Papers reviewed and determined not to be word norm studies. Use the flag icon to report errors or suggest re-inclusion.
18625 papers
This work introduces a new approach to checking treebank consistency. Derivation trees based on a variant of Tree Adjoining Grammar are used to compare the annotation of word sequences based on their structural similarity. This overcomes the problems of earlier approaches based on using strings of words rather than tree structure to identify the appropriate contexts for comparison. We report on the result of applying this approach to the Penn Arabic Treebank and how this approach leads to high precision of error detection. 1
Noun phrases (nps) are a crucial part of natural language, and can have a very complex structure. However, this np structure is largely ignored by the statistical parsing field, as the most widely used corpus is not annotated with it. This lack of gold-standard data has restricted previous efforts to parse nps, making it impossible to perform the supervised experiments that have achieved high performance in so many Natural Language Processing (nlp) tasks. We comprehensively solve this problem by manually annotating np structure for the entire Wall Street Journal section of the Penn Treebank. The inter-annotator agreement scores that we attain dispel the belief that the task is too difficult, and demonstrate that consistent np annotation is possible. Our gold-standard np data is now available for use in all parsers. We experiment with this new data, applying the Collins (2003) parsing model, and find that its recovery of np structure is significantly worse than its overall performance. The parser's F-score is up to 5.69% lower than a baseline that uses deterministic rules. Through much experimentation, we determine that this result is primarily caused by a lack of lexical information. To solve this problem we construct a wide-coverage, large-scale np Bracketing system. With our Penn Treebank data set, which is orders of magnitude larger than those used previously, we build a supervised model that achieves excellent results. Our model performs at 93.8% F-score on the simple task that most previous work has undertaken, and extends to bracket longer, more complex nps that are rarely dealt with in the literature. We attain 89.14% F-score on this much more difficult task. Finally, we implement a post-processing module that brackets nps identified by the Bikel (2004) parser. Our np Bracketing model includes a wide variety of features that provide the lexical information that was missing during the parser experiments, and as a result, we outperform the parser's F-score by 9.04%. These experiments demonstrate the utility of the corpus, and show that many nlp applications can now make use of np structure.
Abstract During prolonged exercise, ratings of perceived exertion (RPE) and affect (pleasure; activation) play an important role in performance, especially in hot conditions. Although various mechanisms have been tried to manage the effects of heat (e.g. cooling jackets), the purpose of the current research was to assess the effect of a cooling collar on RPE, affect, thermal sensation, and running performance in the heat. Participants ( n = 8) wore a cold collar, uncooled collar or no collar as they completed three 90‐min preloaded treadmill time‐trials in the heat (30.5±0.1°C), which included 75 min at ∼60% [Vdot] O 2max followed by a 15‐min time‐trial. Affect, RPE, and thermal sensation were measured throughout. Performance during the time‐trial was significantly improved in the cold collar compared with the uncooled collar and no‐collar trials; however, pleasure and activation ratings were highest for the no‐collar trial. Findings suggest that a cooling collar may improve performance, but not necessarily “feel good”. Practically, it seems necessary to ensure sufficient education about the benefits, on balance, of the cold collar, while the performance improvements outweighed the reported discomfort of athletes in this study. In addition, the cold collar may prove quite beneficial for individuals who work in hot climates and or conditions, as it may be effective in minimizing discomfort associated with heat for other groups as well.
