Sandrine Zufferey "Introduction to Experimental Linguistics"

1.2. Characteristics of experimental research

In this section, we will first stress the fact that experimental research must be based on a research question that makes it possible to formulate precise hypotheses. We will then see that in order to empirically assess a hypothesis, an experimental study must manipulate variables of interest while controlling other variables, which may influence the outcome of the experiment. Finally, we will discuss some methodological aspects of data collection, so that they can be analyzed through the use of statistics. These points will be elaborated in detail in the chapters dedicated to these different aspects.

1.2.1. Research questions and hypotheses

We have already emphasized that experimental research is part of a scientific process. It builds on existing knowledge in a research field and aims to increase such knowledge by studying a research question generated on the basis of an existing theory. A scientific research question identifies the potential cause for a phenomenon and postulates a cause to effect relation between the cause and phenomenon. For example, the question “how do we understand a text?” is not a research question, as it is too vague. Such a question corresponds to a general research topic, from which many research questions can emanate. On the other hand, a question such as “what is the role of memory in readers’ comprehension of a text?” is a research question that can be investigated empirically. This question identifies a cause – memory – and a consequence – text comprehension –, and establishes a relation between the two.

Once the research question has been defined, it is necessary to transform it into a research hypothesis, which corresponds to an empirically testable statement. In other words, the hypothesis must be confirmed or rejected on the basis of objective data. In order to do this, the research hypothesis must be operationalized, that is, it is necessary to specify which variables will be examined and how these variables will be measured, in order to collect relevant data for the experiment.

If we go back to our example above, memory is still a vague concept. As a matter of fact, a distinction is generally made between long-term memory, short-term memory and working memory. Working memory is a system that simultaneously stores and processes verbal elements (verbal working memory) or visual elements (visuospatial working memory). It is typically the verbal working memory that we use for reading, for deciphering and for putting together the words in a sentence. The operational hypothesis should therefore define what type of memory will be the object of study, verbal working memory, for example.

In the same way, the operational hypothesis should explain the way in which reading comprehension will be measured. Reading comprehension involves many steps, from deciphering words to relating these words in a sentence, and then to a text. Therefore, it is impossible to measure reading comprehension in only one way or with one type of experiment. We need to narrow down this notion to a more precise variable, corresponding to a process involved in reading comprehension that can be measured. For example, this could be the elements included in the readers’ representation of the text and stored in memory once the reading has finished. One way to assess comprehension would be to ask questions about the text at the end of reading and count the number of correct answers.

Let us look at a few more examples to understand what a research hypothesis is:

1В (1) Bilinguals have different cognitive abilities from monolinguals.

2В (2) Reading and understanding a text is difficult for children.

The above-mentioned hypotheses cannot be the basis of experimental research since they do not meet the criteria listed above. Their terms are too vague, they specify neither the cause nor the effect, and do not specify any measure to rely on so as to draw conclusions.

In order to be tested empirically, these hypotheses could be transformed into (3) and (4):

1В (3) Bilinguals perform better than monolinguals at a cognitive flexibility task.

2 (4) When reading a text, 10–12-year-old children draw fewer inferences than 14–16-year-old teenagers.

In these two examples we see that the vague terms used in (1) and (2) have been transformed into accurate terms in (3) and (4). Cognitive skills became performance during a cognitive flexibility task, and understanding atext became drawing inferences. By doing this, measures for quantifying the variables were defined. In addition, (4) specifies which groups will be included and compared in the study. Finally, both (3) and (4) indicate a clear relationship between variables.

In summary, a research hypothesis is based on existing knowledge in order to establish a relationship between two or more variables. It must also be operationalized, that is, clearly defining the measures that will be used for quantifying the variables being examined to verify the hypothesis.

The construction of a good research hypothesis is the result of different stages, among which the most important are conceptualizing the hypothesis, on the basis of knowledge acquired in the field, and then operationalizing the hypothesis. We will discuss the specific stages for conceptualizing a hypothesis in Chapter 6 (#uc8df27c8-0777-528b-815a-dd0c80557d4b), which is devoted to the practical aspects of an experiment. We will discuss the stages involved in the operationalization of a hypothesis in Chapter 2 (#u04c2cebf-2633-50e5-824e-a9e10e997cd2).

1.2.2. Manipulation of variables

Let us now go back to the example of the influence of working memory on reading comprehension. In this example, the variable verbal working memory can be observed in two ways. The first possibility would be to measure the skills of the people taking part in the experiment by using a verbal working memory test. According to this evaluation and its results, participants could be sorted into groups. By doing so, every participant is included under a variable modality (e.g. high competence or low competence) depending on his/her own characteristics, as some people have better working memory capacities than others. In this case, the variable is simply observed during research.

