On the role of prototypes in appellate legal argument (abstract)

A history of AI and Law in 50 papers: 25 years of the international conference on AI and Law

“…His main point here is that a "pure exemplar-based" theory of precedent would have to be consistent with the Goodhart view of ratio decidendi, and therefore subject to the critique of Cross in a case such as Bourhill v Young. Four approaches to exemplar-based reasoning are considered: (i) structural similarity [134]; (2) dimensional analysis [16]; (3) the nearest neighbor classification rule [172,266]; and (4) my own prototype-plus-deformation model [178]. For each approach, Karl argues that the authors of these studies have adopted the Goodhart view, implicitly and unavoidably, thus making it impossible to represent the distinction that Cross wants to make in Bourhill v Young.…”

Some arguments about legal arguments

“…Sridharan [Mc-Carty and Sridharan, 19801 [McCarty and Sridharan, 19811 [McCarty and Sridharan, 19821. There w&s a long hiatus through the late 1980's and early 1990's, during which time I worked primarily on knowledge representation problems, but I returned to the subject again two years ago [McCarty, 19951. The problem all along has been to build a computational model of the arguments that occur in the majority and dissenting opinions in a series of corporate tax cases in the United States dating from the 1920's and 1930's, and beginning with Eisner v. Macomber, 252 U.S. 189 (1920).…”

“…The problem all along has been to build a computational model of the arguments that occur in the majority and dissenting opinions in a series of corporate tax cases in the United States dating from the 1920's and 1930's, and beginning with Eisner v. Macomber, 252 U.S. 189 (1920). The reader should consult the original papers, especially [McCarty, 19951, for the full story. These papers are dense with technical detail, however, and it may be difficult to see in them my overall thesis about legal argument.…”

“…Second, machine learning algorithms are based on very simple data representations: e.g., feature vectors, for most algorithms [Quinlan, 19931, or Horn clauses, for the most complex algorithms yet discovered [Muggleton, 19921. But to adequately represent the cases in a legal knowledge base, we need an extremely rich knowledge representation language [McCarty, 19891. How is it possible to learn anything in such a complex environment?…”

“…Liiwise, coherent concepts should be (i) easy to compute with and (ii) easy to learn. The trick is to formalize these ideas within (a) a knowledge rep resentation language that is expressive enough for the legal domain [McCarty, 19891, in which (b) the problematic concepts are defined by prototypes [McCarty, 19951. In my 1995 paper, I extended my formal model of legal argument up to the edge of the theory of coherence, but no further. The situation is essentially the same today.…”