Reviewed by: From poets to padonki: Linguistic authority and norm negotiation in modern Russian culture Anastassia Zabrodskaja Ingunn Lunde and Martin Paulsen, eds. From poets to padonki: Linguistic authority and norm negotiation in modern Russian culture. Bergen: University of Bergen, 2009. [Slavica Bergensia, 9.] The issues discussed in From poets to padonki became a part of my life in 1999 when I began my university studies and using Russian and Estonian became my everyday reality. Choosing different languages for different purposes, code-switching, and multilingual language play have all been part of my daily language use for the last eleven years. This collection challenges definitions of what can be meant by "language", "standard" language, and the linguistic "norm". It acquaints readers with a wide range of linguistic phenomena in modern Russian culture. Padonki refers to a subculture within the Russian-speaking Internet (Runet), whose representatives use erratic spellings for words, aiming at creating a comic effect. Their nickname padonki itself illustrates such a trend-it is an alteration of podonki 'dregs'. Padonki is also characterized by gratuitous use of profanity and a penchant for obscene subjects. During the last decade, a body of literature has emerged proposing that (socio)linguists direct their attention away from the traditional focus of linguistics, i.e., language as a bounded system, towards broader semiotic resources, to see what is really going on when people use "language" (Stroud 2003, Jacquemet 2005, Shohamy 2006, Makoni and Pennycook 2007, Blommaert 2010). The notion of "language" becomes especially questionable in cases of (multilingual) computer-mediated communication. The last decade has also witnessed rising scholarly interest in language on and of the Internet in general and in e-mails and postings on Internet discussion forums or message boards in particular (e.g., Koutsogiannis and Mitsikopoulou 2003, Palfreyman and al Khalil 2003, Hinrichs 2006, Dorleijn and Nortier 2009, Androutsopoulos 2006, 2009). García (2009: 32) instead of "language" offers a more suitable term for the multiple discursive practices-languaging, i.e., "social practices that are actions performed by our meaning-making selves." For her, dialects, pidgins, creoles, and academic language [End Page 153] are all examples of languaging, as there are differences between language practices at home, in communities, and in academic contexts. I would argue that standard languages are idealized constructs, and none can remain unaffected by language contacts during its entire history. While studying language use by individuals, it is important to shift "from focus on structure to focus on function-from focus on linguistic form in isolation to linguistic form in human context" (Hymes 1974: 77). The volume under review offers fascinating reading for (socio)linguists, who work with different manifestations of real language practices rather than seeking for linguistic norm descriptions. Comprising 17 contributions written in English and Russian, the book summarizes analyses of the norm in modern Russian language culture based on data from multiple sources: "literary fiction, Internet slang, literary criticism and aesthetics, writers' blogs, linguistic play, and various arenas for 'talk about talk,' such as the classroom, blogs, the media, the courtroom, etc." (11). The book opens with Ingunn Lunde and Martin Paulsen's excellent general introduction, an insightful synthesis of how various approaches to the standard Russian language, its norms, and linguistic standards see the history, development, and future of the Russian language culture. The first paper, "Living norms" by Henning Andersen, gives a comprehensive overview of the history and development of the notion of language norms. Andersen reviews historical contributions to the understanding of norms. Making a distinction between declarative and deontic norms and dividing them further into explicit and implicit norms, he highlights the leading ideas in the field of (socio)linguistic studies that concern the position of norms. On the example of excerpts from Soviet grammars and dictionaries, he illustrates how norms can be governed from above by agencies of the state. "The standard norms include only prescribed and permitted forms", and all other forms must be avoided (25). Andersen also comments on spoken language standards, noting that "the [Russian] language continues to be spoken effectively in its numerous variants all over the inherited Russian language area as well as in the diasporas, old and new" (32). Martin Paulsen...
Annotated corpora are essential for almost all NLP applications. Whereas they are expected to be of a very high quality because of their importance for the followup developments, they still contain a considerable number of errors. With this work we want to draw attention to this fact. Additionally, we try to estimate the amount of errors and propose a method for their automatic correction. Whereas our approach is able to find only a portion of the errors that we suppose are contained in almost any annotated corpus due to the nature of the process of its creation, it has a very high precision, and thus is in any case beneficial for the quality of the corpus it is applied to. At last, we compare it to a different method for error detection in treebanks and find out that the errors that we are able to detect are mostly different and that our approaches are complementary. 1
The paper describes the basic strategies behind the word and semantic level alignment in the Bulgarian-English treebank. The word level alignment has taken into consideration the experience within other NLP groups in the context of the Bulgarian language specific features. The semantic level alignment builds on the word level alignment and is represented in the framework of the Minimal Recursion Semantics. 1
Journal Article Building A Lexical Database with Multiple Outputs: Examples from Legacy Data and from Multimodal Fieldwork Get access Nick Thieberger Nick Thieberger University of Melbourne (thien@unimelb.edu.au) Search for other works by this author on: Oxford Academic Google Scholar International Journal of Lexicography, Volume 24, Issue 4, December 2011, Pages 463–472, https://doi.org/10.1093/ijl/ecr027 Published: 10 September 2011
The article deals with processing of diminutives in the Pralex lexical database, in particular with their allomorphism, homonymy and word-forming relations. Moreover, it discusses various definitions of diminutive meanings and provides their detailed classification in particular types.
