Papers reviewed and determined not to be word norm studies. Use the flag icon to report errors or suggest re-inclusion.
18265 papers
Abstract: This technical note describes the US Army Research Laboratory (ARL) Arabic Dependency Treebank (AADT) for the purpose of documenting its release. The AADT was derived from existing Arabic treebanks distributed by the Linguistic Data Consortium using constituent-to-dependency conversion software written at ARL. Earlier versions of the AADT, as well as parsers trained from it, have been used in several published ARL research efforts, and, by releasing the data, we hope to facilitate additional Arabic language processing research by the greater community.
We release Galactic Dependencies 1.0—a large set of synthetic languages not found on Earth, but annotated in Universal Dependencies format. This new resource aims to provide training and development data for NLP methods that aim to adapt to unfamiliar languages. Each synthetic treebank is produced from a real treebank by stochastically permuting the dependents of nouns and/or verbs to match the word order of other real languages. We discuss the usefulness, realism, parsability, perplexity, and diversity of the synthetic languages. As a simple demonstration of the use of Galactic Dependencies, we consider single-source transfer, which attempts to parse a real target language using a parser trained on a “nearby” source language. We find that including synthetic source languages somewhat increases the diversity of the source pool, which significantly improves results for most target languages.
Background Conventionally, it is believed that high-frequency auditory information is important for speech understanding. This is only partly true, as recent studies have demonstrated the importance of low-frequency information. This research was taken up to develop, standardize, and validate auditory low-frequency word lists in Hindi, an Indian language. Material and Methods The first phase of the study involved collection of bisyllabic words followed by verification by a native linguist. Words were then short-listed based on familiarity ratings given by 10 adult native speakers; those words were recorded and the best recorded words selected through subjective and objective analysis. Then, using Fast Fourier Transform and k-means clustering, words with more energy below 1.5 kHz were isolated. Finally, equally difficult 10 word lists were generated by obtaining psychometric function curves. Finally, lists were administered on 40 adult normal hearing particip Results Results showed a similar trend of increase in speech identification scores with increase in SL across all lists except list 4. During the final phase, developed lists were validated on 10 simulated low-frequency cochlear hearing loss participants. Hearing loss was simulated using Matlab and National Institute for Occupational Safety and Health (NIOSH) software. Results of validation revealed that auditory low-frequency word lists were sensitive enough to tap the speech understanding difficulty in the simulated condition. Conclusions The developed word lists can be used clinically to assess communication ability in individuals with rising hearing loss. The word lists also have the potential to assess the performance after amplification provided to individuals with rising hearing loss.
Detecting automatically the cause relations of a text may be useful in question answering tasks and event information extraction. The aim of this paper is to study how to detect coherence relations of the cause subgroup (CAUSE, RESULT and PURPOSE). TO achieve this aim we have used the Rhetorical Structure Theory (RST) and some automatic linguistic information from different tools developed by IXA Group. We have used a corpus of 60 scientific abstracts, the Basque RST Treebank (Iruskieta et al., 2013), of different domains: science, medicine and terminology. A linguist has annotated all the signals of that corpus and described the most important problems in such task. To report the reliability of this annotator, two linguists have annotated the signals of the cause subgroup and all the annotations were compared and evaluated. After that, a superannotator has harmonized all the signals of those cause relations. Finally, we show the most important signals for such relations.
Semantic similarities are a cross-field research in Natural Language Processing and Ontologies with some possible fallout in Artificial Intelligence. Formerly, similarities were computed following a syntactical treatment to support case-based reasoning. Textual similarities are now guided by semantic machineries, offering various ways to compute relatedness measures. In this paper, we present both a logical and a visual framework aiming to reason with them. For that reason, we introduced FLH±, a fragment of description logic underpinning the well-known lexical database Wordnet. We illustrated this framework with the path length relatedness, one of the historical similarity measures occurring in a taxonomy. The core of our framework orchestrates the computation of similarity scores supported by REVERB, STANFORD CORENLP and WORDNET:SIMILARITY APIs and interfaces global similarities in graphical way by positioning them on segments. We also depicted some experimental results to confront our computational framework with some empirical data.
Feedforward Neural Network (FNN)-based language models estimate the probability of the next word based on the history of the last N words, whereas Recurrent Neural Networks (RNN) perform the same task based only on the last word and some context information that cycles in the network. This paper presents a novel approach, which bridges the gap between these two categories of networks. In particular, we propose an architecture which takes advantage of the explicit, sequential enumeration of the word history in FNN structure while enhancing each word representation at the projection layer through recurrent context information that evolves in the network. The context integration is performed using an additional word-dependent weight matrix that is also learned during the training. Extensive experiments conducted on the Penn Treebank (PTB) and the Large Text Compression Benchmark (LTCB) corpus showed a significant reduction of the perplexity when compared to state-of-the-art feedforward as well as recurrent neural network architectures.
