Syntactic Generalization Research Articles

Transfer learning of syntactic structures for building taxonomies for search engines

We apply a paradigm of transfer learning to build a taxonomy of entities intended to improve search engine relevance in a vertical domain. The taxonomy construction process starts from the seed entities and mines available source domains for new entities associated with these seed entities. New entities are formed by applying the machine learning of syntactic parse trees (their generalizations) to the search results for existing entities to form commonalities between them. These commonality expressions then form parameters of existing entities, and are turned into new entities at the next learning iteration. To match natural language expressions between source and target domains, we use syntactic generalization, an operation which finds a set of maximal common sub-trees of constituency parse trees of these expressions.Taxonomy and syntactic generalization are applied to relevance improvement in search and text similarity assessment. We conduct an evaluation of the search relevance improvement in vertical and horizontal domains and observe significant contribution of the learned taxonomy in the former, and a noticeable contribution of a hybrid system in the latter domain. We also perform industrial evaluation of taxonomy and syntactic generalization-based text relevance assessment and conclude that a proposed algorithm for automated taxonomy learning is suitable for integration into industrial systems. The proposed algorithm is implemented as a component of Apache OpenNLP project.

Old Norse as anNPLanguage: With Observations on the Common Norse and Northwest Germanic Runic Inscriptions

AbstractThe differences between languages with definite determiners and those without definite determiners have been the object of much research. Recently, Bošković has uncovered a number of one‐ and two‐way syntactic generalizations which serve to set these two typological groups apart. He accounts for the contrasts observed by proposing that languages with definite determiners (‘DP’ languages) have aDPlayer, while determiner‐less languages (‘NP’ languages) lack thisDPprojection altogether. Since the Old Germanic languages did not have fully grammaticalized definite articles, a reasonable hypothesis to explore might be that they wereNPlanguages in Bošković's sense. In this paper we explore this hypothesis, focusing mainly on Old Norse; where possible we also discuss Old Norse's historical predecessors Common Norse and Northwest Germanic (both of which are attested exclusively in runic inscriptions). The specificNPproperties considered in this paper are (A) the absence of a fully grammaticalized definite article, (B) syntactic discontinuities and free word order, (C) the absence of double adnominal genitives with transitive deverbal nouns, (D) the absence of clitic doubling and (E) the presence of radical argument drop. The paper's approach is guided by basic generative assumptions, but its focus is descriptive rather than theoretical.

Syntactic generalization with novel intransitive verbs

To understand how children develop adult argument structure, we must understand the nature of syntactic and semantic representations during development. The present studies compare the performance of children aged 2;6 on the two intransitive alternations in English: patient (Daddy is cooking the food/The food is cooking) and agent (Daddy is cooking). Children displayed abstract knowledge of both alternations, producing appropriate syntactic generalizations with novel verbs. These generalizations were adult-like in both flexibility and constraint. Rather than limiting their generalizations to lexicalized frames, children produced sentences with a variety of nouns and pronouns. They also avoided semantic overgeneralizations, producing intransitive sentences that respected the event restrictions and animacy cues. Some generated semantically appropriate agent intransitives when discourse pressure favored patient intransitives, indicating a stronger command of the first alternation. This was in line with frequency distributions in child-directed speech. These findings suggest that children have early access to representations that permit flexible argument structure generalization.

The course of actualization

Actualization is traditionally seen as the process following syntactic reanalysis whereby an item's new syntactic status manifests itself in new syntactic behavior. The process is gradual in that some new uses of the reanalyzed item appear earlier or more readily than others. This article accounts for the order in which new uses appear during actualization. Five corpus-based case studies are presented involving reanalysis and actualization in different functional domains of grammar. These include the reanalysis of all but, far from , and Dutch verre van to adverbial downtoners, and the reanalysis of fun and key from nouns to adjectives. It is shown that actualization proceeds from one environment to another on the basis of similarity relations between environments. The similarity relations may involve broad syntactic generalizations but also superficial similarities to existing patterns, including even an item's uses prior to reanalysis. Because actualization is guided by local and global analogies to existing uses, one determinant of the course of actualization is the locus of reanalysis, as it defines the first uses of an item under change, on which subsequent uses can be modeled. It also follows that the course of actualization is both item-specific and language-specific. The findings presented challenge the concept of reanalysis, which appears less abrupt than usually assumed. Further, it is argued that the findings fit best with usage-based models of language, which attribute a prominent role to similarity-based organization in grammar, and in which an item's use can be subject to multiple, potentially conflicting generalizations.