This paper introduces the META-NORD project which develops Nordic and Baltic part of the European open language resource infrastructure. META-NORD works on assembling, linking across languages, and making widely available the basic language resources used by developers, professionals and researchers to build specific products and applications. The goals of the project, overall approach and specific action lines on wordnets, terminology resources and treebanks are described. Moreover, results achieved in first five months of the project, i.e. language whitepapers, metadata specification and IPR management, are presented. 1
This paper briefly depicts major achievements in the Chinese language information processing and roughly reviews the computational linguistic research for the recent 20 years in China.The author questions the current methodologies such as POS tagging and treebank for Chinese.The paper presents some new ideas about the construction of Chinese data resources.The authors suggest that for Chinese we should address deep and semantic annotation instead of current shallow and syntactic one.The future annotation will include targeted tackling,diversified content,stepwise procedure,and non-professional annotators.The paper predicts some eye-catching features: merge of technologies and human-centered computing.
How to quickly locate the interested information in the XML database under a certain twig pattern is a popular research topic.To solve the problem that the TwigStack algorithm for handling the case with parent-child nodes would come out with massive intermediate results,an improved twig pattern query algorithm of cTwigStack was proposed,which was based on caching the non-leaf nodes and delaying the leaf nodes output.The experimental results on Treebank dataset indicate that the proposed algorithm can achieve the most accurate results of the queries that contain the ancestor-descendant relationships below branching nodes.Besides,compared with the present algorithm,it is also highly effective when processing parent-child relationships below branching nodes.
International audience
For the task of automatic treebank conversion, this paper presents a feature-based approach which encodes bracketing structures in a treebank into features to guide the conversion of this treebank to a different standard. Experiments on two Chinese treebanks show that our approach improves conversion accuracy by 1.31 % over a strong baseline. 1
Text clustering is of substantial importance to information retrieval.The method of applying the information of syntactic distribution to text clustering is presented,in order to avoid the complex clustering algorithm whileenabling the linguistic interpretation of clustering features and the results of clustering.According to the dependency Treebank,ten dependency relations are suggested with distinctive distribution between oral and written Chinese By using five of them as clustering feature,the similarity of spoken and written classes achieves 71.98% and 83.13%,respectively.The experiment result shows that the proposed method of applying dependency relations to text clustering is feasible and effective.
Transition-based dependency parsers generally use heuristic decoding algorithms but can accommodate arbitrarily rich feature representations. In this paper, we show that we can improve the accuracy of such parsers by considering even richer feature sets than those employed in previous systems. In the standard Penn Treebank setup, our novel features improve attachment score form 91.4 % to 92.9%, giving the best results so far for transitionbased parsing and rivaling the best results overall. For the Chinese Treebank, they give a signficant improvement of the state of the art. An open source release of our parser is freely available.
This paper proposes a method to improve the accuracy of bilingual texts (bitexts) dependency parsing by using an auto-generated bilingual treebank created with the help of statistical machine translation (SMT) systems. Previous bitext parsing methods use human-annotated bilingual treebanks that are costly and troublesome to obtain. In the proposed method, we use an auto-generated bilingual treebank to train the parsing models. First, an SMT system is used to translate a monolingual treebank into the target language; then, a monolingual parser for the target language is used to parse the translated sentences. Since the auto-translated sentences and auto-parsed trees in the auto-generated bilingual treebank are far from perfect, the bilingual constraints are not sufficiently reliable. To overcome this problem, we propose a method to verify the reliability of the constraints using a large amount of target monolingual and bilingual unannotated data. Finally, we design a set of effective bilingual features for parsing models on the basis of the verified constraints. We conduct the experiments using a standard test data. The experimental results show that our bitext parser significantly outperforms monolingual parsers. Moreover, our method is still able to provide improvement when we use a larger monolingual treebank containing over 50 000 sentences. We also test the proposed method with different SMT systems and the results show that our method is very robust to the noise. In particular, the proposed method can be used in a purely monolingual setting with the help of SMT. That is, it does not need the human translation of the test set as previous methods do.