情绪调节指的是个体对情绪的发生、体验与表达进行调控的能力和过程。良好的情绪调节能力有利于个体保持愉快的心境、改善不利的心境。现有的情绪调节研究大多都局限于负性情绪调节，而正性情绪调节的研究一直很少，当前研究综合行为和电生理方法，对被试在进行正性情绪认知重评调节时的唤醒度和颧肌肌电(zygomatic electromyography，简称zEMG)进行分析。结果发现两个实验中当进行正性情绪上调调节时，唤醒度和肌电指标相比维持调节时都有显著的增强，而下调与维持时肌电差异不显著。两个实验结果一致的证明正性情绪上调效应显著，而下调效应不显著。这表明相对于抑制正性情绪，人们可能更习惯和倾向于增强自身的正性情绪。 Emotion regulation refers to the ability and the process that individual adjusts and controls the occurrence, experience and expression of emotion. Good emotion regulation ability is beneficial for individuals to keep pleasant mood and improve the bad one. Although there are many studies investigating emotion regulation, they are mostly about negative emotion regulation, and few are known about the study of positive emotion regulation. In the current study, we use behavioral and electrophysiology measure of arousal rating and zygomatic electromyography (zEMG) to index the variation of positive emotion regulation. We collect arousal rating in each trial after participants regulate their emotion in Experiment 1 and the activity of zEMG in Experiment 2 in which the par-ticipants regulate their emotions induced by International Affective Picture System pictures. The results indicate that up-regulated effect of positive emotion regulation is significant, but down- regulated effect is not. This suggests that people may be desirable and habitual to increase their positive emotion rather than inhibit it.
Detecting automatically the cause relations of a text may be useful in question answering tasks and event information extraction. The aim of this paper is to study how to detect coherence relations of the cause subgroup (Cause, Result and Purpose). To achieve this aim we have used the Rhetorical Structure Theory (RST) and some automatic linguistic information from different tools developed by IXA Group. We have used a corpus of 60 scientific abstracts, the Basque RST Treebank (Iruskieta et al., 2013), of different domains: science, medicine and terminology. A linguist has annotated all the signals of that corpus and described the most important problems in such task. To report the reliability of this annotator, two linguists have annotated the signals of the cause subgroup and all the annotations were compared and evaluated. After that, a superannotator has harmonized all the signals of those cause relations. Finally, we show the most important signals for such relations.
Due to the constant increasing of electronic textual information, modern society needs for the automatic processing of natural language (NL). The main purpose of NL automatic text processing systems is to analyze and create texts and represent their content. The purpose of the paper is the development of linguistic and software bases of an automatic system for processing English publicistic texts. This article discusses the examples of different approaches to the creation of linguistic databases for processing systems. The author gives a detailed description of basic building blocks for a new linguistic processor: lexicalsemantic, syntactical and semantic-syntactical. The main advantage of the processor is using special semantic codes in the alphabetical dictionary. The semantic codes have been developed in accordance with a lexical-semantic classification. It helps to precisely define semantic functions of the keywords that are situated in parsing groups and allows the automatic system to avoid typical mistakes. The author also represents the realization of a developed linguistic database in the form of a training computer program.
This volume takes as its central organizing principle the foundational understanding about community knowledge that challenges “narrow conceptions of language, literacy, personal stories, bounded or contained learning contexts (e.g., home, community, schools), hegemonic cultural and linguistic norms, quantitative and static views of ‘resources,’ and limited attention to the agency, identities, and strategic actions of diverse students and their families as they traverse contexts” (daSilva Iddings, this volume). The touchstone to this approach is “Funds of Knowledge” as it has been conceptualized for nearly twenty-five years. This chapter will briefly summarize the approach as it has evolved and will lay out the programmatic implementation as it unfolded within CREATE.
In recent years there has been a lot of interest in cross-lingual parsing for developing treebanks for languages with small or no annotated treebanks. In this paper, we explore the development of a cross-lingual transfer parser from Hindi to Bengali using a Hindi parser and a Hindi-Bengali parallel corpus. A parser is trained and applied to the Hindi sentences of the parallel corpus and the parse trees are projected to construct probable parse trees of the corresponding Bengali sentences. Only about 14% of these trees are complete (transferred trees contain all the target sentence words) and they are used to construct a Bengali parser. We relax the criteria of completeness to consider well-formed trees (43% of the trees) leading to an improvement. We note that the words often do not have a one-to-one mapping in the two languages but considering sentences at the chunk-level results in better correspondence between the two languages. Based on this we present a method to use chunking as a preprocessing step and do the transfer on the chunk trees. We find that about 72% of the projected parse trees of Bengali are now well-formed. The resultant parser achieves significant improvement in both Unlabeled Attachment Score (UAS) as well as Labeled Attachment Score (LAS) over the baseline word-level transferred parser.
