EMNLP 2015 Papers

Monolingual alignment is the task of pairing semantically similar units from two pieces of text. We report a top-performing supervised aligner that operates on short text snippets. We employ a large feature set to (1) encode similarities among semantic units (words and named entities) in context, and (2) address cooperation and competition for alignment among units in the same snippet. These features are deployed in a two-stage logistic regression framework for alignment. On two benchmark data sets, our aligner achieves F1 scores of 92.1% and 88.5%, with statistically significant error reductions of 4.8% and 7.3% over the previous best aligner. It produces top results in extrinsic evaluation as well.

We introduce a new approach to argumentation mining that we applied to a parallel German/English corpus of short texts annotated with argumentation structure. We focus on structure prediction, which we break into a number of subtasks: relation identification, central claim identification, role classification, and function classification. Our new model jointly predicts different aspects of the structure by combining the different subtask predictions in the edge weights of an evidence graph; we then apply a standard MST decoding algorithm. This model not only outperforms two reasonable baselines and two datadriven models of global argument structure for the difficult subtask of relation identification, but also improves the results for central claim identification and function classification and it compares favorably to a complex mstparser pipeline.

In this paper we present the first ever, to the best of our knowledge, discourse parser for multi-party chat dialogues. Discourse in multi-party dialogues dramatically differs from monologues since threaded conversations are commonplace rendering prediction of the discourse structure compelling. Moreover, the fact that our data come from chats renders the use of syntactic and lexical information useless since people take great liberties in expressing themselves lexically and syntactically. We use the dependency parsing paradigm as has been done in the past (Muller et al., 2012; Li et al., 2014). We learn local probability distributions and then use MST for decoding. We achieve 0.680 F1 on unlabelled structures and 0.516 F1 on fully labeled structures which is better than many state of the art systems for monologues, despite the inherent difficulties that multi-party chat dialogues have.

We investigate dual decomposition for joint MAP inference of many strings. Given an arbitrary graphical model, we decompose it into small acyclic sub-models, whose MAP configurations can be found by finite-state composition and dynamic programming. We force the solutions of these subproblems to agree on overlapping variables, by tuning Lagrange multipliers for an adaptively expanding set of variable-lengthn-gram count features. This is the first inference method for arbitrary graphical models over strings that does not require approximations such as random sampling, message simplification, or a bound on string length. Provided that the inference method terminates, it gives a certificate of global optimality (though MAP inference in our setting is undecidable in general). On our global phonological inference problems, it always terminates, and achieves more accurate results than max-product and sum-product loopy belief propagation.

Neural networks have been shown to improve performance across a range of natural-language tasks. However, designing and training them can be complicated. Frequently, researchers resort to repeated experimentation to pick optimal settings. In this paper, we address the issue of choosing the correct number of units in hidden layers. We introduce a method for automatically adjusting network size by pruning out hidden units through‘1,1 and ‘2,1 regularization. We apply this method to language modeling and demonstrate its ability to correctly choose the number of hidden units while maintaining perplexity. We also include these models in a machine translation decoder and show that these smaller neural models maintain the significant improvements of their unpruned versions.

This paper proposes a novel hierarchical recurrent neural network language model (HRNNLM) for document modeling. After establishing a RNN to capture the coherence between sentences in a document, HRNNLM integrates it as the sentence history information into the word level RNN to predict the word sequence with cross-sentence contextual information. A two-step training approach is designed, in which sentence-level and word-level language models are approximated for the convergence in a pipeline style. Examined by the standard sentence reordering scenario, HRNNLM is proved for its better accuracy in modeling the sentence coherence. And at the word level, experimental results also indicate a significant lower model perplexity, followed by a practical better translation result when applied to a Chinese-English document translation reranking task.

We ask how much information a human translator adds to an original text, and we provide a bound. We address this question in the context of bilingual text compression: given a source text, how many bits of additional information are required to specify the target text produced by a human translator? We develop new compression algorithms and establish a benchmark task.

Extracting named entities in text and linking extracted names to a given knowledge base are fundamental tasks in applications for text understanding. Existing systems typically run a named entity recognition (NER) model to extract entity names first, then run an entity linking model to link extracted names to a knowledge base. NER and linking models are usually trained separately, and the mutual dependency between the two tasks is ignored. We propose JERL, Joint Entity Recognition and Linking, to jointly model NER and linking tasks and capture the mutual dependency between them. It allows the information from each task to improve the performance of the other. To the best of our knowledge, JERL is the first model to jointly optimize NER and linking tasks together completely. In experiments on the CoNLL’03/AIDA data set, JERL outperforms state-of-art NER and linking systems, and we find improvements of 0.4% absolute F1 for NER on CoNLL’03, and 0.36% absolute precision@1 for linking on AIDA.

We propose FINET, a system for detecting the types of named entities in short inputs—such as sentences or tweets—with respect to WordNet’s super fine-grained type system. FINET generates candidate types using a sequence of multiple extractors, ranging from explicitly mentioned types to implicit types, and subsequently selects the most appropriate using ideas from word-sense disambiguation. FINET combats data scarcity and noise from existing systems: It does not rely on supervision in its extractors and generates training data for type selection from WordNet and other resources. FINET supports the most fine-grained type system so far, including types with no annotated training data. Our experiments indicate that FINET outperforms state-of-the-art methods in terms of recall, precision, and granularity of extracted types.