There has been a rapid increase in the volume of research on data-driven dependency parsers in the past five years. This increase has been driven by the availability of treebanks in a wide variety of languages—due in large part to the CoNLL shared tasks—as well as the straightforward mechanisms by which dependency theories of syntax can encode complex phenomena in free word order languages. In this article, our aim is to take a step back and analyze the progress that has been made through an analysis of the two predominant paradigms for data-driven dependency parsing, which are often called graph-based and transition-based dependency parsing. Our analysis covers both theoretical and empirical aspects and sheds light on the kinds of errors each type of parser makes and how they relate to theoretical expectations. Using these observations, we present an integrated system based on a stacking learning framework and show that such a system can learn to overcome the shortcomings of each non-integrated system.
Previous research suggests that neural and behavioral responses to surprised faces are modulated by explicit contexts (e.g., "He just found $500"). Here, we examined the effect of implicit contexts (i.e., valence of other frequently presented faces) on both valence ratings and ability to detect surprised faces (i.e., the infrequent target). In Experiment 1, we demonstrate that participants interpret surprised faces more positively when they are presented within a context of happy faces, as compared to a context of angry faces. In Experiments 2 and 3, we used the oddball paradigm to evaluate the effects of clearly valenced facial expressions (i.e., happy and angry) on default valence interpretations of surprised faces. We offer evidence that the default interpretation of surprise is negative, as participants were faster to detect surprised faces when presented within a happy context (Exp. 2). Finally, we kept the valence of the contexts constant (i.e., surprised faces) and showed that participants were faster to detect happy than angry faces (Exp. 3). Together, these experiments demonstrate the utility of the oddball paradigm to explore the default valence interpretation of presented facial expressions, particularly the ambiguously valenced facial expression of surprise.
Part-of-speech (POS) is an indispensable feature in dependency parsing. Current research usually models POS tagging and dependency parsing independently. This may suffer from error propagation problem. Our experiments show that parsing accuracy drops by about 6 % when using automatic POS tags instead of gold ones. To solve this issue, this paper proposes a solution by jointly optimizing POS tagging and dependency parsing in a unique model. We design several joint models and their corresponding decoding algorithms to incorporate different feature sets. We further present an effective pruning strategy to reduce the search space of candidate POS tags, leading to significant improvement of parsing speed. Experimental results on Chinese Penn Treebank 5 show that our joint models significantly improve the state-of-the-art parsing accuracy by about 1.5%. Detailed analysis shows that the joint method is able to choose such POS tags that are more helpful and discriminative from parsing viewpoint. This is the fundamental reason of parsing accuracy improvement. 1
This study investigated cognitive and emotional effects of syncopation, a feature of musical rhythm that produces expectancy violations in the listener by emphasising weak temporal locations and de-emphasising strong locations in metric structure. Stimuli consisting of pairs of unsyncopated and syncopated musical phrases were rated by 35 musicians for perceived complexity, enjoyment, happiness, arousal, and tension. Overall, syncopated patterns were more enjoyed, and rated as happier, than unsyncopated patterns, while differences in perceived tension were unreliable. Complexity and arousal ratings were asymmetric by serial order, increasing when patterns moved from unsyncopated to syncopated, but not significantly changing when order was reversed. These results suggest that syncopation influences emotional valence (positively), and that while syncopated rhythms are objectively more complex than unsyncopated rhythms, this difference is more salient when complexity increases than when it decreases. It is proposed that composers and improvisers may exploit this asymmetry in perceived complexity by favoring formal structures that progress from rhythmically simple to complex, as can be observed in the initial sections of musical forms such as theme and variations.
OBJECTIVES: This research compared sensory processing and personality traits involved in deciding to try a novel fruit (guava) in adults and children. DESIGN: The research employed an age, sex, and food neophobia matched between-participant design to examine sensory decision making in choosing to eat a novel fruit. METHODS: Forty-four adults (Study 1) and 68 children (Study 2) took part. In each study, participants were separated into two groups to investigate whether prior assessment of a familiar and liked fruit (apple) that shares similar visual characteristics to the target novel fruit (guava) increased the likelihood that an individual would decide to try it. All participants completed appetitive and familiarity ratings by sensory stages: vision, smell, and touch, prior to trying (tasting) the fruit. Participants (or their parents) also completed the general and food neophobia scales and adults also completed the sensation-seeking scale. RESULTS: Twenty-eight adults (64%) tried the guava and 16 did not (36%). In the second study, 22 children decided not to try the novel fruit (32%). Significant predictors of whether the adult tried the target fruit were Thrill and Adventure Seeking, Experience Seeking, General Neophobia, and 'appealing to touch'. In children, Food Neophobia, concurrent presentation of a familiar fruit alongside the target and visual assessment of the target predicted decision to try the novel fruit. CONCLUSIONS: This study suggests that touch is pertinent to adults' decision to try a novel fruit, whereas visual cues appear to be more important for children.
