Human Measurement Techniques in Speech and Language Pathology

Methods for Research and Clinical Practice

Rietveld Toni

• Publisher: Routledge
• Year: 2020
• ISBN:
  9781000336450
• Edition: 1
• Topic: Psychologie
• Subtopic: Neuropsychologie

Human Measurement Techniques in Speech and Language Pathology gives an overview of elicitation methods in the assessment and diagnosis of speech and language disorders and explains approaches to the qualification of the obtained data in terms of agreement and reliability.

Despite technological advances in the assessment and diagnosis of speech and language disorders, the role of human judgements is as important as ever. Written to be accessible to students, researchers and practitioners alike, the book not only provides an overview of elicitation procedures of human judgement such as visual analog scaling, Likert scaling etc. but also presents methodological and statistical approaches to quality assessment of judgements. The book introduces statistical procedures for processing scores obtained in paired comparisons and in the context of signal detection theory, and introduces software relevant for the calculation of a large number of coefficients of reliability and agreement.

Featuring a wealth of reader-friendly pedagogy throughout, including instructions for using SPSS and R software, clarified by many illustrations and tables, example reports, and exercise questions to test the readers understanding, it is an ideal companion for advanced students and researchers in the field of speech pathology.

1
Measuring in speech pathology

1.1 What is measuring?

Measuring is a procedure to obtain useful information on objects and processes. A good example is provided by the so-called speech chain. The speech chain concept is well known among speech therapists and phoneticians. This concept expresses the fact that language and speech go through a large number of stages between ‘planning’ and ‘formulation’ of a message by the speaker on one side, and ‘understanding’ by the listener on the other (see Levelt, 1989; Gafos & Van Lieshout, 2020). After the message is formulated, muscle commands from the central nervous system are generated. The effects of these commands are the initiation of airflow, settings of the laryngeal system resulting in either phonation or non-phonation, and the movement and positioning of the articulators. The auditory system and the central nervous system of the listener process the acoustic signals in such a way that the listener understands the message of the speaker. Disturbances, however, can occur in each of the stages and substages of the speech chain. That is why both the clinician and the researcher have reason to carry out measurements in one or more of the stages that make up the speech chain.

In this context, three related terms play a role: ‘measure’, ‘measurement’ and ‘measuring’. They can be defined in the following way:

• Measure is a unit by which an object or process is measured. Examples of measures include Hz (hertz, number of oscillations per second) and hPa (hectopascal, pressure in weather systems or in the oral cavity).
• Measuring is the activity of obtaining data by using a measure. Examples include measuring pitch in Hz with a pitch algorithm and judging perceived nasality by using a scale ranging from 0 to 10.
• Measurement is the result of measuring.

The word ‘measuring’ may suggest the use of physical instruments and direct access to the object investigated or the process in question; however, that is not necessarily the case and is often even impossible. According to Stevens (1951), one of the early measurement theorists, measuring is ‘assigning numbers to objects on the basis of rules’. The set of these numbers (or, more generally, symbols) is called a scale. When a listener is asked to judge the speech sound of a speaker and says, e.g., ‘I rate this speaker as moderately nasal’, a measurement was carried out by a human observer. The human observer assigned a scale value to a stimulus, the object. The observer might have used an internally defined scale, such as one ranging from 0 (does not sound nasal) to 3 (sounds moderately nasal) to 6 (sounds very nasal). Numbers were used by the observer when giving her judgement. In this case, that means that higher numbers express higher degrees of nasality; that was one aspect of the ‘rule’. The rules used were possibly observer-specific, which means that a nasality rating of 2 given by Observer A is not necessarily the same as a rating of 2 given by Observer B.

Instrumental (‘objective’) measurements do, in fact, the same as raters. Instruments assign scale values to objects. Instrumental measurements are carried out on the basis of rules. For example, measuring the fundamental frequency of a speech sound entails the following rules: a) determine the fundamental period (T0) of a speech sound, measured in seconds, e.g. 0.01 seconds; b) divide 1 by this number of seconds (1/T0); and c) the ratio yields 100 Hz. There is a difference of 3 Hz between 100 and 103 Hz, and also a difference of 3 Hz between 4000 and 4003 Hz. From a physical point of view, these differences are equal. However, the difference of 3 Hz does not necessarily have the same perceptual effect at both levels (around 100 and 4000 Hz). A difference of 3 Hz around 100 Hz is well perceivable, but a difference of 3 Hz around 4000 Hz is not. This example illustrates that instrumental measurements do not always reflect perceived effects. Both instrumental and human measurements are essential in explaining the working of the human auditory system.

It is often suggested that the values or scores arising from instrumental measurements, even when carried out with different instruments or procedures, will always be the same. However, using an instrument like a pitch meter does not always guarantee that the same values will be obtained. An example of a speech signal realized after neck cancer may clarify this.

Creaky voice is characterized, among other things, by irregular voice pulses; see Figure 1.1. Some pitch meters allow different options for labelling speech as ‘voiced’; one of the parameters of these options is the (ir-)regularity of the voice pulses. Thus, with different parameter settings, one might obtain different voice decisions and consequently different measures of voice irregularity, called jitter.

Figure 1.1 Creaky voice: waveform showing aperiodicity in a speech signal realized after neck cancer. With thanks to Dr. Lisette van der Molen, Netherlands Cancer Institute, Amsterdam, the Netherlands.

This means that all measurements, be they subjective or instrumental, involve the following requirements: validity, sensitivity, reliability, and agreement. We use the term ‘object’ for all ‘things’ that have been judged or rated: speech samples, therapy conditions, etc.

1.2 Validity, sensitivity, reliability and agreement

Human judgements on speech and language properties and attributes can be obtained in many different ways. The choice is guided by a number of considerations:

• The informativity of the measurements: e.g. a two-alternative forced choice between ‘good’ or ‘bad’ is less informative than a scale ranging from 1 to 10.
• The ease of the task for the raters: e.g. giving an absolute judgement on a scale is often more difficult than a task which involves making paired comparisons such as ‘A is different from B,’ ‘A is better than B,’ etc.
• The cognitive load of the task: the extent to which cognitive processing resources, e.g. memory, are involved in the task.
• The statistical analysis of the data: the availability of procedures with which the data can be analyzed in such a way that the research question can be answered.
• The expected validity and sensitivity of the measurement instrument, the reliability of the scores (= measurement without errors) and the agreement between (for instance) raters.

1.2.1 Validity

Validity is an extremely important aspect of a measurement. The concept of validity means that a measurement should measure what it is meant to measure. Nasality provides a nice example. A well-known misconception is that perceived nasality can be assessed by measuring nasal airflow, i.e. the flow of air through the nostrils. This is not correct. It is merely the coupling of the nasal cavity to the oral cavities that results in the nasal quality of a speech segment. As a result, a device which only measures the magnitude of nasal airflow does not provide valid measurements of perceived nasality.

Seven types of validity (and two subtypes) are conventionally distinguished:

1. Face validity: this type of validity is quite obvious: does the instrument at first sight look as if it measures the domain adequately? For example, the percentage of stuttered syllables is not a good measure of severity of aphasia.
2. Construct validity: the association between the measurements which are meant to assess a domain and other measurements focusing o...