The First Update to eg2X — eg2A
| Level 2 | Level 2A | Level 2B | Level 2C | Level 2D |
Read Document 2X
and Homework 2A.
The first update to the basic grammar eg2X is concerned primarily with expansion of the definitions <VERB> and <OBJECT>, and the introduction of <TENSE> into <ASSERTION>. Some verbs carry their tense as part of the word, and are assigned the category TV (builds). Some verbs, i.e. infinitives, V, carry no tense (e.g. build in to build and will build). As a main verb in ASSERTION, a V requires the presence of a preceding "tense word" W (e.g. may in They may build a tower). For the occurrence of a tense word we add the element <TENSE> to <ASSERTION>. The value of <TENSE> is either a word with the dictionary category assignment W (written <*W> in a BNF definition of the grammar) with its optional left and right adjuncts, or <*NULL>. To go along with the introduction of <TENSE> in the definition of <ASSERTION>, a change in the defition of <VERB> is required. The symbol <*TV> in the definition of <VERB> in the grammar eg2X.txt is replaced by <VVAR> (verb variants) whose value is either <*TV> or <*V>. The new restriction WVC1 checks that if the value of <VVAR> is <*TV>, then <TENSE> is empty (No one builds a stone, *No one may builds a stone), and if the value of <VVAR> is <*V>, then a tense word occurs in <TENSE> (One may build a tower, *One build a tower). English also has so-called compound tenses, formed with be and have (He is going, He has gone; also The tower was built). Since the forms of be and have in these constructions are themselves tensed, we treat the compounds as composed of tensed be and have followed by an object, which contains the second verb of the compound with its object. Thus, in They had built a tower, the main verb is had with object built a tower, in which built is the verb, and a tower is its object. To treat the compound tenses with have (e.g. has gone) we add to <OBJECT> the option <VENO>, which occurs only with forms of the verb have. In the dictionary, the verb have in all its forms has VENO as a value of its OBJLIST attribute. VENO is defined as follows:
The first element <LVENR> consists of a past participle with its optional left and right modifiers. The third element of <VENO> is <OBJECT>. Optional sentence adjuncts <SA> can occur between the verb and its object (They have finally investigated the tower), or after the object (They have investigated the tower finally). To treat the compound tenses with be (e.g. is going) the option <OBJECTBE> is added to <OBJECT>. <OBJECTBE> has the option <VENPASS> for the passive construction (A tower was built), the option <VINGO> for a present participle with its object (They are building a tower), and the option <OBJBE> for the non-verbal objects of be, here represented by a lone adjective <*ADJ> (The tower is old), and a noun string <NSTG> (The tower is an artifact). In parsing, rather than testing for the literal forms of be (is, are, etc. ) we introduce the attribute VBE, which is added to the dictionary entry for each form of the verb be. For example, the entry for be is:
The ATTRIBUTE list of the grammar is updated to include VBE. Note that the updated ATTRIBUTE list is present in the grammar eg2A which is now used in compiling the source dictionary wd2A.src. Another change to the grammar involves the replacement of WVERBOBJ with the two disqualify restrictions DVC1 and DVC2. Using the EDIT function DVC1 rules out <OBJECT> options that do not appear on the OBJLIST attribute in the dictionary entry of the governing verb. Similarly, DVC2 uses the EDIT function to rule out <PASSOBJ> options that do not appear on the POBJLIST attribute in the dictionary entry of the governing passive verb. Both DVC1 and DVC2 are Disqualify restrictions. A Disqualify restriction is executed before the specified structure is built. Thus DVC1, for example, is more efficient than the well-formedness restriction WVERBOBJ, since WVERBOBJ had to wait for the object to be built before checking whether it was appropriate for the given verb. A new type list VERBAL is defined so that the governing verb can be located regardless of its specific form.
VERBAL is used in DVC1. In addition to the above changes related to the verb, several other additions have been made that will help in parsing the tower text. Since there may appear in the subject position with a VBE verb (there is a tower), the literal THERE is added as an option of <SUBJECT> with its occurrence restricted by WTHERE. The restriction WAGREE-VO checks number agreement between a noun in <OBJBE> and its governing verb. This will rule out *They are a tower and *They is towers. But note - this rules out some acceptable occurrences, e.g. They are a problem. Also, there is a minor change in <LPRO>. The atomic symbol <*D> (for an adverb, classed in the dictionary as D) is added as an option to <LPRO> to handle phrases such as only they, even he. The number of adverbs which may appear immediately before a pronoun is small; a later version of the grammar will restrict their appearance by using adverb subclasses. Below is a summary of the changes made to form eg2A.txt from eg2X.txt, and the changes to form the dictionary wd2A.src from wd2X.src. GRAMMAR CHANGES
DICTIONARY CHANGES
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