Emotion recognition algorithms for spoken dialogue applications typically employ lexical models that are trained on labeled in-domain data. In this paper, we propose a domainindependent approach to affective text modeling that is based on the creation of an affective lexicon. Starting from a small set of manually annotated seed words, continuous valence ratings for new words are estimated using semantic similarity scores and a kernel model. The parameters of the model are trained using least mean squares estimation. Word level scores are combined to produce sentence-level scores via simple linear and non-linear fusion. The proposed method is evaluated on the SemEval news headline polarity task and on the ChIMP politeness and frustration detection dialogue task, achieving state-of-theart results on both. For politeness detection, best results are obtained when the affective model is adapted using in domain data. For frustration detection, the domain-independent model and non-linear fusion achieve the best performance. Index Terms: language understanding, emotion, affect, affective lexicon
BACKGROUND: The neuroanatomic basis of affective processing deficits in Huntington disease is insufficiently understood. We investigated whether Huntington disease-related deficits in emotion recognition and experience are associated with specific changes in grey matter volume. METHOD: We assessed grey matter volume in symptomatic patients with Huntington disease and healthy controls using voxel-based morphometry, and we correlated regional grey matter volume with participants' affective ratings. RESULTS: We enrolled 18 patients with Huntington disease and 18 healthy controls in our study. Patients with Huntington disease showed normal affective experience but impaired recognition of negative emotions (disgust, anger, sadness). The patients perceived the emotions as less intense and made more classification errors than controls. These deficits were correlated with regional atrophy in emotion-relevant areas (insula, orbitofrontal cortex) and in memory-relevant areas (dorsolateral prefrontal cortex, hippocampus). LIMITATIONS: Our study was limited by the small sample size and the resulting modest statistical power relative to the number of tests. CONCLUSION: Our study sheds new light on the importance of a cognitive-affective brain circuit involved in the affect recognition impairment in patients with Huntington disease.
This paper introduces Chart Inference (CI), an algorithm for deriving a CCG category for an unknown word from a partial parse chart. It is shown to be faster and more precise than a baseline brute-force method, and to achieve wider coverage than a rule-based system. In addition, we show the application of CI to a domain adaptation task for question words, which are largely missing in the Penn Treebank. When used in combination with self-training, CI increases the precision of the baseline StatCCG parser over subjectextraction questions by 50%. An error analysis shows that CI contributes to the increase by expanding the number of category types available to the parser, while self-training adjusts the counts. 1
In this paper, we describe and compare two statistical parsing approaches for the hybrid dependency-constituency syntactic representation used in the Quranic Arabic Treebank (Dukes and Buckwalter, 2010). In our first approach, we apply a multi-step process in which we use a shift-reduce algorithm trained on a pure dependency preprocessed version of the treebank. After parsing, the dependency output is converted into the hybrid representation. This is compared to a novel one-step parser that is able to learn the hybrid representation without preprocessing. We define an extended labelled attachment score (ELAS) as our performance metric for hybrid parsing, and report 87.47 % (F1 score) for the multi-step approach, and 89.03 % (F1 score) for the onestep integrated algorithm. We also consider the effect of using different sets of morphological features for parsing the Quran, comparing our results to recent work on Modern Standard Arabic.
We present an approach of identifying the most prominent text/sentences using various shallow linguistic features, taking degree of connectiveness among the text units into consideration so as to minimize the poorly linked sentences in the resulting summary. As per the limitations of the current summarizing systems, the summary generated by those systems contains poorly linked sentences and are not topically salient. Thus, the paper aims at highlighting the effect of lexical chain scoring after the nouns and compound nouns are chained by searching for lexical cohesive relationships between words in the text using WordNet and using lexicographical relationships such as synonymy and hyponyms. In this paper, our algorithm ranks sentences based on the sum of the scores of the words in each sentence involving approaches like term frequencies, location of sentence in the text, cue words and phrases, word occurrences, and measuring lexical similarity(measuring chain score, word score and finally sentence score) for ranking the text units. We then identified and extracted high scored sentences and then the Vector Space approach is used to measure the relatedness/similarity between the extracted sentence and the topic words involving again the WordNet lexical database relationships to prioritise the topically related sentences. A threshold angle between the two vectors is predefined experimentally to which the ranked/scored sentences to be dropped and which the significant sentences with ranking/scores higher than threshold to be extracted. Note that the value of threshold is predetermined based on the percentage of output summary required to be generated.
