Using the meanings of verbs to select the countability of English noun phrases

Francis Bond,	Kentaro Ogura,	Satoru Ikehara,	Satoshi Shirai

NTT Network Information Systems Laboratories

[ 1993, Vol.6, p.61 (1993.9). ]
[ In Proceedings of 1993 Fall IEICE Meeting, Vol.6, p.61 (September, 1993). ]

INDEX

This article discusses the restrictions that three kinds of verbs (including predicates such as be extinct) place on the countability of English noun phrases used generically. The first kind are verbs that operate over a set of individuals, for example gather and collect. The second predicate something of an entire class, for example evolve and&e extinct. The third are verbs whose objects tend to refer to an entire class, such as like and prefer.

First, the different ways of expressing genericness in English noun phrases are discussed, particularly with respect to the countability of the nouns heading them. Second, the ways verbs restrict how a NP can be expressed are described, with some explanation as to why. Finally, an example is given of the use of these restrictions in automatic translation from Japanese to English.

An NP used to make a statement about an entire class is generic. NPs can be countable or uncountable; the behaviour of nouns within these two classes will depend on the nouns' countability. An NP which is within the scope of a denumerator is countable. Denumerators are the natural numbers 1, 2, 3,... numerical quantifiers such as both, dozen, score, and quantifiers such as a/an, each, every, either, many, several and (a) few(er) (Bond 1993).

A generic NP (with a countable head noun) can generally be expressed in three ways: GEN 'a', A mammoth is a mammal, GEN 'the'. The mammoth is a mammal, and GEN , Mammoths are mammals (Huddleston 1984). The NPs heading GEN 'a' and 'the' are countable, GEN 'a' because it falls within the scope of 'a', GEN 'the' because it takes as predicate a countable NP. GEN NPs are not within the scope of a denumerator and so are uncountable. NPs headed by uncountable nouns have no countable form and can only be expressed by GEN ^ (eg: Furniture is expensive). Nouns that can be both countable and uncountable only take GEN , although they can be either singular (or plural ( (Cake is delicious/ Cakes are delicious). These combinations are shown in table 1.

GEN type	Noun Countability			Noun Phrase Countability
	Countable	Both	Uncountable
'a' 'the'	a mammoth the mammoth mammoths	*a cake *the cake cake/cakes	*a furniture *the furniture furniture	Countable Countable Uncountable

The following discussion will focus on NPs headed by countable nouns, because only they can be expressed in more than one way.

Verbs such as gather and collect place the most severe restrictions on the possible generic uses, only GEN can be used. This is true of the subject for intransitive use. Mammoths collect around water holes and the object for the transitive use, I collect mammoths. This is due to the meaning of the verb; a collection must contain 2 or more members for it to be a collection. Therefore, if I collect mammoths I must collect more than one, so the NP is represented in English by an uncountable noun phrase.

Note that it is possible to collect a single mammoth, but it should be explicitly specified, for example I only collected one mammoth..

Verbs which predicate over a whole class, such as evolve, can take GEN or 'the'. Mammoths evolved into elephants. The mammoth evolved into the elephant. For these verbs, the subject must be the entire class as a whole. Individual mammoths are not the class as a whole so GEN 'a' cannot be used, thus the ungrammaticality of *A mammoth evolved into an elephant.

Finally, verbs such as like and prefer can take GEN 'a', as well as GEN and 'the', for example I like a nice mammoth. However, this usage seems to be restricted to uses when the object is modified; in the unmodified case I like a mammoth, the natural interpretation is that there is a mammoth that I like, not that I like mammoths in general.

Type:	Example:	Subject:	Object:
I II III	collect (vi) evolve like	GEN (vi) GEN , 'the' none	GEN (vt) none GEN , 'the','a'

Countability and number are not explicitly marked in Japanese, so they must be generated when translating from Japanese to English. The restrictions described in section 3 have been implimented as follows in the experimental machine translation system ALT-J/E.

First, in the absence of other information verbs of the above three types are assumed to be used generically unless the NPs are specifically modified, that is "watashi-ha zou-ga suki da" 'I-TOP elephant-OBJ like' is assumed to refer to all elephants, not some specific elephant. If 'elephant' is modified by a modifier that identifies it specifically, for example "kono-zou" 'this-elephant', then obviously it is not considered to be generic.

Second, as GEN is acceptable for all environments, all generic NPs are translated as such, that is as a bare uncountable noun phrase. The sentence above thus becomes I like elephants. Evaluation has shown that this improves both the fidelity and readability of the translations. Research is now continuing on how to identify verbs of the three types, and determining under what conditions if any GEN 'the' should be generated.

The semantic properties of verbs place restrictions on the ways noun phrases can be expressed generically in English. These restrictions are used to improve the quality in the Japanese to English machine translation system ALT-J/E.

Bond, Francis et al. 1993.

Determination of whether an English noun phrase is countable or not using 6 levels of lexical countability. Proceedings of the IPSJ. Tokyo: IPSJ.

Huddleston, Rodney. 1984.

Introduction to the Grammar of English. Cambridge: University Press.