A second possibility would be to manipulate the variable verbal working memory, by implementing conditions within the experiment where this variable has different modalities. In our example, the manipulation of the independent variable would aim at restricting the use of verbal working memory in some of the participants, in order to see the impact of such manipulation on reading comprehension, as compared to other participants whose working memory has not been restricted during the reading. A common task used for manipulating verbal working memory is to ask people to momentarily memorize different series of letters while reading the text, to report them and then to memorize others. Having to remember a series of letters while reading the text reduces the verbal working memory storage capacity used for reading and makes it possible to show a connection, if existent, between working memory and comprehension.

In general, in experimental research, the aim is to manipulate all the variables involved in the hypotheses. However, due to practical or ethical reasons, this is not always possible. For example, age, socio-economic level, bilingualism, etc., cannot be manipulated because they are inherent in people. When variables can be manipulated, the decision to manipulate them, as well as the way in which to manipulate them, must follow ethical principles, ensuring that research will not harm the participants during the test. The cost/benefit relationship must be clearly considered when pondering the possibility of manipulating a variable or not. For example, imagine that you formulate a hypothesis stating that in stressful situations, people tend to speak faster than in non-stressful situations. In order to study the influence of stress on articulation rate, you could decide to manipulate the participant’s stress level. To set up a stressful condition, you could imagine putting some of the participants in a dark room in front of an audience booing at them. In experimental terms, such manipulation would be adequate, in the sense that a high level of stress would most likely result from your manipulation. On the other hand, it would be totally inappropriate from an ethical point of view. Actually, this type of manipulation would affect the participants to a much larger extent than needed, and they would probably not leave the experiment unscathed. Although this is an extreme example, it illustrates the fact that an experiment should not leave an impact trace on the participants once the experiment is over. We will develop this point in Chapter 6 (#uc8df27c8-0777-528b-815a-dd0c80557d4b), which is devoted to the practical aspects of an experiment.

1.2.3. Control of external variables

We have seen that when operationalizing research hypotheses, variables need to be defined with accuracy. The main purpose of such a definition is to isolate the variables studied within the experiment, in order to reach a reliable conclusion as to the relationship between them. In parallel, and for the same purpose, it is necessary to control the other variables, known as external variables, which could influence the variables and the results obtained in the experiment. External variables can be multiple and we will return to them in Chapters 2 (#u04c2cebf-2633-50e5-824e-a9e10e997cd2) and 6 (#uc8df27c8-0777-528b-815a-dd0c80557d4b), where we will discuss hypotheses and the practical aspects of an experiment. However, it is generally acknowledged that the characteristics of the participants are variables which may interfere with the variables investigated in an experiment.

Going back to the example of the influence of memory on reading comprehension, we may assume that educational level, general cognitive abilities, age, reading habits, etc., can influence both memory and reading comprehension. Likewise, the characteristics of the material used in the experiment may have an influence on the results. If, in the above-mentioned example, we use very simple text and questions, it is possible that everyone answers the questions perfectly well, regardless of their memory skills. On the contrary, if the text and the questions are very complicated, it is possible that very few people will be capable of answering. In these cases, we risk not finding a connection between memory and reading comprehension, not because the link doesn’t exist, but because the material used for the experiment is not suitable for evidencing such a link.

1.2.4. The notions of participants and items

To attenuate these potential problems, and to reduce the importance of the characteristics of the participants or the material employed, experimental research is based on data collected from a large number of people, using a broad palette of materials. Referring back to our example, it would be necessary to test a large number of people by means of a comprehension test. This test should contain multiple texts and different questions for each of them. In general, the material used in an experiment is defined as a set of items (the texts or the questions in our example are items). The ideal number of participants, as well as the number of items necessary to undertake proper research, is a complex question, which we will address in Chapter 6 (#uc8df27c8-0777-528b-815a-dd0c80557d4b).

Furthermore, experimental research is generally carried out by recruiting naive participants, who ignore the goals of the experiment and who have zero expertise in the subject under study. This precaution aims to try to control certain cognitive biases that could influence research results. The first bias is related to the fact that the participants who know the research hypothesis may try to base their answers on this hypothesis. Should this happen, the results obtained could suffer from what is called confirmationbias. Rather than answering naturally, participants could provide answers based on the hypothesis to confirm it, not because the assumption is correct, but rather because it seems adequate to them (even if this is not the case). The second bias is related to the fact that participants may want to help the researcher. If the participants know or suspect the goal of an experiment beforehand, the results obtained in this second scenario may not correspond to reality, but rather to the answers that the participants presume are expected.