Simple rules can explain discrimination of putative recursive syntactic structures by a songbird species

According to a controversial hypothesis, a characteristic unique to human language is recursion. Contradicting this hypothesis, it has been claimed that the starling, one of the two animal species tested for this ability to date, is able to distinguish acoustic stimuli based on the presence or absence of a center-embedded recursive structure. In our experiment we show that another songbird species, the zebra finch, can also discriminate between artificial song stimuli with these structures. Zebra finches are able to generalize this discrimination to new songs constructed using novel elements belonging to the same categories, similar to starlings. However, to demonstrate that this is based on the ability to detect the putative recursive structure, it is critical to test whether the birds can also distinguish songs with the same structure consisting of elements belonging to unfamiliar categories. We performed this test and show that seven out of eight zebra finches failed it. This suggests that the acquired discrimination was based on phonetic rather than syntactic generalization. The eighth bird, however, must have used more abstract, structural cues. Nevertheless, further probe testing showed that the results of this bird, as well as those of others, could be explained by simpler rules than recursive ones. Although our study casts doubts on whether the rules used by starlings and zebra finches really provide evidence for the ability to detect recursion as present in "context-free" syntax, it also provides evidence for abstract learning of vocal structure in a songbird.

Usage-based vs. rule-based learning: the acquisition of word order inwh-questions in English and Norwegian

This paper discusses different approaches to language acquisition in relation to children's acquisition of word order in wh-questions in English and Norwegian. While generative models assert that children set major word order parameters and thus acquire a rule of subject-auxiliary inversion or generalized verb second (V2) at an early stage, some constructivist work argues that English-speaking children are simply reproducing frequent wh-word+auxiliary combinations in the input. The paper questions both approaches, re-evaluates some previous work, and provides some further data, concluding that the acquisition of wh-questions must be the result of a rule-based process. Based on variation in adult grammars, a cue-based model to language acquisition is presented, according to which children are sensitive to minor cues in the input, called micro-cues. V2 is not considered to be one major parameter, but several smaller-scale cues, which are responsible for children's lack of syntactic (over-)generalization in the acquisition process.

Learning Animacy Hierarchy Effects in Sesotho Double Object Applicatives

Researchers have long debated the means by which children learn the argument structure of verbs. Making syntactic generalizations often entails learning the semantics of different verbs, complicating and delaying the acquisition process. This study investigates four- to twelve-yearolds' and adults' knowledge of animacy hierarchy restrictions on postverbal word order in Sesotho double object applicatives, constructions where verb semantics is kept constant. Performance on forced-choice elicited production tasks showed that four-year-olds have early knowledge of the animacy hierarchy restrictions, providing evidence of syntactic generalization even on low-frequency constructions. Although there were no verb frequency effects, performance was also better on the highest-frequency animacy constructions. The results suggest that learning restrictions on verb-argument structure is facilitated when verb semantics is not a confound, but that construction frequency also plays a role in mastering the argument structure of verbs.

Subject-auxiliary inversion: A natural category

This article argues that subject auxiliary inversion in English (SAI) provides an example of a syntactic generalization that is strongly motivated by a family of closely related functions. Recognition of the functional properties of each subconstruction associated with SAI allows us to predict many seemingly arbitrary properties of SAI: e.g., its (partial) restriction to appear in main clauses, the fact that the inversion only involves the first auxiliary, and the fact that its use in comparatives is more limited. The dominant feature of SAI, being non-positive, is also argued to motivate the syntactic form of SAI. It is suggested that attention to the rich data inherent in language and to findings in categorization research simultaneously serves to reinforce and benefit our understanding of both language and categorization more generally.

Higher order generalization and its application in program verification

Generalization is a fundamental operation of inductive inference. While first order syntactic generalization (antidunification) is well understood, its various extensions are often needed in applications. This paper discusses syntactic higher order generalization in a higher order language λ2 l1r. Based on the application ordering, we prove that least general generalization exists for any two terms and is unique up to renaming. An algorithm to compute the least general generalization is also presented. To illustrate its usefulness, we propose a program verification system based on higher order generalization that can reuse the proofs of similar programs.

Towards a formal taxonomy of hybrid uncertainty representations

In this paper we present some ideas about how to formally relate various uncertainty representations together in a taxonomic structure, capturing both syntactic and semantic generalization. Fuzziness and nonspecificity are presumed as primitive concepts of uncertainty, and transitive and intransitive methods operating with nonspecificity and fuzziness are introduced to generate a base class of hybrid uncertainty representational forms. Additive, maximal, and interval constraints then complete the characterization of the most important hybrid forms.

Pac-learning non-recursive prolog clauses

Recently there has been an increasing amount of research on learning concepts expressed in subsets of Prolog; the term inductive logic programming (ILP) has been used to describe this growing body of research. This paper seeks to expand the theoretical foundations of ILP by investigating the pac-learnability of logic programs. We focus on programs consisting of a single function-free non-recursive clause, and focus on generalizations of a language known to be pac-learnable: namely, the language of determinate function-free clauses of constant depth. We demonstrate that a number of syntactic generalizations of this language are hard to learn, but that the language can be generalized to clauses of constant locality while still allowing pac-learnability. More specifically, we first show that determinate clauses of log depth are not pac-learnable, regardless of the language used to represent hypotheses. We then investigate the effect of allowing indeterminacy in a clause, and show that clauses with k indeterminate variables are as hard to learn as DNF. We next show that a more restricted language of clauses with bounded indeterminacy is learnable using k-CNF to represent hypotheses, and that restricting the “locality” of a clause to a constant allows pac-learnability even if an arbitrary amount of indeterminacy is allowed. This last result is also shown to be a strict generalization of the previous result for determinate function-free clauses of constant depth. Finally, we present some extensions of these results to logic programs with multiple clauses.

