Generalizing Automatically Generated Selectional Patterns

Frequency information on co-occurrence pa t te rns can be att tomatically collected from a syntactically analyzed corpus; this information can then serve as the basis for selectional constraints when analyzing new text; from the same domain. Tiffs information, however, is necessarily incomplete. We report on measurements of the degree of selectional coverage obtained with ditt\~rent sizes of corpora. We then describe a technique for using the corpus to identify selectionally similar terms, and for using tiffs similari ty to broaden the seleetional coverage for a tixed corpus size. 1 I n t r o d u c t i o n Selectional constraints specify what combinat ions of words are acceptable or meaningful in particular syntactic relations, such as subject-verb-object or headmodifier relations. Such constraints are necessary for the accurate analysis of natural language text+ Accordingly, the acquisition of these constraints is an essential yet t ime-consuming par t of port ing a natural language system to a new domain. Several research groups have a t t empted to au tomate this process by collecting co-occurrence pa t te rns (e.g., subject-verb-ol)ject patterns) from a large t ra ining corpus. These pat terns are then used as the source of seleetional constraints in attalyzing new text. The initial successes of this approach raise the question of how large a t raining corpus is required. Any answer to this question must of course be relative to the degree of coverage required; the set of selectional pat terns will never be 100% complete, so a large corpus will always provide greater coverage. We a t t empt to shed to some light on this question by processing a large corpus of text from a broad domain (business news) and observing how selectional coverage increases with domain size. In many cases, there are l imits on the amount of t raining text, available. We therefore also consider how coverage can be increased using a tixed amount of text. The most s traightforward acquisition procedures build selectional pat terns containing only the specific word combinat ions found in the t raining corpus. (areater coverage can be obtained by generalizing fl'om the pat terns collected so tha t pat terns with semantically related words will also be considered acceptable. In most cases this has been (lotto using manually-created word classes, generalizing fi'oul specific words to their classes [12,1,10]. If a pre-existing set of classes is used (as in [10]), there is a risk tha t the classes awdlable may not match the needs of the task. If classes are created specifically to capture selectional constraints, there lnay be a substant ial manual I>urden in moving to a new domain, since at least some of the semantic word classes will be domain-specillc. We wish to avoid this manual component by auto: mari ta l ly identifying semantically related words. This can be done using the co-occurrence data, i.e., by idea: tifying words which occur in the same contexts (for example, verbs which occur with the same subjects and objects). From the co-occurrence data o110 Call coiil.pute a similarity relation between words [8,7]. This similarity information can then be used in several ways. One approach is to form word clusters based on this similarity relation [8]. This approach was taken by Sekine et al. at UMIST, who then used these chlsters to generalize the semantic pat terns [11]. l 'ereira et al. [9] used a variant of this approach, "soft clusters", in which words can be members of difl'erent clusters to difl'eren t degrees. An al ternat ive approach is to use the word similarity information directly, to inDr information about the likelihood of a co-occurrence pat tern from information abont pat terns involving similar words. This is the approach we have adopted for our current experiments [6], and which has also been employed by 17)agan et al. [2]. We corl:lttttl;e from the co+occurrence data a "confitsion matr ix" , which measures the interchangeability of words in particular contexts. We then use the confllsion matr ix directly to geueralize the selllantic patterns. 2 A c q u i r i n g S e m a n t i c P a t t e r n s Based on a series of experitnents over tile past two years [5,6] we have developed the following procedure