After the Enthymeme Machine. A Reply to Walton and Reed's "Argumentation Schemes and Enthymemes"

In 2005 Douglas Walton and Chris Reed published an article in Synthese that opens with a wish and closes with a partial renunciation. The wish is the "enthymeme machine": a program able to read an argumentative text, such as a newspaper editorial, and pick out its unstated premises and conclusions. The renunciation concerns full automation. Such a machine would have to understand natural language, and the authors knew that capability was not available. Before going on, I invite readers to read Walton and Reed's article in full: what follows takes its ten cases and its vocabulary for granted. Here is how Walton and Reed described their "logician's dream":

For purposes of finding missing premises in incomplete arguments expressed in texts of natural language discourse it would be a logician's dream to have an automated enthymeme machine. The machine would be a software entity that could be applied to incomplete arguments in any chunk of discourse comprised by a natural language text, like a newspaper editorial. Assuming that there is some way of identifying arguments with existing premises and (if stated) a conclusion, the function of an enthymeme machine is to pick out the unstated premises and conclusions in these existing arguments. The task is one of identifying the commitments that could be ascribed to an arguer as a basis for posing critical questions about the argument, based on the given text of discourse. But automation of an enthymeme machine, if reasonable reliability and domain independence are required, becomes extremely difficult if the machine must be presumed to have a capability of natural language understanding. (Walton and Reed 2005, 340)

Twenty years later the missing condition has largely been met. Language models read an editorial, recognize its structure, and propose its unstated premises. For some time I have been working with one of these models inside SPES, a set of skills built for argument analysis and for teaching it. The acronym stands for Sistema di Procedure Esperte in Skill; in English, Structured Procedures for Expert Skills.

The purpose of this essay is to take Walton and Reed's program seriously and record what happens once their condition is satisfied. Some of the article's requests find an answer, others change shape, and one remains open. The work I describe owes a debt to that article, and I present it as its continuation.

To make the comparison checkable, each section includes a worked example based on one of the cases the article discusses. The schemes use the notation of SPES analyses: P for stated premises, PI for implicit premises, C for the conclusion, with the epistemic type and status of each premise in square brackets.