Have you thought about exploring the field of "argument mining"?

Thank you for the hint! It seems that AI algorithms, at least on the longer term, will provide a solution to the problems of subjectivity in discourse analysis.

My favorite reference, so far, is this: Article Argumentation mining: How can a machine acquire common sense...

The problem that I found in argument mining is that it seems to assume a "closed digital dataset", like all the Twitter tweets that relate to Brexit. In reality, all argumentation seems to relate to a huge semantic network that extends from social media to all texts that have ever been written. In order to understand a Twitter argument you need to connect it to scientific, political and popular discourses, and each of these discourses is connected to one or more ideologies or philosophies that have emerged over thousands of years. It might be difficult for an AI alrogithm, for example, to note that a specific Brexit argument is actually the same argument that was presented by the ancient Greeks, who felt uncomfortable being part of the Roman empire 2000 years ago - and that both arguments are based on the same ideological or philosophical grounds.

It might be also too difficult for AI algorithms to analyze 1) the subject positions that different discourses suggest or impose to audiences and social actors, 2) the fallacious argumentation lines in which an argument is based, for example, on the position of authority of the person who is giving the statement, and not in any real arguments. Both of these tasks require "reading from between-the-lines", which might be difficult for AI.

Due to these challenges, it seems that we would first need a good manual process and methodology for analyzing memes, discourses and ideologies, and only after reaching that goal, we would be ready to automate that methodology. However, it would still be very interesting to experiment with the tools provided by argument mining and see, whether at least parts of the discourse analysis tasks could be automated!

I suggest using some stylometric methods; especially keyword and collocation analyses may be of help in all sorts of discourse research.

Pasi Malmi I had in mind analyses such as this one: Article Framing fracking: Semantic frames as meta-argumentative indi...

which pay attention to context and use rather diverse corpora to reconstruct recurring arguments in controversies. But I see your point: certainly algorithms can't do the whole job (which, as you describe it, is quite ambitious!).

Do you think, and this is really only a remark, it at all possible that the actual evolution of the above discourse is something like a (quasi-)organic process, in which new ideas come up as ad-hoc responses to earlier ideas? And our post-hoc 'analysis' of this discourse process would actually impose a (fictitious) linear (or, if you like, multi-linear) structure (or narrative) on this discourse set rather than discover its evolution? As you would find it in any type of historiographic writing?

Michal Místecký and Johanna Miecznikowski , thank you for very good hints! The problem of using these methods is that the study of the evolution of discourses, memes, paradigms, competences, technologies, words, proverbs, languages, norms, values etc etc has been studied from the point of view of

- Linguistics and AI (you know better than I do)

- General theory of evolution and memetics (Dawkins 1976)

- Philosophy of science (Kuhn & Lakatos)

- Economics (Nelson & Winter 1982)

- Administrative sciences (Aldrich & McKelvey 1983, Malmi 1988/1992/2009)

- Sociology (eg Althusser & Habermas)

- Anthropology and folklore (study of rituals, stories, decorations etc)

- History (eg the study of narratives)

Therefore the task of combining all of the best methodologies from these fields is an incredibly challenging one. Also, there seems to be a relatively difficult learning curve for an administrative scientists like myself for starting to use the methodology of linguistics and AI.

Pasi Malmi I understand your point; as I see it, it is not very difficult to get to know linguistic methods. If you wish, I can send you references to papers treating the topic. As to combination of methodologies, I think it is better to stick to one approach, as multiple viewpoints may confuse the research and lead to contradictory results.

Andreas Wirag, I agree with you that the emergence of new memes, discourses and ideologies normally happens as an "ad-hoc response" to earlier memes, discourses and ideologies. I have used a method, in which I first use my existing knowledge for making a model (a theory or a hypothesis) about what kind of earlier memes, discourses and ideologies have contributed to the creation of certain meme or decision (eg the decision of the Finnish government not to recommend facemasks in summer 2020). The next step is then to study texts to see, whether empirical reality matches my model, or whether I have to improve it based on the observations. This is a hermeneutic spiral in which the model of reality evolves step by step during empirical research.

I know there is a risk that my initial model would too much drive my search of empirical evidence, so that I only seek for data that will support the initial model. However, empirical reality will normally offer enough deviations from what the researcher expected. Therefore, it seems to me sufficient that the researcher is not intentionally ignoring conflicting evidence and alternative interpretations and models. Do you still see, that the methodological risks are too big - and can you propose any alternative approach? What did you mean by your comment of historiography?

P.S. A sociologist cannot call the evolution of discourses as "evolution of discourses" without taking a big risk of being labeled a user of an invalid biological analogy. The prejudice against evolutionary approaches, is so strong that sociologists do not normally notice that the term of Habermas (discursive element) is almost a synonym for a meme (Dawkins 1976), and discourses are almost synonyms for memeplexes.

I'm really not an expert on this topic, to be honest! But I know that historiographers, or, rather, historians who think about history-writing, have pointed out that their 'discovery' of history is always necessarily an 'invention' of this history. Each historian is unwillingly forced to impose a logical, chronological, linear order on history - which was never part of the actual, often chaotic unfolding of events as they really happened. So there are as many histories as there are historians, I suppose. But, at the same time, I am sure that a constant, hermeneutic comparison of model/theory and empirical data should lead historians - and semiotic scholars - to the best possible account of the actual unfolding of events (or memes).

Andreas Wirag, thanks for the clarification! I believe the strongest current among historiographers has been "narrativism" during the recent years. It means that there are an infinite amount of interpreting the historical events (and texts). However, newer approaches point out, that some interpretations are better and more convincing than others. A clear methodology for deciding, which interpretation is the most convincing, does not seem to exist yet (?)

Yes, thank you! This is exactly what I had in mind! In the attempt to linearise - through a clear-cut narrative (or a linear theory) - a sequence of natural events that are in themselves non-linear, or even spontaneously emerging as responses to previous events, there is no ideal, or most plausible, interpretation. All interpretations are equally limited in their (ultimately unfeasible) attempt to impose a linear structure on a non-linear unfolding of events. (Unless of course you choose to report the unfolding of these events themselves, which would be tautological). - Whether this is good news or bad news I cannot tell. (As a footnote, I am sure that humanities scholars would argue that any of these interpretations, or why it was chosen, is ultimately a product of the writer's ideology or of their personal interests).

Discourse and text: linguistic and intertextual analysis within discourse analysis .

Dear Read it would be helpful.

Dear all, I wish to thank you for your comments! Based on them, we managed to integrate some ideas of Norman Fairclough to the second edition of our research report, which is published here: https://agilepublishing.fi/artikkelit/article-10 (available also at Research Gate).