espanolCon el desarrollo de la informatica, en la investigacion del lenguaje se introdujo la teoria y metodologia de redes complejas, que transforma el sistema de la lengua en las redes complejas compuestas de nodos y enlaces para hacer un analisis cuantitativo de la estructura de la lengua. El desarrollo de la gramatica de dependencias proporciona un apoyo teorico a la construccion del corpus anotado (treebank), por lo que el analisis estadistico con las redes complejas se hace posible. Este articulo presenta la teoria y metodologia de las redes complejas y construye las redes sintacticas de dependencia a base del corpus anotado (treebank) de las expresiones orales del examen EEE-4 (Examen del Espanol como Especialidad - Nivel 4). Mediante el analisis de las caracteristicas generales de las redes, incluyendo el numero de nodos, los enlaces, el grado medio, la longitud media de los caminos, la distribucion de grados y la centralizacion, tiene como objetivo descubrir la diferencia y similitud potencial entre las expresiones orales de distintos niveles. Ademas, con el analisis de conglomerados, esta investigacion pretende demostrar la capacidad discriminatoria de las variables de las redes complejas y proporcionar una referencia potencial para el trabajo de calificacion. EnglishWith the development of information technology, the theory and methodology of complex network has been introduced to the language research, which transforms the system of language in a complex networks composed of nodes and edges for the quantitative analysis about the language structure. The development of dependency grammar provides theoretical support for the construction of a treebank corpus, making possible a statistic analysis of complex networks. This paper introduces the theory and methodology of the complex network and builds dependency syntactic networks based on the treebank of speeches from the EEE-4 oral test. According to the analysis of the overall characteristics of the networks, including the number of edges, the number of the nodes, the average degree, the average path length, the network centrality and the degree distribution, it aims to find in the networks potential difference and similarity between various grades of speaking performance. Through clustering analysis, this research intends to prove the network parameters’ discriminating feature and provide potential reference for scoring speaking performance.
The article deals with the phenomenon of diglossia in the context of development of Greek language in the Hellenistic period, observes in diachrony the correlation between the linguistic norm of Atticism and Koine Greek – from the beginning of theory of Atticist mimesis to the times of second sophistry; determines the specificity of Koine Greek of the New Testament and its relation to, on the one hand, Atticism canon and literary Koine Greek, and on the other hand, – to the language of the early Patristic literature.
Dataset for TACL submission "The Galactic Dependencies Treebanks: Getting More Data by Synthesizing New Languages".<br> The scripts and model parameters for replicating this dataset are available at https://github.com/gdtreebank/gdtreebank.
Universal Dependencies (UD) are gaining much attention of late for systematic evaluation of cross-lingual techniques for crosslingual dependency parsing. In this paper we present our work in line with UD. Our contribution to this is manifold. We extend UD to Indian languages through conversion of Pān inian Dependencies to UD for the Hindi Dependency Treebank (HDTB). We discuss the differences in annotation in both the schemes, present parsing experiments for both the formalisms and empirically evaluate their weaknesses and strengths for Hindi. We produce an automatically converted Hindi Treebank conforming to the international standard UD scheme, making it useful as a resource for multilingual language technology.
Nonostante una secolare tradizione lessicografica, la lingua latina manca ancora di risorse lessicali di tipo computazionale aggiornate allo stato dell’arte. Ciò è strettamente connesso alla limitata disponibilità di corpora testuali latini annotati linguisticamente, sulla cui base empirica possano essere costruite nuove risorse lessicali. Tuttavia, una serie di progetti mirati allo sviluppo di avanzate risorse linguistiche per il latino (tra cui alcune treebank) è stata avviata nel corso dell’ultimo decennio. In questo articolo, presentiamo Latin Vallex, un lessico di valenza per il latino realizzato in stretta connessione con l’annotazione semantico-pragmatica di due treebank latine comprensive di testi di epoche e generi diversi. Ciò consente di connettere biunivocamente le strutture valenziali registrate nel lessico e le loro occorrenze nei dati testuali delle treebank.
We describe results of investigation of a specific type of discontinuous constructions, namely non-projective constructions concerning verbs and their arguments. This topic is especially important for languages with a relatively free word order, such as Czech, which is the language we have primarily worked with. For comparison, we have included some results for English. The corpora used for both languages are the Prague Czech-English Dependency Treebank and the Prague Dependency Treebank, which are both annotated at a dependency syntax level as well as a deep (semantic) level, including verbs and their valency (arguments). We are using traditionally defined non-projectivity on trees with full linear ordering, but the two levels of annotation are innovatively combined to determine if a particular (deep) verb -argument structure is non-projective. As a result, we have identified several types of discontinuities, which we classify either by the verb class or structurally in terms of the verb, its arguments and their dependents. In addition, we have quantitatively compared selected phenomena found in Czech translated texts (in the PCEDT) to the native Czech as found in the original Prague Dependency Treebank.
Tokenizer, POS Tagger, Lemmatizer and Parser models for all 50 languages of Universal Depenencies 2.0 Treebanks, created solely using UD 2.0 data (http://hdl.handle.net/11234/1-1983). The model documentation including performance can be found at http://ufal.mff.cuni.cz/udpipe/users-manual#universal_dependencies_20_models. To use these models, you need UDPipe binary version at least 1.2, which you can download from http://ufal.mff.cuni.cz/udpipe. In addition to models itself, all additional data and value of hyperparameters used for training are available in the second archive, allowing reproducible training.
Nonostante una secolare tradizione lessicografica, la lingua latina manca ancora di risorse lessicali di tipo computazionale aggiornate allo stato dell’arte. Cio e strettamente connesso alla limitata disponibilita di corpora testuali latini annotati linguisticamente, sulla cui base empirica possano essere costruite nuove risorse lessicali. Tuttavia, una serie di progetti mirati allo sviluppo di avanzate risorse linguistiche per il latino (tra cui alcune treebank) e stata avviata nel corso dell’ultimo decennio. In questo articolo, presentiamo Latin Vallex, un lessico di valenza per il latino realizzato in stretta connessione con l’annotazione semantico-pragmatica di due treebank latine comprensive di testi di epoche e generi diversi. Cio consente di connettere biunivocamente le strutture valenziali registrate nel lessico e le loro occorrenze nei dati testuali delle treebank.
