1 Research methods

Figure 1.1: A conceptual model of Research Methods (Hodotics).

Research Methods are about the theory, methods and practice of conducting research. I like to call the discipline that deals with research methods Hodotics, but it hasn’t caught on yet. One way of categorising different aspects of research methods is represented in Figure 1.1. You can think of research methods as the combination of:

The research process: the process of conducting a research project, from determining the research context to communicating results.
Project management: the process of managing a research project, from project planning to writing-up.
Digital skills: all research involves using computers (at least at some point if not throughout) so that computer literacy and digital skills are nowadays a fundamental aspect of research.
Philosophy: all research is not performed in a vacuum and a lot of philosophical questions shape the entire research process.
Ethics: all research is not performed in a social vacuum and ethical considerations are a fundamental aspect of research.

Let’s zoom in on the research process, as represented in Figure 1.2.

Figure 1.2: A conceptual model of research process.

Context: the research context includes several aspects of the research process, including the background (i.e. the previous literature and current knowledge) and the rational of the study (from the general topic to specific research questions/hypotheses).
Data acquisition: this is the process of gathering data to be used in the study. Data acquisition covers many different types of processes, from experimental set-ups to corpus queries.
Data analysis is the process of analysing the acquired data using qualitative, quantitative or a mixed methods. This part of the research process also include interpreting the output of such analysis.
Communication: finally, the last step in a research process cycle is to communicate what was done and what was learned, both to the research community and to the wider public.

1.1 Empirical research

Empirical research is one approach to research. This type of research focusses on learning about the Universe through data and observation.

Empirical research

The word empirical is related to experience, and in the context of research it basically means “based on experience (i.e. data and observation)”.

Lear more about the etymology of empirical here.

1.2 Axes of research

There are two main “axes” of empirical research types: exploratory vs corroboratory and descriptive vs explanatory.

Exploratory vs corroboratory research

Exploratory research is about exploring the data looking for patterns, associations, features and so on. This type of research is also known as “hypothesis generating” because exploration can lead to the formulation of new hypotheses.
Corroboratory research (aka as confirmatory research) is about checking expectations against data. It is also known as “hypothesis testing” because it is about testing hypotheses using data.

While there is still a lot of prejudice against exploratory research (typical sentiments are “it doesn’t have theory”) it is an important way of doing research, as recognised by important scholars. For example, Tukey (1980) stressed the importance of both approaches. The other axis of research is the descriptive/explanatory axis.

Descriptive vs explanatory research

Descriptive research is about describing facts through observation and collection of data. In other words, descriptive research is about the what.
Explanatory research is about explaining facts, i.e. understanding why they are the way they are. In other words, explanatory research is about the why.

Similarly to the exploratory/corroboratory axis, there is still prejudice against descriptive research (again, typical sentiments are that “it doesn’t have theory”). Within linguistics, several scholars have shown that both descriptive and explanatory research are fundamental and that both need conceptual and methodological theories to function. Indeed, Dryer (2008) talks about descriptive theories and explanatory theories, granting both the same status.

1.3 Research objectives

Orthogonal to the two research axes from the previous section, we can classify research instances based on their objectives. There are in principle three types of research objectives: establishing facts, improving the fit of a framework to the facts, and comparing the fit of difference frameworks to the facts. Each has its merits and to improve our understanding of the Universe we need all three, although there is nothing wrong for any one study to focus just on one or two!

Establish facts

Research can establish facts and fill a gap in the knowledge of one or more phenomena.

The aim of establishing facts is to accumulate evidence of particular events, features, associations.

Examples:

What are the uses of the Sanskrit verb gam ‘to go’?
What is the duration of vowels in Mawayana (Arawakan)?
Do people interact with AI as with other people?

Improve fit of framework to facts

Research can improve the fit of a specific framework to established facts. Usually this is done to fine-tune a framework in light of new evidence but it also just works when you want to test new expectations/hypotheses. When the facts do not match the expectations, researchers modify the framework to accommodate the results.

In some cases, a framework can be totally abandoned in light of the facts, or a new one could be developed.

Examples:

Strong exemplar-based models preclude the possibility of abstract representations, but certain categorisation tasks seem to involve abstract representations so these must be included in exemplar-based models.

Compare fit of different frameworks to facts

This objective allows researcher two compare two or more frameworks in light of empirical results. The main prerequisite for this approach is that each framework must have different expectations in relation to the phenomenon at hand.

When different frameworks entail different and exclusive hypotheses, one can test the hypotheses with data: the results might help excluding certain hypotheses and keep others. The frameworks that generate the excluded hypotheses have to be abandoned (unless they can be modified to fit the new results, see above, while still be different enough from other frameworks).

Examples:

There are two possible models for the bilingual lexicon: Word association and concept mediation. Which one better describes and explains the data?
A strict feed-forward architecture of grammar does not allow phonetic details to be sensitive to morphological structure, while some exemplar-based models allow that.

Each of the three objectives are important in research, but note that in order to really advance our understanding of things the third objective is fundamental: it is only by directly comparing different frameworks that we can accumulate knowledge and weed out inaccurate explanations. Every time you read about a study, ask yourself which of these objectives the study is setting to address.

Quiz 1

Select the appropriate research types for the following study: Previous research showed that in several Euroasiatic languages, vowels followed by voiced consonants tend to be longer than vowels followed by voiceless consonants. We investigate this tendency in Quechua.
Descriptive, exploratory. Descriptive, corroboratory. Explanatory, corroboratory.
Which of the following studies aims to improve the fit of a framework to the data? (Thanks to András Bárány for suggesting the second example)
We set out to test whether gestural timing is affected by foot-structure (as per the foot-sensitivity hypothesis) or not (as per the segmental hypothesis). In particular, we expect V-to-V timing to be stable within but not across feet independent of intervening segments if the foot-sensitivity hypothesis holds, while the timing should be affected by intervening segments both within and across feet if the segmental hypothesis holds. According to one hypothesis, there is one operation, Agree, which assigns Case features to an argument (accusative to objects, in the example) and at the same time gets the argument's person, number, and gender features. This means that in an English sentence like John sees her, her gets accusative case from a functional head (v) and v in turn gets the object's features — these are not spelled out in English; there is never any object agreement. Other languages are different in this respect: for example, in Hungarian, all direct objects have accusative case and the verb can show object agreement but it doesn't always do so. So there are a few theoretical options: either in all of these languages Case and agreement happen but case is not always realised (in English, case is sometimes realised but agreement never is; in Hungarian, object case is always realised, but object agreement only sometimes is), or Agree is actually not both Case and feature-agreement at the same time.