9 13.0 12.8 2.4

Control 31 12 19 44.0 17.8 14.8 3.3

(28)

Since the data for controls and patients were gathered and analysed as part of an ongoing project by Carolina Méndez Orellana (2015) the criteria for inclusion and exclusion of patients and controls were the same as in her study. The exclusion criteria were the following:

• Severe developmental dyslexia • Severe hearing deficit

• Severe perceptual visual disorders • Recent psychiatric history

• Severe motor disability

• History of neurological disease

Patients were recruited through the outpatient clinics at the Hospital Clínico Pontificia Universidad Católica, Complejo Asistencial Hospital Sótero del Rio and Complejo Asistencial Barros Luco, all located in Santiago, Chile. Healthy control subjects were found among friends, family and colleagues of the researchers.

Patient participation in the study was subject to requirements of the medical ethics committee boards of each centre: Comité de Ética Servicio Metropolitano Sur Oriente (for Sótero del Rio), Comité de Ética Servicio Metropolitano Sur (for Barros Luco) and Comité Ético Científico de la Facultad de Medicina, Universidad Católica, which also approved control subjects’ participation in the study. All participants were given documents detailing the study and were required to sign informed consent forms before participating. In

addition, any control participants over the age of 65 were tested with the Mini Mental State Examination (Folstein, Folstein & McHugh, 1975) with a required score of at least 24 out of 30 for inclusion in the study.

4.2. Testing

Testing was completed according to the instructions given in the ALEA manual, as explained in Section 2.3. All recordings were made on a digital voice recorder and saved under

anonymised Study IDs to protect the participants’ identities. Interviews took place at the university, in the hospitals where patients were admitted or in the participants’ homes. Recordings were subsequently transcribed and analysed by the author of this study according to the ALEA method. Although recordings were originally obtained from

additional patients, a number of the samples could not be used as they failed to reach the 300-word threshold. In her study, Méndez Orellana (2015) includes patients with

arteriovenous malformations in addition to patients with gliomas, as she intends to assess how other brain lesions may affect the bold fMRI signal. Therefore, three patients with arteriovenous malformations whose spontaneous speech had originally been included in this study were removed from the analysis, since their results are not relevant to studying gliomas.