Search

The existing relation extraction approaches select the relevant partial dependency structures and exhibit the limitation associated with long-distance dependencies. Moreover, due to the repeated use of the irrelevant redundant information and the lack of consideration of the key semantic details, the extraction of relations is relatively complex when the entities overlap. To address this limitation and effectively exploit the relevant information while ignoring irrelevant information, this paper proposes a simple but effective multistream semantics-guided dynamic aggregation graph convolution network (SG-DAGCN) to realize the extraction of overlapping relations. The proposed model constructs the entity relation graphs by enumerating the possible candidates and external auxiliary information and adaptively manages the relevant substructure. Subsequently, this framework models the relational graphs between the entities through a dynamic aggregation graph convolution module and gradually produces the discriminative embedded features and a refined graph through the dynamic aggregation of nodes. The proposed approach can effectively leverage the rich multiscale structural information and capture the long-distance dependencies between overlapping entities in long sentences. The results of the experiments conducted on two typical benchmark datasets show that the proposed model can achieve a high level of performance and outperform other state-of-the-art methods in both qualitative and quantitative aspects.

In wh-in-situ languages like Mandarin Chinese, in which the wh-phrase remains in a canonical argument position in wh-questions, syntactic theories generally posit that a covert dependency between the in-situ position and a clause-initial syntactic operator must nonetheless hold at logical form. Wh-in-situ languages and wh-fronted languages are in this way abstractly similar. This paper investigates whether the processing of Mandarin wh-in-situ questions indeed involves constructing a long-distance dependency. Using the multiple-response speed-accuracy tradeoff (SAT) paradigm, we show that Chinese wh-in-situ questions incur more processing costs than their non-wh counterparts. Furthermore, the length of the covert dependency affects only processing accuracy, but not processing speed. This pattern suggests a content-addressable memory process underlying the construction of wh-in-situ dependencies, similar to overt long distance dependencies in English. © 2013 Springer Science+Business Media Dordrecht.

To make sense of a sentence, a reader must keep track of dependent relationships between words, such as between a verb and its particle (e.g. turn the music down). In languages such as German, verb-particle dependencies often span long distances, with the particle only appearing at the end of the clause. This means that it may be necessary to process a large amount of intervening sentence material before the full verb of the sentence is known. To facilitate processing, previous studies have shown that readers can preactivate the lexical information of neighbouring upcoming words, but less is known about whether such preactivation can be sustained over longer distances. We asked the question, do readers preactivate lexical information about long-distance verb particles? In one self-paced reading and one eye tracking experiment, we delayed the appearance of an obligatory verb particle that varied only in the predictability of its lexical identity. We additionally manipulated the length of the delay in order to test two contrasting accounts of dependency processing: that increased distance between dependent elements may sharpen expectation of the distant word and facilitate its processing (an antilocality effect), or that it may slow processing via temporal activation decay (a locality effect). We isolated decay by delaying the particle with a neutral noun modifier containing no information about the identity of the upcoming particle, and no known sources of interference or working memory load. Under the assumption that readers would preactivate the lexical representations of plausible verb particles, we hypothesised that a smaller number of plausible particles would lead to stronger preactivation of each particle, and thus higher predictability of the target. This in turn should have made predictable target particles more resistant to the effects of decay than less predictable target particles. The eye tracking experiment provided evidence that higher predictability did facilitate reading times, but found evidence against any effect of decay or its interaction with predictability. The self-paced reading study provided evidence against any effect of predictability or temporal decay, or their interaction. In sum, we provide evidence from eye movements that readers preactivate long-distance lexical content and that adding neutral sentence information does not induce detectable decay of this activation. The findings are consistent with accounts suggesting that delaying dependency resolution may only affect processing if the intervening information either confirms expectations or adds to working memory load, and that temporal activation decay alone may not be a major predictor of processing time. [ABSTRACT FROM AUTHOR]

Three eye movement experiments investigated the processing of the syntactic ambiguity in strings such as the information that the health department provided , where the that -clause can be either a relative clause (RC) or the start of a nominal complement clause (CC; the information that the health department provided a cure ). The experiments tested the prediction that comprehenders should avoid the RC analysis because it involves an unforced filler-gap dependency. Readers showed difficulty upon disambiguation toward the RC analysis, and showed facilitated processing of the ambiguous material itself when the CC analysis was available; both patterns suggest rapid initial adoption of the CC analysis in preference to the RC analysis. The strength of the bias of a specific head noun (e.g., information ) to appear with a CC did not modulate these effects, nor were these effects reliably modulated by the tendency of an ambiguous string to be completed off-line as a CC or an RC. These results add to the evidence that structural principles guide the processing of filler-gap dependencies. [ABSTRACT FROM AUTHOR]

Wh-in-situ constructions in Mandarin Chinese, as opposed to their English counterparts that front wh-phrases to the beginning of the sentence, have the same word order as regular non-wh declaratives. We argue that despite their surface word order, processing wh-in-situ constructions involves constructing a covert non-local syntactic dependency between the in-situ wh-phrase and a higher scope position at a clause boundary, leading to behavioral patterns similar to those associated with the processing of overt dependencies. In two comprehension experiments, we showed that the process of linking an in-situ wh-phrase and its scope position induces a similarity-based memory interference effect if another clause boundary position intervenes. In addition, a set of sentence completion studies also showed that the production of wh-in-situ constructions is heavily modulated by the increased working memory burden that results from planning and maintaining a non-local dependency. [ABSTRACT FROM AUTHOR]

Semantic annotation of Web objects is a key problem for Web information extraction. The Web contains an abundance of useful semi-structured information about real world objects, and the empirical study shows that strong two-dimensional sequence characteristics and correlative characteristics exist for Web information about objects of the same type across different Web sites. Conditional Random Fields (CRFs) are the state-of-the-art approaches taking the sequence characteristics to do better labeling. However, as the appearance of correlative characteristics between Web object elements, previous CRFs have their limitations for semantic annotation of Web objects and cannot deal with the long distance dependencies between Web object elements efficiently. To better incorporate the long distance dependencies, on one hand, this paper describes long distance dependencies by correlative edges, which are built by making good use of structured information and the characteristics of records from external databases; and on the other hand, this paper presents a two-dimensional Correlative-Chain Conditional Random Fields (2DCC-CRFs) to do semantic annotation of Web objects. This approach extends a classic model, two-dimensional Conditional Random Fields (2DCRFs), by adding correlative edges. Experimental results using a large number of real-world data collected from diverse domains show that the proposed approach can significantly improve the semantic annotation accuracy of Web objects. [ABSTRACT FROM AUTHOR]