We present a novel model for the task of joint mention extraction and classification. Unlike existing approaches, our model is able to effectively capture overlapping mentions with unbounded lengths. The model is highly scalable, with a time complexity that is linear in the number of words in the input sentence and linear in the number of possible mention classes. Our model can be extended to additionally capture mention heads explicitly in a joint manner under the same time complexity. We demonstrate the effectiveness of our model through extensive experiments on standard datasets.

Cross-document co-reference resolution (CCR) computes equivalence classes over textual mentions denoting the same entity in a document corpus. Named-entity linking (NEL) disambiguates mentions onto entities present in a knowledge base (KB) or maps them to null if not present in the KB. Traditionally, CCR and NEL have been addressed separately. However, such approaches miss out on the mutual synergies if CCR and NEL were performed jointly. This paper proposes C3EL, an unsupervised framework combining CCR and NEL for jointly tackling both problems. C3EL incorporates results from the CCR stage into NEL, and vice versa: additional global context obtained from CCR improves the feature space and performance of NEL, while NEL in turn provides distant KB features for already disambiguated mentions to improve CCR. The CCR and NEL steps are interleaved in an iterative algorithm that focuses on the highest-confidence still unresolved mentions in each iteration. Experimental results on two different corpora, news-centric and web-centric, demonstrate significant gains over state-of-the-art baselines for both CCR and NEL.

We present a general framework for comparing multiple groups of documents. A bipartite graph model is proposed where document groups are represented as one node set and the comparison criteria are represented as the other node set. Using this model, we present basic algorithms to extract insights into similarities and differences among the document groups. Finally, we demonstrate the versatility of our framework through an analysis of NSF funding programs for basic research.

Multi-label text categorization is a type of text categorization, where each document is assigned to one or more categories. Recently, a series of methods have been developed, which train a classifier for each label, organize the classifiers in a partially ordered structure and take predictions produced by the former classifiers as the latter classifiers’ features. These predictions-asfeatures style methods model high order label dependencies and obtain high performance. Nevertheless, the predictionsas-features methods suffer a drawback. When training a classifier for one label, the predictions-as-features methods can model dependencies between former labels and the current label, but they can’t model dependencies between the current label and the latter labels. To address this problem, we propose a novel joint learning algorithm that allows the feedbacks to be propagated from the classifiers for latter labels to the classifier for the current label. We conduct experiments using real-world textual data sets, and these experiments illustrate the predictions-as-features models trained by our algorithm outperform the original models.

Distributed word representations are very useful for capturing semantic information and have been successfully applied in a variety of NLP tasks, especially on English. In this work, we innovatively develop two component-enhanced Chinese character embedding models and their bigram extensions. Distinguished from English word embeddings, our models explore the compositions of Chinese characters, which often serve as semantic indictors inherently. The evaluations on both word similarity and text classification demonstrate the effectiveness of our models.

We present an unsupervised method to find lexical variations in Roman Urdu informal text. Our method includes a phonetic algorithm UrduPhone, a featurebased similarity function, and a clustering algorithm Lex-C. UrduPhone encodes roman Urdu strings to their phonetic equivalent representations. This produces an initial grouping of different spelling variations of a word. The similarity function incorporates word features and their context. Lex-C is a variant of k-medoids clustering algorithm that group lexical variations. It incorporates a similarity threshold to balance the number of clusters and their maximum similarity. We test our system on two datasets of SMS and blogs and show an f-measure gain of up to 12% from baseline systems.

This paper presents a new algorithm to automatically solve algebra word problems. Our algorithm solves a word problem via analyzing a hypothesis space containing all possible equation systems generated by assigning the numbers in the word problem into a set of equation system templates extracted from the training data. To obtain a robust decision surface, we train a log-linear model to make the margin between the correct assignments and the false ones as large as possible. This results in a quadratic programming (QP) problem which can be efficiently solved. Experimental results show that our algorithm achieves 79.7% accuracy, about 10% higher than the state-of-the-art baseline (Kushman et al., 2014).

Wikipedia is the largest collection of encyclopedic data ever written in the history of humanity. Thanks to its coverage and its availability in machine-readable format, it has become a primary resource for largescale research in historical and cultural studies. In this work, we focus on the subset of pages describing persons, and we investigate the task of recognizing biographical sections from them: given a person’s page, we identify the list of sections where information about her/his life is present. We model this as a sequence classification problem, and propose a supervised setting, in which the training data are acquired automatically. Besides, we show that six simple features extracted only from the section titles are very informative and yield good results well above a strong baseline.

Improving the search and browsing experience in PubMed r is a key component in helping users detect information of interest. In particular, when exploring a novel field, it is important to provide a comprehensive view for a specific subject. One solution for providing this panoramic picture is to find sub-topics from a set of documents. We propose a method that finds sub-topics that we refer to as themes and computes representative titles based on a set of documents in each theme. The method combines a thematic clustering algorithm and the Pool Adjacent Violators algorithm to induce significant themes. Then, for each theme, a title is computed using PubMed document titles and theme-dependent term scores. We tested our system on five disease sets from OMIM r and evaluated the results based on normalized point-wise mutual information and MeSH r terms. For both performance measures, the proposed approach outperformed LDA. The quality of theme titles were also evaluated by comparing them with manually created titles.

This paper addresses text categorization problem that training data may derive from a different time period from the test data. We present a learning framework which extends a boosting technique to learn accurate model for timeline adaptation. The results showed that the method was comparable to the current state-of-theart biased-SVM method, especially the method is effective when the creation time period of the test data differs greatly from the training data.