This article shows that the Web can be employed to obtain frequencies for bigrams that are unseen in a given corpus. We describe a method for retrieving counts for adjective-noun, noun-noun, and verb-object bigrams from the Web by querying a search engine. We evaluate this method by demonstrating: (a) a high correlation between Web frequencies and corpus frequencies; (b) a reliable correlation between Web frequencies and plausibility judgments; (c) a reliable correlation between Web frequencies and frequencies recreated using class-based smoothing; (d) a good performance of Web frequencies in a pseudodisambiguation task. 1.

Abstract not found

This paper explores the determinants of adjective-noun plausibility by using correlation analysis to compare judgements elicited from human subjects with five corpus-based variables: co-occurrence frequency of the adjective-noun pair, noun frequency, conditional probability of the noun given the adjective, the log-likelihood ratio, and Resnik's (1993) selectional asso- ciation measure. The highest correlation is obtained with the co-occurrence frequency, which points to the strongly lexicalist and collocational nature of adjective-noun combinations.

In this article we investigate logical metonymy, that is, constructions in which the argument of a word in syntax appears to be different from that argument in logical form (e.g., enjoy the book means enjoy reading the book, and easy problem means a problem that is easy to solve). The systematic variation in the interpretation of such constructions suggests a rich and complex theory of composition on the syntax/semantics interface. Linguistic accounts of logical metonymy typically fail to describe exhaustively all the possible interpretations, or they don't rank those interpretations in terms of their likelihood. In view of this, we acquire the meanings of metonymic verbs and adjectives from a large corpus and propose a probabilistic model that provides a ranking on the set of possible interpretations. We identify the interpretations automatically by exploiting the consistent correspondences between surface syntactic cues and meaning. We evaluate our results against paraphrase judgments elicited experimentally from humans and show that the model's ranking of meanings correlates reliably with human intuitions.

Previous work on the induction of se- lectional preferences has been mainly carried out for English and has concentrated almost exclusively on verbs and their direct objects. In this paper, we focus on class-based models of selec- tional preferences for German verbs and take into account not only direct objects, but also subjects and prepositional complements. We evaluate model performance against human judgments and show that there is no single method that overall performs best. We explore a variety of parametrizations for our mod- els and demonstrate that model combi- nation enhances agreement with human ratings.

In this paper, we investigate the interaction of phonological and syntactic constraints on the realization of Information Structure in Greek, a free word order language. We use magnitude estimation as our experimental paradigm, which allows us to quantify the influence of a given linguistic constraint on the acceptability of a sentence. We present results from two experiments. In the first experiment, we focus on the interaction of word order and context. In the second experiment, we investigate the additional effect of accent placement and clitic doubling. The results show that word order, in contrast to standard assumptions in the theoretical literature, plays only a secondary role in marking the Information Structure of a sentence. Order preferences are relatively weak and can be overridden by constraints on accent placement and clitic doubling. Our experiments also demonstrate that a null context shows the same preference pattern as an all focus context, indicating that `...

Binding theory is the component of grammar that regulates the interpretation of noun phrases. Certain syntactic configurations involving picture noun phrases (PNPs) are problematic for the standard formulation of binding theory, which has prompted competing proposals for revisions of the theory. Some authors have proposed an account based on structural constraints, while others have argued that anaphors in PNPs are exempt from binding theory, but subject to pragmatic restrictions. In this paper, we present an experimental study that aims to resolve this dispute. The results show that structural factors govern the binding possibilities in PNPs, while pragmatic factors play only a limited role. However, the structural factors identified differ from the ones standardly assumed.

As agents approach animal-like complexity, evaluating them becomes as difficult as evaluating animals. This paper describes the application of techniques for characterizing animal behavior to the evaluation of complex agents. We describe the conditions that lead to the behavioral variability that requires experimental methods. We then review the state of the art in psychological experimental design and analysis, and show its application to complex agents. We also discuss a specific methodological concern of agent research: how the robots versus simulations debate interacts with statistical evaluation. Finally, we make a specific proposal for facilitating the use of scientific method. We propose the creation of a web site that functions as a repository for platforms suitable for statistical testing, for results determined on those platforms, and for the agents that have generated those results. Keywords: agent performance, complex systems, behavioral indeterminacy, repl...

We analyzed a corpus of spoken dialogues to identify cases in which it appears that the addressee of an utterance containing an anaphoric pronoun does not have enough evidence to resolve that pronoun. Based on these cases, we propose two conditions under which such 'sloppy' anaphoric references are innocuous. We ran preliminary experiments indicating that subjects find anaphoric pronouns occurring in ambiguous contexts fulfilling these conditions significantly easier to process than pronouns occurring in minimally different contexts in which they are not.

this paper, we use the term topology to refer to a row of slots. Filler items are terminal nodes of an unordered syntactic tree called mobile. In the simplest case, the mobile consists of two layers: the bottom layer specifying the filler items, which are all connected to a single root node in the top layer. In more complicated cases, the mobile may span additional layers of nodes. In a typical model application, the `lexicon' of filler items contains many different entries. However, these entries belong to a small number of classes, and the item-to-slot binding process is sensitive to class membership. That is, certain slots only accept items of certain classes. Already in 1975, Merrill Garrett proposed a slot-and-filler model for syntactic speech errors that involve item miss-orderings of words or phrases (as in Although murder is a form of suicide). He had discovered that the likelihood of such errors is independent of the distance between the permuted elements in the surface string, whereas other types of exchanges (e.g. between phonemes) occur predominantly between elements in neighboring positions. He took this contrast as evidence for two distinct levels of processing: a `functional' level where grammatical relationships between constituents are established, vs. a `positional' level where constituents are ordered from left to right. This distinction has been adopted, in one way or another, by many students of grammatical encoding (e.g. Kempen & Hoenkamp (1987), Levelt (1989), Bock & Levelt (1994) and Kempen & Harbusch (1998)). In Kempen & Hoenkamp 's version, the constituents of unordered functional structures acquired their linear position by binding themselves to a slot in a 'holder' which is similar to a topology in more recent literature (see Kathol (2000) in...