Finally, in experimental research, participants are generally assigned to conditions in a random manner. This means that every person has the same chances of being included under one condition of the experiment or another. This random assignment offers additional protection against the effect of uncontrolled external variables. In addition to testing a large number of people, randomly distributing them to the different conditions reduces the probability that external variables could systematically influence the results. However, this random assignment is only feasible when all variables are manipulated. When one or more variables are simply observed, participants must be included in one condition or another on the basis of their own characteristics, such as gender or age, for instance. In this case, we speak of quasi-experimental research, since it is not possible to control all the variables. Leaving this question aside, experimental and quasi-experimental research is very similar, and the elements developed in the following chapters apply to both types of research.

1.2.5. Use of statistics and generalization of results

The last essential characteristic of experimental research concerns the way in which data is analyzed. Experimental research aims to collect quantitative data that can be statistically analyzed. As we will see in Chapter 7 (#u45fdd36d-d225-53e3-9134-bba5a7e08a6a), quantitative data can be described using different indicators, such as the mean, for example. Based on these descriptive indicators, it is possible to obtain an overview of the data collected, to summarize and illustrate them, in order to communicate the results with simplicity.

At the second stage, data is used to draw conclusions about the research hypotheses. In experimental linguistics, the aim is to study and understand a linguistic phenomenon for a specific population. Since it is impossible to test an entire population, researchers collect data from a representative sample. Through the use of inferential statistics, it is possible to determine whether the results of a particular sample are applicable to the whole population. This process is called generalization.

1.3. Types of experiment in experimental linguistics

Experimental research can be applied to all areas of linguistics, even if historically some areas have used such a methodology more consistently than others. Research questions vary widely between linguistic fields, meaning that many different methods and measures can be used in experimental linguistics. In this book, we do not aim to offer a detailed presentation of every research field and the methods associated with each, but rather to provide an overview of the principles of experimental methodology and the available techniques for linguists. Here, we will introduce some major classes of experiments that can be carried out in linguistics, and we will then develop these in every dedicated chapter.

In general, the experimental studies carried out in linguistics can be classified depending on the aspect of the language under study. Alternately, we will discuss studies on linguistic production and those relating to language comprehension. We will see that the study of comprehension poses many challenges, since this process is not directly observable. For this reason, research on language comprehension is based on the observation of indirect measures, which can be explicit or implicit. We will also see that it is possible to study comprehension by observing different stages of this process, either while it is in progress or once it has been completed.

1.3.1. Studying linguistic productions

The first type of linguistic experiment aims to investigate language production, all the manifestations of language that are produced by individuals in a certain language. Although these manifestations can be collected from diverse corpora and then studied through corpus analysis (see Zufferey (2020) for a detailed presentation of these methods), in some cases, the data contained in the corpus is not enough for studying a linguistic phenomenon. Some rare phenomena practically do not appear, if at all, in a corpus. What is more, the use of observation of naturally produced data is not suitable for showing the influence of a variable on the emergence of a specific linguistic phenomenon, as we have already seen. To counter this, different experiments can be implemented in order to study the production of linguistic phenomena. In these experiments, the goal is to purposefully elicit the emergence of certain linguistic structures, while controlling the context in which such structures appear. The experimental study of linguistic production will be described in further detail in Chapter 3 (#u757dc413-f351-5b1a-8315-0d05a70e5cca).

1.3.2. Explicit and implicit measures of comprehension

The second type of experiments used in experimental linguistics include studies conducted on the mechanisms involved in language processing and comprehension. Such processes are numerous and range from the organization of the lexicon, to the comprehension of a text or a discourse. It is therefore the most broadly studied aspect in experimental linguistics. Unlike some aspects of the production component, the language comprehension component is unique, in that it cannot be directly assessed through mere observation. It is outright impossible to directly observe the processes involved in the comprehension of a text, for example. This is why it is necessary to find a way to measure these processes indirectly, based on indicators that can be associated with them.