PREDICTIVE LEFT‐TO‐RIGHT PARSING OF A RESTRICTED VARIANT OF TAG(LD/LP)

By means of a factorization of linear precedence (LP) information in tree‐adjoining grammar (TAG), the possibility to capture syntactic generalizations can be enhanced, especially with respect to languages which exhibit relative free word‐order. In TAG(LD/LP) LP‐information is factorized in a way reminiscent of the ID/LP format of CFGs. However, in contrast to its context‐free counterpart, in TAG(LD/LP) entanglement of constituents is allowed by admitting not only the permutation of sister nodes, but also of non‐sister nodes in a structural description. In this paper a generalization of the predictive left‐to‐right parser for TAG is presented which is able to handle a restricted variant of TAG(LD/LP), which allows the permutation of sister nodes and non‐sister nodes within elementary structures, the basic building blocks of TAG(LD/LP).

Topic as a grammatical function in Chinese

This paper contests the inconsistant use of ‘topic’ in referring to a syntactic as well as a semantic notion in the literature of Chinese linguistics. We propose that ‘topic’ be reserved as a grammatical function, a syntactic notion parallel to ‘subject’ and ‘object’, as it is used in LFG, Lexical-Functional Grammar. We propose the term ‘frame’ to refer to the semantic or discoursal function encoded by the syntactic topic. Furthermore, we provide evidence for the argument that TOPIC, as a grammatical function in LFG, should not be considered subcategorizable in Mandarin Chinese, contrary to the proposal made by Huang (1989), who observed a small number of verbs requiring topics. We illustrate some of the syntactic generalizations that our analysis can maintain, which would be missed by the analysis of subcategorizable TOPIC. Finally we employ the theories of lexical diffusion and linguistic interaction (Hsieh 1990) to give a tentative account of the linguistic factors facilitating the idiosyncratic behavior of these verbs.

Lexical categories in Japanese

The lexical feature system proposed by Chomsky (1970), [+/-V, +/-N], makes it possible to capture a number of significant syntactic and morphological generalizations. The feature system can minimally distinguish the four major English categories, but in Japanese, there are six categories commonly recognized. It is shown that the feature system is capable of dealing with the Japanese lexical categories. Those properties that resist ready categorization by the feature system are shown to derive from general principles of the grammar, thus making categorial distinction based on those properties unnecessary.

Structure Dependence in Grammar Formation

Structure Dependence in Grammar Formation

Clarification and the child's learning of grammar

The notion of clarification in language addressed to children was introduced by Ferguson, following a distinction made by Hockett between what he called frequency and clarity norms. The term is intended by Ferguson to be a descriptive label for some of the characteristics of motherese, but it has functional implications. If mothers ‘clarify’ their speech, particularly by slowing it down so that what is normally unstressed or reduced in speech becomes stressed or full, then relationships among grammatical items often obscured in fast speech may become apparent in a ‘clarified’ variety. This paper reviews work by Newport, Gleitman and Gleitman (NGG) which appears to indicate that there is a correlation between certain clarified forms in maternal speech — auxiliary-initial structures — and growth of auxiliaries in the child. Evidence is examined, from corpora available to the author, which bears on maternal use of auxiliaries in English, and the characteristics of auxiliary growth in children's language. We will consider types of auxiliaries and their phonetic realizations, in detail, in relation to three questions: (a) How ‘clarified’ are auxiliaries in input language? (b) How close is the relationship between mother's use of auxiliaries and their growth in the child? (c) To what extent is the child's use of auxiliaries based on what we might want to call syntactic generalizations?

Recursive definitions of partial functions and their computations

The object of this paper is to present a syntactic and semantic model for recursive definitions, and to study the relation between their computed functions and their fixpoints. The recursive definitions that we consider are syntactic generalizations of those introduced in [2] by Kleene and in [5] by McCarthy. Each recursive definition yields two classes of fixpoint partial functions (“fixpoints over D υ {ω}” and “fixpoints over D”), and a class of computed partial functions obtained by applying different computation rules to the recursive definition. In this work we first describe the relations between the computed functions and the fixpoints over D υ {ω} (Theorem 1), and between the computed functions and the fixpoints over D (Theorem 2). Our main interest is in the class of fixpoints (over D or D υ {ω}) which are also computed functions of the recursive definition.

Recursive definitions of partial functions and their computations

The object of this paper is to present a syntactic and semantic model for recursive definitions, and to study the relation between their computed functions and their fixpoints. The recursive definitions that we consider are syntactic generalizations of those introduced in [2] by Kleene and in [5] by McCarthy. Each recursive definition yields two classes of fixpoint partial functions (“fixpoints over D υ {ω}” and “fixpoints over D”), and a class of computed partial functions obtained by applying different computation rules to the recursive definition. In this work we first describe the relations between the computed functions and the fixpoints over D υ {ω} (Theorem 1), and between the computed functions and the fixpoints over D (Theorem 2). Our main interest is in the class of fixpoints (over D or D υ {ω}) which are also computed functions of the recursive definition.