Prepositional phrase (PP) consists of two parts which are a preposition as the leading part and a word or phrase as the tail part. In accordance with this fact, this paper proposes a new approach for identifying PP. In this method, PP identification is transformed into the collocation identification of preposition itself and the right boundary word. The Cascaded Conditional Random Fields (CCRFs) is used in this approach. With the Penn Chinese Treebank 5.1 as our experiment corpus, the F <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sub> rises to 94.63%. This approach obtains breakthrough in this specific field as the current F <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sub> is about 8.6% higher than any publicly published paper.
We want to demonstrate on some selected linguistic issues that classical structural and functional linguistics even with its seemingly traditional approaches has something to offer to a formal description of language and its applications in natural language processing and to illustrate by a brief reference to Functional Generative Grammar (on the theoretical side of CL) and Prague Dependency Treebank (on the applicational side) a possible interaction between linguistics and CL.
We employ syntactic parsing to describe and to discover lexico-grammatical features of English regional varieties. In the absence of suitable Treebanks, automatically parsed corpora (tree jungles) can be used. As an example we focus on Indian English, using the International Corpus of English (ICE), and the British National Corpus (BNC). We use a largely corpus-driven method. There are few differences in frequencies of syntactic relations between the corpora, but considerable differences when taking the intricate relations between grammar and lexis into account. We describe differences in the use of zero articles, verb-preposition constructions, and ditransitive verbs. We show that relatively small corpora can be used to discover subtle lexico-grammatical differences.
While some visual objects prompt strong affective responses (e.g., guns and ice cream), most objects are thought to be affectively neutral. Last year we reported evidence for the existence of “micro-valences” (Lebrecht & Tarr, VSS, 2010): that nominally neutral objects actually possess subtle valences that we hypothesize form an integral part of object perception. In the current experiment we used fMRI to investigate: a) the extent to which micro-valences are coded within the extended visual object recognition network (Bar, 2007); b) how micro-valences are neurally instantiated with respect to valence strength and direction. Using slow event-related fMRI, participants viewed an object picture for 500ms and evaluated the object's “pleasantness” on each trial. Participants were shown 120 everyday, nominally neutral objects (e.g., teapots and clocks) and 120 strongly valenced objects (e.g., gold and a skull). Objects were assigned to these conditions based on mean valence ratings acquired in a prior experiment with a different population of participants. Individualized ratings for all objects were also acquired for our fMRI participants during a post-scan session. Regions of interest for further analysis were identified using two independent localizers: a) objects versus scrambled objects; b) strongly valenced objects versus minimally valenced objects (e.g., paperclips). Two results stand out. First, somewhat consistent with previous findings, lateral regions of PFC and regions of medial OFC are selective to a positive versus negative comparison for strongly valenced objects. Second, and intriguingly, almost all participants show selectivity for micro-valence objects, comparing positive to negative, in a region adjacent to the region for strongly valenced objects. We posit that intrinsic to visual object perception, object valence – for all objects – is evaluated in PFC. This valence metric forms one of many associated object properties that can influence subsequent perceptual and non-perceptual object-related processing.
This paper proposes a method for shallow parsing on the basis of CRF and transformation-based error-driven learning.The method is applied to Penn Chinese Treebank and gets a good performance of chunking identification.First,CRF model is used to identify chunks to acquire candidate transformation rules by error-driven learning.Then,an evaluation function is used to filter candidate transformation rules.And last,transformation rules are used to revise the chunking results of CRF.The experimental results show that this approach is effective,and outperforms the single CRF-based approach in shallow parsing.Precision,recall and F-values are improved respectively.
