Compositional Semantics - Computational Linguistics and Talking Robots - page 121

In the PSG graph, the modifier-modified relation is not expressed directly, but
must be deduced from the dominance and precedence constellations of the NP,
the ADJ, and the N nodes. In the DG graph, the nodes the and little are spec-
ified indiscriminately as depending on girl ,and girl is specified as depending
on sleep . In a DBS graph, however, any “ ” line is defined specifically to rep-

resent the relation between a modifier and a modified. Given that little is an
Aand girl is an N, the DBS graph in 7.1.2 characterizes the relation between
little and girl explicitly as A

|

| N, i.e., as adnominal (3.5.3, 3.5.4).

7.2 Absorption and Precipitation of Function Words

Another difference between the PSG and DG graphs on the one hand and the
DBS graph on the other is in the treatment of function words. For example, the
definite article the appears as a node in the PSG and DG graphs of 7.1.2, but
not in the DBS graph. This is because the PSG and DG graphs are language
analyses, while the DBS graph is an analysis of content. For a representation of
content it does not matter whether a certain semantic value such as definiteness
is coded by means of a function word, as in English, or by means of a suffix,
as in Romanian. 6
Function words of English are determiners like the, a(n), some, every,
all ; auxiliary forms like have, had, am, are, were ; prepositions like in, on,
under, above, below ; conjunctions like and, or, that, when; and punctuation
signs like comma, full stop, question mark, and exclamation mark.
In accordance with Surface Compositionality, function words are lexically
analyzed as proplets which contain certain attributes and/or values represent-
ing semantic properties of the word in question. In the hear mode, a function
word and its content word are fused and their properties are inherited by the
proplet resulting from such a function word absorption . For example, the aux-
iliary was introduces the value past into the resulting verb proplet, the deter-
miner the introduces the value def into the resulting noun proplet, etc. In the
speak mode, these same values are used for function word precipitation . 7

3 In contrast, Arc Pair grammar (Johnson and Postal 1980), Relational grammar (Perlmutter 1980), and
Functional grammar (Dik 1989, 1997) define grammatical relations such as “subject” and “object” as
elementary, as in the DBS approach to the semantic relations of structure.
4 We are referring here to DG as originally presented by Tesnière (1959). In recent work on DG, such
as Liu (2009b) and Hudson (2010), dependencies are differentiated by annotating different roles.
5 The differentiated interpretation of different graph structures was pioneered by Frege (1879) in his
Begriffsschrift .
6 For example, om (man) is marked as definite with the suffix -ul ,asin om-ul (the man).
7 Function word absorption has the practical advantage that the high frequency words with low seman-
tic significance (Zipf 1932, 1935, 1949) are moved out of the way of a meaningful retrieval.