The purpose of this paper is to discuss the social legitimacy of the non-dominant variety of French that is used in Belgium (henceforth ‘Belgian French’). As will be detailed, Francophone Belgians’ attitudes have shifted from early 19th c. – late 20th c. purism and subsequent linguistic subjection to France to more recent acceptation of endogenous traits and increasing distance from the Hexagonal model. Nevertheless, these attitudes remain characterized by a “double distance” from both Hexagonal and Belgian French. The idea that French is viewed by Francophone Belgians as a polycentric/polynomic language will thus be questioned: do they really consider that there is a legitimate Belgian variety of French? What is the relevance of the national criterion in the way they define linguistic norms? What other criteria lie behind the definition and legitimization of their linguistic norms?
This dataset introduces a companion reproducibility Java console program, called HESML_vs_SML_test.jar, of the work introduced by Lastra-Díaz and García-Serrano [1]. This latter work introduces the Half-Edge Semantic Measures Library (HESML), and carries-out an experimental survey between HESML V1R2, the Semantic Measures Library (SML) 0.9 [2] and the WNetSS [4] semantic measures libraries. The HESML_vs_SML_test.jar program runs the set of performance and scalability benchmarks detailed in [1] and generates the figures and tables of results reported in the aforementioned work, which are also enclosed as complementary files of this dataset (see files below). Licensing note: The 'HESML_vs_SML_test.jar' program is based on the HESML V1R2 [3], SML 0.9 [2] and WNetSS [4] semantic measures libraries, and it includes these libraries in its distribution, as well as WordNet 3.0 [6] and the SimLex665 [5] dataset. Thus, if you use this dataset, you should also cite the works related to these resources. References: [1] Lastra-Díaz, J. J., and García-Serrano, A. (2016). HESML: a scalable ontology-based semantic similarity measures library with a set of reproducible experiments and a replication dataset. To appear in Information Systems Journal. [2] Harispe, S., Ranwez, S., Janaqi, S., and Montmain, J. (2014). The Semantic Measures Library: Assessing Semantic Similarity from Knowledge Representation Analysis. In E. Métais, M. Roche, & M. Teisseire (Eds.), Proc. of the 19th International Conference on Applications of Natural Language to Information Systems (NLDB 2014) (Vol. 8455, pp. 254–257). Montpelier, France: Springer. http://dx.doi.org/10.1007/978-3-319-07983-7_37 [3] Lastra-Díaz, J. J., & García-Serrano, A. (2016). HESML V1R2 Java software library of ontology-based semantic similarity measures and information content models. Mendeley Data, v2. https://doi.org/10.17632/t87s78dg78.2 [4] Ben Aouicha, M., Taieb, M. A. H., and Ben Hamadou, A. (2016). SISR: System for integrating semantic relatedness and similarity measures. Soft Computing, 1–25. http://dx.doi.org/10.1007/s00500-016-2438-x [5] Hill, F., Reichart, R., & Korhonen, A. (2015). SimLex-999: Evaluating Semantic Models with (Genuine) Similarity Estimation. Computational Linguistics, 41(4), 665–695. http://dx.doi.org/10.1162/COLI_a_00237 [6] Miller, G. A. (1995). WordNet: A Lexical Database for English. Communications of the ACM, 38(11), 39–41. http://dx.doi.org/10.1145/219717.219748
FrameNet is a semi-formal lexical database. There are various attempts to formalize this database. This paper presents a partial formalization of FrameNet in situation theory. This mathematical theory of information may provide a fruitful basis for making FrameNet accessible to other fields such as natural language processing, computational linguistics and informational retrieval.
Research finds we make spontaneous trait inferences from facial appearance, even after brief exposures to a face (i.e., less than or equal to 100 ms). We examined spontaneous impressions of criminality from facial appearance, testing whether these impressions persist after repeated presentation (i.e., one to three exposures) and increased exposure duration (100, 500, or 1,000 ms) to the face. Judgement confidence and response times were recorded. Other participants viewed the faces for an unlimited period of time, rating trustworthiness, dominance and criminal appearance. We found evidence that participants spontaneously make criminal appearance attributions. These inferences persisted with repeated presentation and increased exposure duration, were related to trustworthiness and dominance ratings, and were made with high confidence. Implications are discussed.
In many natural language processing based intelligent systems, parsing is the first task to perform. However, in the next stages, many systems often have the capacity of processing a limited number of parsed structures. The problem is to determine what parsed sentences can be recognized by a system. The decision of syntactic structures which can be processed by a system is consider as the task of "classification" of a parsed sentence into one of given classes of recognizable parses. In this paper we deal with this issue by proposing a method for mapping Vietnamese chunked sentences to a set of pre-defined shallow structures. Also, we tag lexicons and chunk phrases of the original sentences using our Functional Part-of-Speech (FPOS) tagset with Apache OpenNLP tools (Tokenizer, POS Tagger, Chunker). Based on the foundation of Functional Grammar, we define new lexical tags and combine with Penn-Treebank tagset to build our FPOS tagset. Due to our set of shallow structures is finite, instead of using a parser, we propose a rule-based algorithm for the mapping process. We establish conversion rules according to the reality experiences when using Vietnamese in common communication. The experiment shows that we converse successfully for the major of testing sentences and the algorithm can be applied for different languages.