The first way of collecting these indicators requires the use of explicit tasks in which participants have to reflect upon certain linguistic aspects. For example, this is the case for metalinguistic tasks such as grammaticality or acceptability judgments. This type of task could be used to test the participants’ grammatical knowledge, by showing them syntactically correct or incorrect sentences in compliance with grammatical standards, and asking them to identify errors and justify their choice. While these tasks have the advantage of providing direct access to speakers’ knowledge, they also have the defect of being based on their reflexive skills and their subjective appreciation of their own understanding. These tasks are also particularly complex for certain types of people, especially for children or people with language impairments, for whom it is often very difficult to explain the reasoning behind their decision. Other tasks make it possible to circumvent these problems, by setting up experiments in which the participants have to choose between several illustrations matching a linguistic stimulus. For example, Durrleman et al. (2015) tested the comprehension of relative sentences in people with autism spectrum disorder (ASD), asking them to point to the image corresponding to sentences such as “show me the little boy running after the cat”. Making use of such tasks offers the possibility of studying language comprehension in children and populations suffering from linguistic impairments.

Alternatively, methods for studying comprehension in an implicit manner (without asking the participants directly for a judgment or an explanation of their reasoning) have also been developed. This is the case in action tasks, in which some kinds of behavior adopted on the basis of a linguistic stimulus can be observed. For example, Pouscoulous et al. (2007) tested the understanding of scalar implicatures triggered by words such as quelques (roughly equivalent to some), by asking French-speaking children to arrange tokens in boxes so as to match statements like “quelques cases ont des jetons (some boxes have tokens)”. It is also possible to understand comprehension skills using recall or recognition tasks, in which questions are asked at the end of a reading exercise or after listening to a text or speech fragment. For example, Zufferey et al. (2015a) tested the comprehension of causal relations in children aged 5–8 years, by asking them to answer why questions after every page, when reading a story with them.

1.3.3. Offline and online measures of comprehension

The various tasks listed above, as well as the tasks proposed in the examples presented so far in this chapter, enable access to comprehension once the word, sentence or text has been processed and understood. These measures are described as offline, in that they affect the final interpretations resulting from the comprehension process. On the other hand, online measures allow us to study the processes that come into play in comprehension itself. Such processes have the characteristic of being extremely fast, transient and occurring out of people’s consciousness, therefore remaining inaccessible to traditional offline measures.

Borrowing scientific methods and paradigms from other disciplines, such as psychology, has allowed the study of online processes involved in language comprehension. The majority of online measurement techniques have something in common: they observe the time required for a process, by measuring the reading time or reaction time. These techniques are based on the idea that the time required to complete a process reflects certain characteristics of this process, particularly in terms of complexity. Longer reaction times and reading times are generally associated with a more in-depth processing of the linguistic stimulus. Tasks using these time measures typically involve asking participants to name words, read or produce sentences, or decide whether or not a series of letters matches a word in their language. Studies that have employed such tasks have shown that, at the word level, response times and reading times are influenced by properties such as frequency, length and predictability. Similarly, at the sentence level, reading is influenced by properties such as syntax complexity or the need to produce inferences (Just and Carpenter 1980; Rayner 1998; Smith and Levy 2013).

Studies based on time measures have benefited from significant technological developments since the 1970s, so that today, anyone can easily conduct research from their computer. In addition, new techniques have been developed to enable the recording of eye movement whilst reading or when observing an image. It is thus possible to gain an insight, not only into the time required to read certain words or sentences, but also the exact movements made by the eyes during reading. This data provides additional information, such as the time allotted for different words, the order in which words are fixated or even the eye movements associated with reading certain passages. These eye movement measures can be applied to the study of reading as well as to the study of spoken speech production or comprehension.

Finally, the methods used in the field of neuroscience have also been transferred to experimental linguistics. These methods provide access to the brain activity involved in language-related processes. Using small electrodes placed on the scalp, the electroencephalogram (EEG) records the activity of neurons on the surface of the brain. This technique gives an accurate temporal overview of the activity of neurons associated with a specific linguistic process. Functional magnetic resonance imaging (fMRI) aims to measure the activity of neurons based on their oxygen consumption. It thus provides a precise spatial overview of the brain areas involved in a specific linguistic process.

As we can infer by reading these lines, offline methods are the most accessible to researchers, since they require few technical means. In most cases, offline measures can be collected using paper and pencil tasks. A simple spreadsheet available on every computer can be used for organizing and analyzing the data from such studies. For some statistical tests, a program must be added to the list of necessary tools. Online methods for observing reaction time or reading performance require special software for programming experiments. Things get more complicated when you want to record eye movements. These recordings require the use of expensive tools, that also take time to control. Furthermore, the data from studies on eye movements is much more complex to process. Finally, EEG or fMRI studies are generally reserved for people benefiting from access to such techniques, which are extremely costly in terms of equipment and necessary skills for processing recorded signals. For this reason, such techniques will not be discussed in this book.

