3.3.3 Inheritance hierarchies
Howshould thegrammar account for thefactthatmany verbs expressingtransfer (orintended transfer)appear intwo
possible syntactic frames?
(20) a.Beth gave/handed/sent/offered a teapot to Nancy,
b. Beth gave/handed/sent/offered Nancy a teapot.
(21) a.Beth told a story to Nancy,
b Beth told Nancy a story.In early generative grammar, this alternation was accomplished by means of a derivational rule (often called Dative
Shift)thatoptionallyturned theunderlyingorder (20a, 21a) intothesurfaceorder (20b, 21b);a variantofthisapproach
has been revived more recently (Larson 1988 Larson Larson). The advantage of this treatment is that the verbs in
question need to be supplied only with a single subcategorization feature instead of two; the price is this single
derivational rule.
However, again for a varietyof technical reasons (see references in Levin 1993; Jackendoff 1990b), most linguists have
rejected the derivational approach for this alternation. But this leaves the proble mof why all these verbs have two
distinctsubcategorizationframes. Oneapproach thathas wonsome degreeofacceptance(Pollardand Sag1987; 1994;
Michaelis and Lambrecht 1996) is that the lexicon contains not just the actual lexical items of the language but also
more abstract schemata from which actual items can“inherit”properties. In the present case, the schema might be
statedroughlyas (22); thesubscriptscorrelatethesyntacticvariableswiththesemanticones, in theway theydid in Fig.
1.1.
This says that a verb that means“transfer”accepts both syntactic frames, with the same meaning. Because the words
in(20) and (21) inherittheir properties fro mthissche ma, they“cost less”: theextra subcategorizationcomes ina sense
“for free.”New items in the language with the appropriate meaning inherit these properties as well, with no further
ado; for instance the new verbfaxhas both forms.
As in the discussion of derivational rules versus constraints, we see here an important locus of contention among
alternative frameworks. Given any particular phenomenon, should it be treated as a derivational rule, a lexical
redundancy rule, or an inheritance hierarchy? These are not just matters of notational