This thesis describes unexpected constructions based on THIS and THAT by French and Spanishlearners of English. Chapter 1 raises the issue of the study of THIS and THAT as markers in the twomicrosystems of proforms and deictics. Chapter 2 covers different types of analyses of referencewith THIS and THAT in native English and refers to different theoretical frameworks (Cornish,Cotte, Halliday & Hasan, Kleiber, Fraser & Joly, Lapaire & Rotgé). It crossreferencesrepresentations (anaphora/deixis; endophoricity/exophoricity) with an analysis of functionalrealisations. Chapter 3 broaches the issue of interlanguage analysis, and it shows that a dynamicsystemic approach grounded in the functional distinction of the forms is necessary. Chapter 4 givesdetails about existing annotation tagsets for English corpora (Penn Treebank, Claws7, ICEGB). Itshows the need for a finergrained annotation relying on functional tags and for semanticinformation on the positions (subject v. oblique). Chapter 5 describes the multilayer annotationstructure which is implemented for the analysis of different corpora. It also covers the methods usedto automatically annotate functional categories (as well as their evaluation), and it justifies thechoices made to support corpus interoperability. Chapter 6 offers a regression analysis whichprovides evidence on the tendencies of the operationalised variables (the L1, the written or spokenmode of the corpora and the type of reference). Chapter 7 examines the role of the previously codedlinguistic properties of the analysis. With the use of classifiers, it describes a system for automaticerror analysis. Chapter 8 concludes on the methodologies used in the thesis and their implicationsin linguistic analysis.
The OPT submission to the Shared Task \nof the 2016 Conference on Natural Language \nLearning (CoNLL) implements a \n‘classic’ pipeline architecture, combining \nbinary classification of (candidate) explicit \nconnectives, heuristic rules for non-explicit \ndiscourse relations, ranking and ‘editing’ \nof syntactic constituents for argument identification, \nand an ensemble of classifiers to \nassign discourse senses. With an end-toend \nperformance of 27.77 F1 on the English \n‘blind’ test data, our system advances \nthe previous state of the art (Wang & Lan, \n2015) by close to four F1 points, with particularly \ngood results for the argument identification \nsub-tasks. OPT system results appear \nmore competitive on the new, ‘blind’ \ntest data than on the ‘test’ and ‘development’ \nsections of the Penn Discourse Treebank \n(PDTB; Prasad et al., 2008), which \nmay indicate reduced over-fitting to specific \nproperties of the venerableWall Street \nJournal (WSJ) text underlying the PDTB
Perirhinal cortex (PrC) has been implicated as a brain region in the medial temporal lobes (MTL) that critically contributes to familiarity-based recognition memory, a process that allows for recognition to occur independently of contextual recollection. Informed by neurophysiological research in non-human primates, fMRI, as well as behavioural work in humans, the current thesis research tests the novel hypothesis that PrC cortex functioning also underlies the ability to assess cumulative lifetime familiarity with object concepts that are characterized by a lifetime of experiences. In Chapter 2, a patient (NB) with a left anterior temporal lobe (ATL) lesion that included PrC as well as an amnesic patient (HC) with a bilateral lesion to the hippocampus were tested on their ability to make lifetime familiarity judgements for object concepts (i.e., concrete nouns). Patient NB made abnormal familiarity ratings for objects concepts relative to matched controls, while patient HC produced ratings that did not differ from control participants. In Chapter 3, I tested healthy young adults on a frequency judgement task and lifetime familiarity task while they underwent fMRI. A region in the left PrC tracked both the perceived frequency of recent laboratory exposure as well as perceived lifetime familiarity. Finally, in Chapter 4, I tested whether indeed lifetime familiarity judgements are based on conceptual processing by making use of an associative priming paradigm. Associatively-related primes increased the perceived familiarity of object concepts while also reducing the latency of these judgements. Overall, the results from all three empirical chapters provides evidence that warrants an extension of PrC functioning to include the cumulative assessment of lifetime familiarity with object concepts.
Prague Czech-English Dependency Treebank - Russian translation (PCEDT-R) is a project of translating a subset of Prague Czech-English Dependency Treebank 2.0 (PCEDT 2.0) to Russian and linguistically annotating the Russian translations with emphasis on coreference and cross-lingual alignment of coreferential expressions. Cross-lingual comparison of coreference means is currently the purpose that drives development of this corpus. The current version 0.5 is a preliminary version, which contains (+ denotes new features): * complete PCEDT 2.0 documents "wsj_1900"-"wsj_1949" * Czech-English word alignment of coreferential expressions annotated manually mainly on the t-layer + Russian translations of the original English sentences + automatic tokenization, part-of-speech tagging and morphological analysis for Russian + automatic word alignment between all Czech and Russian words + manual alignment between Russian and the other two languages on possessive pronouns
A known way to improve the accuracy of dependency parsers is to combine several different parsing algorithms, in such a way that the weaknesses of each of the models can be compensated by the strengths of others. For example, voting-based combination schemes are based on variants of the idea of analyzing each sentence with various parsers, and constructing a combined output where the head of each node is determined by “majority vote” among the different parsers. Typically, such approaches combine very different parsing models to take advan- tage of the variability in the parsing errors they make. In this paper, we show that consistent improvements in accuracy can be obtained in a much simpler way by combining a single parser with itself. In particular, we start with a greedy implementation of the Nivre pseudo-projective arc-eager algorithm, a well-known left-to-right transition-based parser, and we combine it with a “mirrored” version of the algorithm that analyzes sentences from right to left. To determine which of the two obtained outputs we trust for the head of each node, we use simple criteria based on the length and position of dependency arcs. Experiments on several datasets from the CoNLL-X shared task and the WSJ section of the English Penn Treebank show that the novel combination system obtains better performance than the baseline arc-eager parser in all cases. To test the generality of the approach, we also perform experiments with a different transition system (arc-standard) and a different search strategy (beam search), obtaining similar improvements in all these settings.
