| | | | | specification is a two-item list of the form |
| | | | | ( { attribute | I path ] [ path [ value } ). |
| , ~rammar writer. | | | | |
| | | | | Here attribute is an atom, path is a list, and value is |
| 2 Grammar Formalism | | | | either an atomic symbol, a list of specifications, or an |
| 2.1 Rules | | | | | abbreviation tbr such a list. | The last case is |
| | | | | distinguished from the first by prefixing the value symbol |
| A rule in I]-PATR is a list of atomic eot~stituent labels | with (,~) when it has an abbreviatory role. Ignoring the |
| that may be followed by speczfications. Specifications are | (\u00a2L>cases, this gives four different kinds of specifications: |
| constraints upon one or more constituents of the rule. in | (attribute vahte), (path value), (attribute path), and (path |
| the simplest case, there are no specifications and the | path). The same feature set can often be specified in |
| labels correspond to symbols in an ordinary phrase | several different ways; in choosing one, we generally try |
| structure rule. For example, the rule | | | to minimize the number of parentheses. |
| S + NP VP | | | | |
| in I) PATR notation is written as | | | | Below is a simple example of a phrase structure rule |
| (S NP VP) | | | | | augmented with specifications. |
| Before a rule is used by the parser, I)-PATI{ compiles it |
| to a feature set. A feature set can be displayed in di[ferent |
| ways--for example, as a matri\u00d7 or as a directed graph. [n |
| this paper, we usually represent feature sets as graphs |
| but the matrix notation will also be used occasionally. [n |
| these graphs, the constituents of the rule are represented |
| by labeled arcs, called attributes. The nodes in the graph |
| are values. A value can be atomic or complex; a complex |
| wfiue is another set of attribute-value pairs. | By |
| convention, the symbol on the left-hand side of a phrase |
| structure rule is represented by the numeric attribute 0. |
| Constituents on the right-hand side of the rule are |
| numbered left-to-right, starting with 1. The above rule |
| D PATR represents as the following feature set, shown |
| here first as a graph and then as the equivalent matrix. |
| O----cat--S i ..... cat--NP 2 .... cat--VP | 0 1 2 | [cat S]-[cat < [cat VF;]I | D-PATR consists of four basic parts: \u2022 A unification package |
| arity--2 | | ar i t,y 2 | I t | \u2022 Interpreter for rules and lexical items |
| | | | | \u2022 Input output routines for directed graphs |
| labels I | I--NP | labels | NP | | \u2022 An Earley style chart parser. |
| --2--VP | | V | | These packages are written in simple Interlisp-D for |
| | | | | transportability to other dialects of LISP. They do not |
| | | | | depend on the features of any particular machine. The |
",
"text": "It provides good facilities for writing and editing grammars as well as many debugging tools for the"
},
"TABREF2": {
"type_str": "table",
"num": null,
"html": null,
"content": "| cat | l/ | | c(~t\\ |
| le_~ | [ / | | -t,_.x~ |
| syntactic arguments. 2.5 l,exieal Rules The expansion of morpheme definitions by means of templates is a straightforward matter: an initial graph acquires additional features and, perhaps, new values for features it already has. I)-['A'I'I{ also allows a more radical transformation by means of lexical titles. A lexical rule is a special kind of template with two attributes: m and out. [n applying a lexical rule to a graph, the latter is J]rst mentioned explicitly only in rules that introduce fillers, such as the relative-clause rule, and in the lexical entries of relative and interrogative pronouns. Other rules are automatically augmented by I)-PAI'R in the appropriate manner when they are compiled to feature sets used by the parser. By deactivating this facility, the grammar writer can also take care of fillers and gaps in a manner of his own choosing. 3 Parsing and Unification D PATR uses an active chart parser that proceeds in a top-down, breadth-first manner. Because the constituents in a rule are feature sets rather than atomic symbols, the task is a bit more complicated than in standard implementations of Earley's algorithm. We unified with the value of in. i n -the object slot in its syntactic frame, even though ....... ~ntics--relatior,-fl-,',d,dion\\ \\ SOMEBODY--arR! 7 sem(lnticsl-out / The effect of the rule is to make a transitive verb lose semantically it remains a two-place predicate. The semantic effect of the rule is to unify arg2 with the subject's semantics and to assign to argl the value SOMEI}OI)Y. This is similar to the analysis of passives in some LI,'G grammars. 2.6 Fillers and Gaps Constructh)ns such as the fb[lowing contain constituents that, semantically and syntactically, fill a vacant slot--a gap--somewhere in the adjacent structure. That paper [ don't intend to read --. Good aw)cados are hard to find --. The neighbor whose car you asked to borrow --called. Is this the company the ht.stogr'am of whose productmrt she wants to display --? From a parser's point of view, there are two main problems to be solved. For the parse to succeed, the filler needs to be available when the incomplete structure is encountered. There must also be a way to etlsure Chat a consider two cases here. |
| designated filler will be consumed by a gap somewhere. A |
| third problem is that, in relative clauses, the filler must |
| contain a relative pronoun. | |
| Many solutions to these problems have been proposed |
| and could be implemented in D PATR. As a convenience, |
| D PAI'It also makes available to the grammar writer a |
| built-in default | mechanism | for distributing | the |
| information about fillers, gaps, and relative pronouns in |
| an appropriate way. | The original idea, conceived by |
| Fernando Pereira, was implemented for gaps in z PATR |
| by Stuart Shieber. The scheme in I)-t)A'I'R is an |
| improvement in that it also handles sentences with |
| nested filler-gap dependencies. | |
",
"text": "If the operation succeeds, the value of out is passed on as the result. Because the values under out cart be linked selectively to the corresponding values under in, the usual result of applying a [exical rule is a metamorphosis of the input graph. As an example. let us consider the graph corresponding to a possible lexieal rule for Passive in Engiish. To make it easier to see the effect of the rule, the graph is turned aroud so that the out values are opposite to the corresponding Ln values;the indeterminates in the middle are unified with their counterparts in the word to which the rule is applied.The default mechanism uses four special features: gapln, gapOut, relIn and relOut. These features need to be"
}
}
}
}