Mystery Number Two - Computational Linguistics and Talking Robots - page 34

Representation 2.1.1, inspired by de Saussure, is made more specific in 2.6.1
by stipulating that the surface is agent-internal, and made more general by us-
ing lexical analysis to represent the meaning plus the morphosyntactic prop-
erties. The formats 2.1.1 and 2.6.1 are alike in that they use graphical means
(vertical line) to indicate the conventional connection between the surface and
the lexical properties.
The lexical proplet in 3.1.1 differentiates the analysis further by using the
format of non-recursive feature structures with ordered 2 attribute-value pairs
( avp s). This simple format is (i) suitable for storage and retrieval in a content-
addressable database, and (ii) allows us to code any interproplet relations, any
morphosyntactic properties, and the convention-based (unmotivated) connec-
tion between the internal surface and the lexical analysis. 3
In a proplet, the lexical and the compositional aspects of a word form are
systematically distinguished by means of characteristic attributes. For exam-
ple, the lexical aspect in the proplet shown in 3.1.1 is represented by the values
of the first four attributes: 1. [sur: eau] specifies the surface ;2. [noun: water]
consists of the core attribute specifying the part of speech, and the core value
specifying the literal meaning represented by a concept name; 3. [cat: sn f]
(for category: singular noun, feminine ) specifies the morphosyntactic proper-
ties contributing directly to the combinatorics, for example, regarding agree-
ment; 4. [sem: mass] (for semantics: mass 4 ) specifies morphosyntactic prop-
erties not directly contributing to the combinatorics.
The compositional aspect of the word form is represented by the remain-
ing attributes. Because the proplet illustrated in 3.1.1 is a lexical proplet, the
compositional attributes have no values yet, but will receive their values dur-

brought into use the term “abstract data type” to refer to “software” data structures, which are de-
fined independently of their hardware implementation. With the increasing abstracting away from the
hardware level, the two terms are now widely used interchangeably. This makes it possible to call the
non-recursive feature structure of proplets a “data structure,” which is a shorter, simpler, and more
intuitive term than “abstract data type.”
2 According to ISO 24610-1 (TC37/SC4), a standard feature structure is a(n unordered) set of attribute-
value pairs (Carpenter 1992), presumably for reasons of mathematical aesthetics. Our alternative of
defining proplets as a(n ordered) list of attribute-value pairs has the advantages of easier readability
and a more efficient computational implementation, for example, for pattern matching, cf. 1.3.1.
3 A precursor of proplets, used in NEWCAT'86 and in FoCL'99, Chaps. 16-18, is a format of ordered
triples, consisting of (i) the surface, (ii) the category, and (iii) the base form. For example, the word
form gave was analyzed as the triple (gave (n' d' a' v) give) , which may be transformed into a
character-separated value (csv) notation (as used in the original WordNet (Miller1996), CELEX2
(Olac Record 1996), and other lexical encodings). The corresponding proplet notation would be
⎡
⎣

⎤
⎦

sur: gave
verb: give
cat:n'd'a'v
...

, which is a flat (non-recursive) feature structure.