Recently, these has been a surge on studying how to obtain partially annotated data for model supervision. However, there still lacks a systematic study on how to train statistical models with partial annotation (PA). Taking dependency parsing as our case study, this paper describes and compares two straightforward approaches for three mainstream dependency parsers. The first approach is previously proposed to directly train a log-linear graph-based parser (LLGPar) with PA based on a forest-based objective. This work for the first time proposes the second approach to directly training a linear graph-based parse (LGPar) and a linear transition-based parser (LTPar) with PA based on the idea of constrained decoding. We conduct extensive experiments on Penn Treebank under three different settings for simulating PA, i.e., random dependencies, most uncertain dependencies, and dependencies with divergent outputs from the three parsers. The results show that LLGPar is most effective in learning from PA and LTPar lags behind the graph-based counterparts by large margin. Moreover, LGPar and LTPar can achieve best performance by using LLGPar to complete PA into full annotation (FA).
Hoarding is a complex and impairing psychiatric disorder and a public health problem. Traditionally it is assessed through observation and interview, but recently a new method has been proposed where living quarters of an individual are visually compared with a set of template images ranked according to the “Clutter Image Rating” (CIR) scale from 1 to 9. However, such an assessment is time-consuming, subjective, and weak in repeatability. We propose an automatic method for classifying hoarding images according to the CIR scale. Since clutter in living quarters (e.g., piles of boxes, newspapers, clothing) corresponds to “busy” areas with lots of edges in captured images, we use the histogram-of-gradients (HOG) descriptor to characterize images and estimate the CIR value using two methods: regression and classification. In 4-fold cross-validation on 620 images that we harvested from the internet, both methods result in mean-absolute CIR error of about 1.2. Given the simplicity of our method, this is an encouraging result as it approximates ratings by trained professionals who admit assigning CIR values within ± 1 CIR point.
Continuous word representations appeared to be a useful feature in many natural language processing tasks.Using fixed-dimension pre-trained word embeddings allows avoiding sparse bag-of-words representation and to train models with fewer parameters.In this paper, we use fixed pre-trained word embeddings as additional features for a neural scoring function in the MST parser.With the multi-layer architecture of the scoring function we can avoid handcrafting feature conjunctions.The continuous word representations on the input also allow us to reduce the number of lexical features, make the parser more robust to out-of-vocabulary words, and reduce the total number of parameters of the model.Although its accuracy stays below the state of the art, the model size is substantially smaller than with the standard features set.Moreover, it performs well for languages where only a smaller treebank is available and the results promise to be useful in cross-lingual parsing.
L’informatisation des alignements textuels est confrontée à la complexité de l’organisation textuelle et discursive. L’architecture modulaire Trame/Cadre issue des recherches menées en textométrie facilite la navigation dans l’espace textuel multilingue. Le flux textuel est représenté par un système de coordonnées sur le texte (la Trame ). Le calcul d’une Trame permet une identification précise des objets ( contenants et contenus ) nécessaires aux repérages contextuels (le Cadre ). La construction d’un Cadre permet de stocker non seulement les découpages du texte mais aussi les annotations produites par différentes procédures informatiques (y compris les alignements) et, éventuellement, de les faire passer d’une procédure de traitement à l’autre. Ces états successifs de traitement induisent la notion de ressource textuelle incrémentale qui conserve la trace de séquences de traitement apportées à la ressource textuelle initiale, avec apport de méthodes quantitatives. Cette approche est implémentée au sein du logiciel Le Trameur qui permet d’explorer les corpus multilingues richement annotés ( treebanks ).
It is now widely acknowledged that, between the fifteenth and the seventeenth centuries, most of the European national grammatical traditions were derived from the long-established Graeco-Latin descriptive and normative framework. However, a thorough investigation into the particulars of such a ‘transfer’, at a time when the first grammars of the English vernacular were progressively translated from the Latin, or written directly in English, is still to be carried out. How were classical linguistic norms practically transferred to English? How was usage then looked upon? Why was it decided that the vernacular should be taught? In this paper, I will examine in some detail how the first English grammarians integrated the vernacular of England into the paradigm of Latin grammar. We shall also study the conditions under which the specificities of English were revealed. But our main concern will be to try and determine whether any challenging position to the prevalent model that had sprung from the Graeco-Roman tradition can be traced back. If this long-established tradition appeared at that time as the only one apt to guarantee the efficiency of grammatical description, it seems that it was consistently challenged even by the most prominent figures.
