Pragmatic sociology and the digital worlds: towards an alternative laboratory
Francis Chateauraynaud and Josquin Debaz (GSPR – EHESS, Paris)
Read the first part of this text The challenges of a semantic and argumentative approach of corpus in the era of “big data”
The requirement for in-depth description and attentiveness to the details of action and argumentation is a constant driver in pragmatic investigations, which rely on a kind of reflexive accumulation of multiple cases. In our investigations, we have built up a collection of more than 50 corpora representing long run controversies on environmental issues.
This collective work seeks to build a convergence between:
- the exploration of large documentary networks,
- the construction of sociological models dealing with the ongoing evolution of actors and arguments,
- and the critical interpretation of contemporary processes by epistemic communities able to create anew kind of reflexivity on complex processes.
By allowing the dynamic interrogation of large evolutionary text databases, a semantic and argumentative approach helps us to redefine the relations between different ways of knowing1.
There are three main challenges. Firstly, economic theory has de-coupled from statistics,weakening the triangle between statistics, economic theory and political practice. Secondly, current management ideologies demand measurements of progress in the public sector, in a context where many dimensions cannot currently be quantified. Finally, while big data is seen as a solution to improving our understanding of how our social, economic and environmental systems intersect, this overlooks the need to refine data and ensure quality before drawing conclusions. We need to expand our ability to measure complexity in systems. (Radermacher, 2016)
How does Prospéro works?
There is not a central algorithm, but a long series of operations designed to realize the different translations between texts complexity and a space of computation based on sociological concepts:
- Indexation with 3 main characteristics : cumulative dictionaries ; distinction of few basic syntax types (entities, qualities, verbal relations, modalities, numbers, function words); constructing an internal representation of each texts as a series of sentences.
- Instantiation of semantic concepts and categorization.
- Distributing texts regarding contextual data (at least three: title, author and date).
- Producing statistical and relational structures, texts abstracts and specific properties (for instance ascription of qualities to the different entities).
- Identifying emerging topics and temporal distribution of concepts.
- Allowing user to compose complex objects like formulas (semantic regular expressions), discursive configurations and to sample his corpus in any sub-corpus he needs for producing relevant comparisons.
- Transfer results to other software (spreadsheets or R, geomatics, network analysis tools like Pajek or Gephi).
In order to produce comparative analyses of a large range of issues, specific tools have been designed for collecting, modelling and monitoring public affairs and controversies, by putting the focus on argumentative and semantic patterns. The proliferation of controversies and conflicts over scientific research, technologies or mega-projects, both at European and national levels, requires new methodologies for social scientists. After a huge “production detour”, we have implemented a digital architecture, not only for overcoming quantitative and qualitative divide (Chateauraynaud and Chavalarias, 2018), but also in order to design techniques based on a distributed cognition model.
Different procedures of knowledge are distributed in four tools, each one built upon a principle of symmetry between scripts or bots on one side, and human researchers on the other side. Prospéro does the main semantic and statistical work; Tiresias takes the role of web crawling, data mining and data crunching; Chéloné manages the access to the collection of corpora;and Marlowe, with its dialogic structure, based on natural language processing, plays the role of an e-sociologist, having heuristic interactions with researchers, and producing reports or chronicles.
The different relations between our digital entities combine manual operations and automatic transfer of data. This digital ecosystem has been designed to avoid the production of black boxes. Nevertheless,users can choose the level of vigilance and explicitation they need, finding a compromise between routine and reflexive feedback. The procedures developed share some key characteristics:
- They are founded on a qualitative and pragmatic approach of arguments and of actors. They involve data-processing tools designed to deal with the textual content of documents and to link the logical and linguistic analysis of statements to statistical analysis.
- A virtue of these techniques is to provide a long run perspective in order to monitor controversies and conflicts through time, as they develop, without defining a list of relevant documents in advance.
- A collection of corpus for various case studies is generated dynamically by a network of scripts and users; as well as building a memory of cases. Ideally, a collaborative approach provides concepts and semantic grids transposable from one corpus to others.
- Among software for qualitative studies, the specificity of Prospéro is to use syntactic and semantic collaborative dictionaries, and to transform them in analytical tools, including text generation by processing natural language reports.
But a question remains unanswered or opened: how do these epistemic formulas offered by the different programs fit the continuous challenges of the web transformations? A first answer lies in the effort to connect the research of information on the web to a sort of open laboratory. Based on both argumentative analysis and historical reconstruction, it provides a critical distance from an automatic use of the resources of the web, while maintaining an interpretive grasp on the real world.
A second dimension concerns the forms of knowledge mediated by artefacts that themselves connect interactivity and collaboration. Prospéro’s ecosystem tries to maintain a balance between human researchers and digital artefacts through the use of symmetrical protocols, without ever eliminating human interpretation from the loop. This point clearly appears in Marlowe scripting protocol. As an experimental sociological digital tool capable of conducting inquiries through discourse analysis2, Marlowe offers an alternative to usual “one-button solution”: the multiple and heuristic natural-language dialogues with users, or in a soliloquy3, creates a parallel lane of producing comprehension and knowledge,avoiding a complete dependency to dominant research engines.
To be continued
Chateauraynaud F and Chavalarias D (2018) L’analyse des grands corpus évolutifs et la sociologie pragmatiquedes controverses. Croiser les méthodes face aux transformations des mondes numériques. Sociologie et Sociétés. Forthcoming.
Madiot B. (2013). Analysing the French Abstracts from the International Conferences of Social Representations. Papers on Social Representations. 22: 10.1-10.23. http://psr.iscte-iul.pt/index.php/PSR/article/view/325
Radermacher W (2016) Systems, resilience, risks and accounting. In:Report of the EEA Scientific Committee Seminar on emerging Systemic Risks. Copenhagen, 24 February.
- Even if socio-Informatics has a particular history linked to the sociology of alerts and controversies, its efficiency has been shown in very different fields, as, among other applications, studies of social representations (Madiot, 2013).
- This capability is tested everyday with the writing of a press review, posted each evening on a blog Marlowe blogue
- Only in French at the moment…