Background and Related Work | Our SR-TSG work is built upon recent work on Bayesian TSG induction from parse trees (Post and Gildea, 2009; Cohn et al., 2010). |
Background and Related Work | A derivation is a process of forming a parse tree . |
Background and Related Work | Figure la shows an example parse tree and Figure lb shows its example TSG derivation. |
Inference | We use Markov Chain Monte Carlo (MCMC) sampling to infer the SR-TSG derivations from parse trees . |
Inference | We first infer latent symbol subcategories for every symbol in the parse trees , and then infer latent substitution sites stepwise. |
Inference | After that, we unfiX that assumption and infer latent substitution sites given symbol-refined parse trees . |
Symbol-Refined Tree Substitution Grammars | As with previous work on TSG induction, our task is the induction of SR-TSG derivations from a corpus of parse trees in an unsupervised fashion. |
Symbol-Refined Tree Substitution Grammars | That is, we wish to infer the symbol subcategories of every node and substitution site (i.e., nodes where substitution occurs) from parse trees . |
Abstract | In this work, we further extend this line of exploration and propose a novel but simple approach, which utilizes a ranking model based on word order precedence in the target language to reposition nodes in the syntactic parse tree of a source sentence. |
Experiments | None means the original sentences without reordering; Oracle means the best permutation allowed by the source parse tree ; ManR refers to manual reorder rules; Rank means ranking reordering model. |
Experiments | On the other hand, the performance of the ranking reorder model still fall far short of oracle, which is the lowest crossing-link number of all possible permutations allowed by the parse tree . |
Introduction | The most notable solution to this problem is adopting syntaX-based SMT models, especially methods making use of source side syntactic parse trees . |
Introduction | One is tree-to-string model (Quirk et al., 2005; Liu et al., 2006) which directly uses source parse trees to derive a large set of translation rules and associated model parameters. |
Introduction | The other is called syntax pre-reordering — an approach that re-positions source words to approximate target language word order as much as possible based on the features from source syntactic parse trees . |
Word Reordering as Syntax Tree Node Ranking | Given a source side parse tree T6, the task of word reordering is to transform Te to T4, so that 6’ can match the word order in target language as much as possible. |
Word Reordering as Syntax Tree Node Ranking | By permuting tree nodes in the parse tree , the source sentence is reordered into the target language order. |
Word Reordering as Syntax Tree Node Ranking | parse tree , we can obtain the same word order of Japanese translation. |
Abstract | This paper presents a higher-order model for constituent parsing aimed at utilizing more local structural context to decide the score of a grammar rule instance in a parse tree . |
Conclusion | This paper has presented a higher-order model for constituent parsing that factorizes a parse tree into larger parts than before, in hopes of increasing its power of discriminating the true parse from the others without losing tractability. |
Higher-order Constituent Parsing | Figure l: A part of a parse tree centered at NP —> NP VP |
Higher-order Constituent Parsing | A part in a parse tree is illustrated in Figure 1. |
Introduction | Previous discriminative parsing models usually factor a parse tree into a set of parts. |
Introduction | It allows multiple adjacent grammar rules in each part of a parse tree , so as to utilize more local structural context to decide the plausibility of a grammar rule instance. |
Background | We focus on methods that perform transformations over parse trees , and highlight the search challenge with which they are faced. |
Background | In our domain, each state is a parse tree , which is expanded by performing all applicable transformations. |
Search for Textual Inference | Let t be a parse tree , and let 0 be a transformation. |
Search for Textual Inference | Denoting by tT and tH the text parse tree and the hypothesis parse tree , a proof system has to find a sequence 0 with minimal cost such that tT lO m. This forms a search problem of finding the lowest-cost proof among all possible proofs. |
Search for Textual Inference | Next, for a transformation 0, applied on a parse tree If, we define arequiredfi, 0) as the subset of 75’s nodes required for applying 0 (i.e., in the absence of these nodes, 0 could not be applied). |
Evaluation | In order to compare both approaches, parse trees generated by BKYc were automatically transformed in trees with the same MWE annotation scheme as the trees generated by BKY. |
MWE-dedicated Features | The reranker templates are instantiated only for the nodes of the candidate parse tree , which are leaves dominated by a MWE node (i.e. |
MWE-dedicated Features | dominated by a MWE node m in the current parse tree p, |
Method | Since EDU boundaries are highly correlated with the syntactic structures embedded in the sentences, EDU segmentation is a relatively trivial step — using machine- generated syntactic parse trees , HILDA achieves an F -score of 93.8% for EDU segmentation. |
Method | HILDA’s features: We incorporate the original features used in the HILDA discourse parser with slight modification, which include the following four types of features occurring in SL, SR, or both: (1) N-gram prefixes and suffixes; (2) syntactic tag prefixes and suffixes; (3) lexical heads in the constituent parse tree ; and (4) PCS tag of the dominating nodes. |
Related work | They showed that the production rules extracted from constituent parse trees are the most effective features, while contextual features are the weakest. |
Experiments | We used syntactic features (i.e., features obtained from the dependency parse tree of a sentence) and lexical features, and entity types, which essentially correspond to the ones developed by Mintz et a1. |
Knowledge-based Distant Supervision | Since two entities mentioned in a sentence do not always have a relation, we select entity pairs from a corpus when: (i) the path of the dependency parse tree between the corresponding two named entities in the sentence is no longer than 4 and (ii) the path does not contain a sentence-like boundary, such as a relative clause1 (Banko et al., 2007; Banko and Etzioni, 2008). |
Wrong Label Reduction | We define a pattern as the entity types of an entity pair2 as well as the sequence of words on the path of the dependency parse tree from the first entity to the second one. |
Inference | Following previous work, we design a blocked Metropolis-Hastings sampler that samples derivations per entire parse trees all at once in a joint fashion (Cohn and Blunsom, 2010; Shindo et al., 2011). |
Introduction | Recent work that incorporated Dirichlet process (DP) nonparametric models into TSGs has provided an efficient solution to the problem of segmenting training data trees into elementary parse tree fragments to form the grammar (Cohn et al., 2009; Cohn and Blunsom, 2010; Post and Gildea, 2009). |
Introduction | Figure 2: TIG-to-TSG transform: (a) and (b) illustrate transformed TSG derivations for two different TIG derivations of the same parse tree structure. |