English's predicate syntax is highly positional. That is, the role of each word in a predicate is determined largely by its position in the phrase. There are a several common patterns to the predicate word order, but the specific pattern that applies to a given verb is essentially idiomatic to that verb, especially with respect to complement words (like the "up" in "pick up"). Our approach to defining the predicate grammar is therefore to define a separate, custom syntax rule for each verb. This makes it easy to add rules for the odd little idioms in English verbs.

For verbs that take indirect objects, the indirect object is usually introduced by a preposition (e.g., PUT KEY IN LOCK). Since we consider the preposition in such a case to be part of the verb's grammatical structure, we write it directly into the grammar rule as a literal. This means that we wouldn't be able to parse input that's missing the whole indirect object phrase (e.g., PUT KEY). We don't want to just reject those without explanation, though, which means we have to define separate grammar rules for the truncated verbs. Some of these cases are valid commands in their own right: UNLOCK DOOR and UNLOCK DOOR WITH KEY are both valid grammatically. But PUT KEY isn't, so we need to mark this as missing its indirect object. We do this by setting the missingRole property for these rules to the role (usually IndirectObject) of the phrase that's missing.

Each VerbRule has several properties and methods that it can or must define:

action [Required] - The associated Action that's executed when this verb is parsed. The base library requires this property.

verbPhrase - The message-building template for the verb. The library uses this to construct messages to describe the associated action. The format is 'verb/verbing (dobj) (iobj) (accessory)'. Each object role in parentheses consists of an optional preposition and the word 'what' or 'whom'. For example, 'ask/asking (whom) (about what)'. Outside of the parentheses, you can also include verb complement words before the first object or after the last, but never between objects: for example, 'pick/picking up (what)'.

missingQ - the template for asking missing object questions. This consists of one question per object, separated by semicolons, in the order dobj, iobj, accessory. You only need as many questions as the verb has object slots (i.e., you only need an iobj question if the verb takes an indirect object). The question is simply of the form "what do you want to <verb>", but you can also include the words "it" and "that" to refer to the "other" object(s) in the verb. "It" will be replaced by it/him/her/them as appropriate, and "that" by that/them. Use it-dobj, it-iobj, it-acc to specify which other object you're talking about (which is never necessary for two-object verbs, since there's only one other object). Put the entire 'it' phrase, including prepositions, in parentheses to make it optional; it will be omitted if the object isn't part of the command input. This is only necessary for objects appearing earlier in the verb rule, since it's resolved left to right.

missingRole - the object role (DirectObject, etc) that's explicitly missing from this grammar syntax. This is for rules that you define specifically to recognize partial input, like "PUT <dobj>". The parser will ask for the missing object when it resolves such a rule.

answerMissing(cmd, np) - the base library calls this when the player answers the parser's question asking for the missing noun phrase. 'cmd' is the Command, and 'np' is the noun phrase parsed from the user's answer to the query. This is called from the base library but isn't required, in that it's purely advisory. The point of this routine is to let the verb change the command according to the reply. For example, in English, we have a generic Put <dobj> verb that asks where to put the dobj. If the user says "in the box", we can change the action to Put In; if the user says "on the table", we can change the action to Put On.

dobjReply, iobjReply, accReply - the noun phrase production to use for parsing a reply to the missing-object question for the corresponding role. Players sometimes reply to a question like "What do you want to put it in?" by starting the answer with the same preposition in the question: "in the box". To support this, you can specify a noun phrase production that starts with the appropriate preposition (inSingleNoun, onSingleNoun, etc).

(Note that the base library doesn't place any requirements on exactly how the verb rules are defined. In particular, you don't have to define one rule per verb, the way we do in English. The English module's one-verb/one-rule approach might not be a good fit when implementing a highly inflected language, since such languages are typically a lot more flexible about word order, creating a wide range of possible phrasings for each verb. It might be easier to for such a language to define a set of universal verb grammar rules that cover the common structures for all verbs, and then define the individual verbs as simple vocabulary words that slot into this universal phrase structure.)

VerbRule(Take)      // original source text
grammar VerbRule(Take) :   VerbProduction      // after macro expansion