Finally, we should point out that the offline and online measures do not provide answers to the same type of research questions. It is therefore important to consider them as complementary measures, which shed different light on the same phenomenon. There are no good or bad measures in experimental linguistics; the choice must be made on the basis of the goals and hypotheses of the research project. More and more often, offline and online measurements are used in parallel in the same study. We will return to these measures, their specific characteristics and the means for combining them, in detail, in Chapters 4 (#u3e247f59-c006-53a9-80b8-cc5d1e063a40) and 5 (#uee748d60-ee5c-5dda-a289-b01987d6379c).

1.3.4. Research designs and experimental designs

Whether for the purpose of studying production or comprehension, research can be categorized according to the general framework in which data collection takes place or, in other words, the experimental design. On the one hand, there are longitudinal designs, in which the same subjects are observed on several occasions, following varying time intervals. This type of design is generally used in studies where a variable cannot be manipulated, but its effect can be observed through time. For example, to study the influence of age on the ability to distinguish sounds between the different languages spoken in the environment of babies growing up in bilingual homes, one possibility would be to test the same bilingual babies at 2 months, 4 months, then 8 months old. Another example of longitudinal design would be the study of the relationship between language development and the development of theory of mind. In this case, language skills and individual differences in theory of mind could be measured in children aged 3 and a half, 4, and 4 and a half, for example.

The major interest of longitudinal studies is that they make it possible to observe changes in real time. However, they also have two significant disadvantages. First, such studies imply that participants must be tested on several occasions in relatively short periods of time. It is thus inevitable to lose participants during the study, due to motivation and availability reasons. Secondly, these studies generate significant costs, since it is necessary to find and then test people repeatedly, and above all, keep in touch with them and convince them to return to the following test sessions.

In order to work around these problems, cross-sectional designs observe different people, who are subjected to different conditions. To use the example of bilingual babies, instead of testing the same babies at different ages, we could simultaneously test groups of babies of different ages. This method would imply making a sort of picture of a situation at a given moment, which would offer indications on the relationship between age and sound perception in bilingual babies. Cross-sectional designs are typically used in quasi-experiments, where the independent variable is not manipulated.

When the independent variable can be manipulated, it is possible to allocate the participants to different conditions, in which manipulation can either be present or absent. Two types of experimental designs can be constructed in this case. In the first, the between-subject design, the participants only take part in one condition or the other. For example, to study the influence of reading goals on reading comprehension, one option would be to carry out an experiment in which a group of people reads a text in order to briefly summarize it, while another group reads the same text in order to answer questions about it. The performance of the two groups can then be compared during a recall task after reading the text. The results of such a task would certainly show that the second group performs better than the first group (as in Schmalhofer and Glavanov (1986), for example).

In the second type of experimental design, the within-subject design, also called repeated-measures design, the participants take part in all the conditions of the experiment. For example, such a design can be used in an experiment on the influence of word frequency on their processing time. In this case, each participant would see frequent words and infrequent words in order to cover all the modalities of the variable frequency. Among other things, this type of design makes it possible to control the external variables associated with the participants, given the fact that everyone falls under all conditions. Between-subject and within-subject designs each have advantages and disadvantages, which will be developed in Chapters 2 (#u04c2cebf-2633-50e5-824e-a9e10e997cd2) and 6 (#uc8df27c8-0777-528b-815a-dd0c80557d4b). For the moment, the main thing is to remember that there are many ways to organize research and that experimental research may adopt different designs, depending on the conditions under which the participants are tested.

1.4. Advantages and disadvantages of experimental linguistics

So far, we have shown how experimental linguistics is set within the scientific process and involves the use of quantitative methodology for studying language. On the basis of these principles, the results from studies in experimental linguistics are considered representative and can be generalized, unlike those from studies based on a qualitative methodology. The possibility of generalizing results is one of the strong points of research in experimental linguistics. However, this approach has a less positive corollary. Due to its empirical and quantitative nature, experimental linguistics needs to measure the linguistic phenomena it intends to study. While this can be relatively simple in some cases, the operationalization of complex processes (let us consider language comprehension, for example) implies the decision to observe certain indicators which could, at some point, not exactly measure what is desired. This issue will be discussed in more detail in Chapter 2 (#u04c2cebf-2633-50e5-824e-a9e10e997cd2). For now, it is important to keep in mind that just because something can be measured, this does not necessarily make it valid or reliable. Conclusions drawn on the basis of inadequate measures may not correspond to reality and may therefore lead to erroneous generalizations.