Political misperceptions pose a serious threat to democracy, making it imperative to understand how to correct such false beliefs.Online ratings could play an important role in this process.Research on bandwagon effects suggests that favorable online ratings should help make corrections more persuasive by fostering trust in such messages.The assumption that online ratings are uniformly persuasive is, however, overly simplistic.I argue that online ratings will not always promote acceptance of corrections.In what I term the social affirmation heuristic, I hypothesize that people will only trust ratings of factual corrections that affirm what they already believe, and vice versa.I further predict that rating trust will influence subsequent trust in corrections.Taken together, this means that belief discrepant ratings can have boomerang effects.Instead of eliciting the bandwagon effects described above, favorable ratings promote distrust of belief discrepant corrections.It is, however, possible for belief uncertainty and de-biasing messages to limit these boomerang effects by reducing reliance on the social affirmation heuristic.I expect these predictions to hold for various kinds of ratings, including both star ratings, which indicate rater favorability toward content, and Likes, which indicate the number of raters who see value in the content.This dissertation uses two studies to test these ideas.The first study uses data from an online experiment conducted with convenience sample of 847 participants.The data for the second study come from a nationally representative sample of 500 participants.With the exception of the hypothesis that belief confidence affects rating iii trust, all hypotheses received robust support in the context of star ratings.Implications of these findings are discussed.iv Dedication This work is dedicated to all the academics out there who keep trying, failing, and getting back up again.May you triumph against the odds and prove your detractors wrong.vi My academic journey would not have come to fruition without the parasocial support from my favorite Disney animated features.These movies provided me with much comfort and solace whenever I faced academic struggles.Many thanks to Frozen, which taught me and still teaches me to 'test the limits and break through' and to 'let go of fear'.Special credit goes to Zootopia for being a very relatable movie that reflects my struggles and triumphs as a foreign student in America.Someday, I hope to be just like Judy Hopps, and prove my naysayers wrong.
The object of this study is the case marking of the subject in early medieval charter Latin. The work explores whether and how the nominative/accusative-type morphosyntactic alignment changed into a semantically motivated (active/inactive) alignment in Late Latin before the disappearance of the case system. It is known that the accusative originally the case of the direct object extended in Late Latin to the subject function in which Classical Latin allowed only the nominative. On this basis, it has been postulated that in Late Latin the nominative/accusative contrast was (re)semanticized so that the nominative came to encode all the Agent-like arguments and the accusative all the Patient-like arguments. The study examines which semantic and syntactic factors determine the selection of the subject case in each subject/finite verb combination in the Late Latin Charter Treebank (LLCT). The LLCT is an annotated corpus of Latin charter texts (c. 200,000 words) written in Tuscany between AD 714 and 869. The central result of the study is that the Latin of the LLCT shows a semantically based morphosyntactic alignment in those parts of nominal declension where the morphological contrast between nominative- and accusative-based forms is morphophonologically intact. The following picture of intransitivity split turns up: the low-animacy subjects of the LLCT occur more often in the accusative than do the agentive high-animacy subjects. Likewise, the accusative percentage of SO subject constructions is higher than that of A/SA subject constructions. The common denominator of the examined semantic variables is likely to be the control exercised by the subject over the verbal process. Syntactic factors seem to influence the case distribution pattern as well. For example, the immediate preverbal position of the subject implies a high retention of the nominative. The immediate preverbal position of SV(O) language is a canonical subject position where the syntactic complexity measured as dependency lengths is at its lowest and the cohesion of the verbal nucleus at its highest. Thus, a by then already marked nominative form results. The control of the subject over the verbal process (semantic variable) and the cohesion of the verbal nucleus (syntactic variable) may be partly conflated, i.e., both may affect the subject case selection in certain conditions.
BACKGROUND: The intraoperative application of focused transthoracic echocardiography (TTE) is often considered to be restricted. Echocardiography with pocket-sized hand held ultrasound systems has been shown to be feasible in various settings. OBJECTIVE: The aim of this study was to investigate the feasibility of the intraoperative application of pocket-sized echocardiography and the comparison of its imaging quality and diagnostic reliability and variability with a standard ultrasound system. METHODS: After written informed consent, TTE was performed on 40 anaesthetised general, vascular, visceral, thoracic surgical and orthopaedic patients according to the FATE protocol: first, with a pocket-sized and second, with a high-end ultrasound system randomly by two anaesthetists. Imaging quality of four basic and three additional FATE views was rated on an established scale from 1 (impossible) to 5 (perfect). Successful TTE was defined, if one basic FATE views would be rated as grade 4 or 5 or alternatively two views as grade 3. Pathologic findings by both ultrasound devices were documented and imaging quality and pathologic findings were compared. RESULTS: All 40 patients presented acceptable imaging quality, resulting in a success rate of 1.0 (97.5%-CI 0.91-1, p= 0.015). The individual imaging ratings of each view were significantly lower with the pocket-sized system, but still showed acceptable imaging quality. With the high-end device more pathologic findings were detected (107 vs. 87), but none of the relevant or severe pathologies were overseen with the hand-held device. CONCLUSION: The application of a pocket-sized echocardiography device for focused intraoperative TTE is feasible and can appropriate be used for the initial evaluation of relevant pathologies in the operating theatre.