We have also seen repeatedly throughout this chapter, that experimental linguistics aim to identify the variables, also called factors, which can influence linguistic processes. For this, it is necessary to establish causal relations between variables by manipulating them. In this respect, experimental linguistics differs from another quantitative method, corpus linguistics, which aims to observe linguistic phenomena on the basis of natural data. While corpus linguistics can only account for a relation between some variables, experimental linguistics also makes it possible to explain the reasons underlying such connections between variables. These two methods are considered complementary, since they take place at different stages of the research process. Implementing a corpus study makes it possible to explore data and to uncover relationships between variables, relationships which can later be investigated experimentally, based on the hypotheses formulated as a result of data observation.

One of the prerequisites for establishing a cause and effect relationship between two variables, is the control of the conditions in which these variables are manipulated, as well as the identification of all the external variables that could influence the results. This necessity has the advantage of enabling solid conclusions as to the relationship between the variables, but risks keeping the experiment too far from reality. In certain comprehension experiments, such as reading tasks, for example, the sentences are presented word by word in order to measure the time allotted to each word. This way of reading differs enormously from natural reading conditions, where it is notably possible to go back in the text. For these reasons, experimental studies may lack ecological validity and not be completely generalizable.

This need for control also implies that each experiment can only investigate a specific hypothesis, in which every variable is operationalized in a certain way. For this reason, a specific experiment can only respond to a narrow research question. It is therefore essential to conduct a lot of different experiments to arrive at the comprehension of a phenomenon. Knowledge can then be built on the accumulation of experimental research related to such a phenomenon.

Beyond these limitations, the experimental methodology is a very important tool for linguistics, as it makes possible to study almost any research hypothesis. When linguistic phenomena are very rare or hardly accessible to the consciousness, it becomes essential for the construction of knowledge about these phenomena.

1.5. Where to access research on experimental linguistics

Before going further in this book, we offer a list of scientific journals publishing studies on experimental linguistics. As this is an extremely large and varied field, we cannot set up an exhaustive list of such resources. We will limit our choice to reputable journals in different fields of application. A large part of these journals originate from, or are related to, the field of psychology. As we have already seen, there is a close connection between psycholinguistics and experimental linguistics, due to the fact that they share common methods and measures. It is therefore unsurprising that studies in experimental linguistics are found in journals classified under the psychology section.

The following journals are excellent sources for finding research in experimental linguistics: Discourse Processes; Journal of Pragmatics; Journal of Phonetics; Journal of Experimental Linguistics; Applied Psycholinguistics; Second Language Research; Studies in Second Language Acquisition; Bilingualism: Language and Cognition; Cognition; The Quarterly Journal of Experimental Psychology; Journal of Memory and Language; Journal of Experimental Psychology: Learning, Memory, and Cognition; Language, Cognition, and Neuroscience; Behavioral and Brain Science; Psychological Science.

1.6. Conclusion

In this chapter, we first saw that the scientific process is based on the observation of concrete phenomena, whose systematicity enhances the development of explanatory theories. On the basis of these theories, it is possible to develop specific predictions which will then be tested, in order to refine or revise the existing theories. We then presented the difference between qualitative and quantitative approaches, in terms of the types of reasoning and possibilities of generalization. We also saw that quantitative research can adopt different types depending on the manner of observing the variables, as well as the control procedures carried out on them.

We then presented the characteristics of experimental research. Such research is based on a research question, making it possible to formulate clear hypotheses as to the relationship between two or more operationalized variables. In order to test these hypotheses, it is necessary to manipulate the variables involved in an experimental study, while controlling the other variables which may influence the results. In parallel, it is essential to test naive individuals, using numerous items, and to distribute the participants randomly under the different conditions. Finally, the data collected in an experimental study is mostly quantitative in nature, so that it can be synthesized and analyzed by means of statistical tests.

Studies carried out in experimental linguistics can examine linguistic production or comprehension. For the study of the latter, we have seen that there is a first differentiation between explicit and implicit measures, depending on the tasks. A second differentiation lies in the processes examined: while offline tasks focus on the results of comprehension, online tasks look into the comprehension processes.

Finally, we discussed the advantages and disadvantages of the experimental approach in linguistics, before suggesting useful resources for becoming familiar with this type of research.