Mathematics and poetry typically operate in different realms, employing language and symbol in apparently disjoint semantic domains. In this article we explore the creative output of the Romanian poet/mathematician Ion Barbu/Dan Barbilian. Although he actively pursued these disciplines at different times, he wrote extensively on the parallels he perceived between mathematics and poetry. We explore both his published poetry and his mathematical research to suggest a role for translation in several ways. On one hand, the process of translation of his poetry from Romanian to English presents interesting challenges in bringing out the rich mathematical allusions therein. Alternatively, his mathematical writing, primarily in German, conforms to the linguistic norms that are now common to mathematics in most cultures, independent of language. Ultimately, we point to ways in which his work aspires to an ideal of inter-semiotic translation between poetry and mathematics.
We propose a transition-based dependency parser using Recurrent Neural Networks with Long Short-Term Memory (LSTM) units. This extends the feedforward neural network parser of Chen and Manning (2014) and enables modelling of entire sequences of shift/reduce transition decisions. On the Google Web Treebank, our LSTM parser is competitive with the best feedforward parser on overall accuracy and notably achieves more than 3% improvement for long-range dependencies, which has proved difficult for previous transition-based parsers due to error propagation and limited context information. Our findings additionally suggest that dropout regularisation on the embedding layer is crucial to improve the LSTM's generalisation.
Abstract A case study based on experience in linguistic investigations using annotated monolingual and multilingual text corpora; the “cases” include a description of language phenomena belonging to different layers of the language system: morphology, surface and underlying syntax, and discourse. The analysis is based on a complex annotation of syntax, semantic functions, information structure and discourse relations of the Prague Dependency Treebank, a collection of annotated Czech texts. We want to demonstrate that annotation of corpus is not a self-contained goal: in order to be consistent, it should be based on some linguistic theory, and, at the same time, it should serve as a test bed for the given linguistic theory in particular and for linguistic research in general.
Brown clustering has been used to help increase parsing performance for morphologically rich languages.However, much of the work has focused on using clustering techniques to replace terminal nodes or as a feature for parsing.Instead, we choose to examine how effectively Brown clustering is for unlexicalized parsing by creating data-driven POS tagsets which are then used with the Berkeley parser.We investigate cluster sizes as well as on what information (e.g.words vs. lemmas) clustering will yield the best parser performance.Our results approach the current state of the art results for the German TBa-D/Z treebank when using parser internal tagging.
This paper describes the submitted English shallow discourse parsing system from the natural language processing (NLP) group of Soochow university (SoNLP-DP) to the CoNLL-2016 shared task. Our System classifies discourse relations into explicit and non-explicit relations and uses a pipeline platform to conduct every subtask to form an end-to-end shallow discourse parser in the Penn Discourse Treebank (PDTB). Our system is evaluated on the CoNLL-2016 Shared Task closed track and achieves the 24.31% and 28.78% in F1-measure on the official blind test set and test set, respectively.
This paper investigates the influence of discourse features on text complexity assessment. To do so, we created two data sets based on the Penn Discourse Treebank and the Simple English Wikipedia corpora and compared the influence of coherence, cohesion, surface, lexical and syntactic features to assess text complexity.
Rhetorical relations are typically expressed by discourse structuring devices that ensure textual cohesion and coherence. Resources such as the PDTB target specifically the annotation of these devices, while describing alternative lexicalizations of such relations (AltLex). Our preparatory work to develop a discourse treebank for Portuguese in the PDTB framework has provided ground for some considerations regarding the status, in intra-sentential coherence, of main verbs that internally carry a causative meaning. We have first focused on the annotation of the rhetorical senses Reason, Result, Pragmatic_justification as expressed explicitly by discourse structuring devices (conjunctions, adverbs, phrases and prepositions), taken as elements that express a two-place semantic relation filled by propositional arguments. However, these relations are also frequently marked by other devices (AltLex).
Discourse relations can either be implicit or explicitly expressed by markers, such as 'therefore' and 'but'. How a speaker makes this choice is a question that is not well understood. We propose a psycholinguistic model that predicts whether a speaker will produce an explicit marker given the discourse relation s/he wishes to express. Based on the framework of the Rational Speech Acts model, we quantify the utility of producing a marker based on the information-theoretic measure of surprisal, the cost of production, and a bias to maintain uniform information density throughout the utterance. Experiments based on the Penn Discourse Treebank show that our approach outperforms stateof-the-art approaches, while giving an explanatory account of the speaker's choice.
While gender identities in the Western world are typically regarded as binary, our previous work (Hicks et al., 2015) shows that there is more lexical variety of gender identity and the way people identify their gender.There is also a growing need to lexically represent this variety of gender identities.In our previous work, we developed a set of tools and approaches for analyzing Twitter data as a basis for generating hypotheses on language used to identify gender and discuss genderrelated issues across geographic regions and population groups in the U.S.A.In this paper we analyze the coverage and relative frequency of the word forms in our Twitter analysis with respect to the National Transgender Discrimination Survey data set, one of the most comprehensive data sets on transgender, gender non-conforming, and gender variant people in the U.S.A.We then analyze the coverage of WordNet, a widely used lexical database, with respect to these identities and discuss some key considerations and next steps for adding gender identity words and their meanings to WordNet.