In this paper, we expand Morzycki (2009)’s claims that degree readings of size adjectives are attributed to syntax. We introduce a corpus-based analysis in Dutch to verify and extend his claim into the semantic domain. Using the LASSY Treebank, we extract syntactic and semantic properties of noun phrases consisting of the adjectives “gigantisch”, “kolossaal”, and “reusachtig ” and manually annotate each adjective-noun pair with a gradable or nongradable label. Using these features, we construct a statistical model based on logistic regression and find that the grammatical role, definiteness, and particular semantic noun groups derived from Cornetto (a Dutch WordNet with referential relations) have a significant effect on the likelihood that an adjective-noun pair is interpreted by the reader to have a degree reading. 1.
This study has a dual focus in that it aims to develop a viable methodology for elicitation experiments in English linguistics, while simultaneously applying the proposed methods to investigate an actual subject, the distribtion of the additive particles 'also' and 'too'. Traditionally, data for linguistic research is gained by sampling natural language corpora. Although this approach is valid and, indeed, has been applied here, elicitation experiments can gain in validity and informative value by additionally introducing questionnaires to accompany corpus research. Online questionnaires particularly are a cost-effective and highly customizable tool to create a linguistic database against which existing data can be tested. For the purpose of this study, I have created six online questionnaires to test three hypotheses about the distribution of 'also' and 'too'. Two interdependent hypotheses assume that the use of the two particles is sensitive to structural properties of the `added constituent' while the third one, the information-structural hypothesis, argues that the use of 'also' and 'too' is controlled by the information structure of the sentence. In addition to the questionnaires, a balanced sample was extracted from the "British National Corpus" and tested against corpus data from previous studies as well as the data elicited online. In the course of this study, the additive particles will firstly be defined in terms of their structural properties, and the hypotheses about their use introduced and explicated. Furthermore, the data elicitation process will be detailed, as well as results from previous studies be taken into account. The hypotheses will subsequently be tested against the data from both corpus research and elicitation per questionnaires, and the outcome discussed. Concluding the study, I will focus on the results of the distribution analysis as well as evaluate the introduction of the online questionnaires and their application in the context of testing the hypotheses against empirical linguistic data.
Treebanks are a necessary prerequisite for many NLP tasks, including, but not limited to, semantic role labeling. For many languages, however, treebanks are either nonexistent or too small to be useful. Time-critical applications may require rapid deployment of natural language software for a new critical language—much faster than the development time of a traditional treebank. This dissertation describes a method for generating a treebank and training syntactic and semantic models using only semantic training information—that is, no human-annotated syntactic training data whatsoever. This will greatly increase the speed of development of natural language tools for new critical languages in exchange for a modest drop in overall accuracy. Using Combinatory Categorial Grammar (CCG) in concert with Propbank semantic role annotations allows us to accurately predict lexical categories in combination with a partially hidden Markov model. By training the Berkeley parser on our generated syntactic data, we can achieve SRL performance of 65.5% without using a treebank, as opposed to 74% using the same feature set with gold-standard data.
The reconstruction of standardized texts in the Prague Dependency Treebank of Spoken Czech enables the comparison of authentic spoken utterances and standardized texts The authors concentrate on the questions: What does the syntactic identity of the Czech spoken and written texts consist of? What syntactic constructions are „natural“ in the spoken and in the written text? What is the difference in the density of the cohesive links, in the explicit and implicite relations between units?
Abstract In this paper, I investigate the phonological similarity of different elements of the phonological pole of multi-word units. I discuss two case studies on slightly different levels of abstractness. The first case study investigates lexically fully-specified V-NP DirObj idioms such as kick the bucket and lose one's cool; the idioms investigated are taken from the Collins Cobuild Dictionary of Idioms (Harper Collins, 2002). The second case study investigates the lexically less specified way -construction, which is exemplified by He fought his way through the crowd (cf. Goldberg, Constructions: A Construction Grammar approach to argument structure, The University of Chicago Press, 1995: Ch. 9), on the basis of data from the British National Corpus 1.0. I show that both patterns exhibit a strong phonological within-pole relation, namely a strong preference for having their slots filled with phonologically similar elements, where phonological similarity is manifested in alliteration patterns. These preferences are statistically significant when compared to chance-level expectations derived both from the corpora and from the CELEX database (Baayen et al., The CELEX Lexical Database (CD-ROM), University of Pennsylvania, 1995) and are explained on the basis of Langacker's concept of syntactic and phonological constituents as well as current exemplar-/usage-based approaches.
In this paper, we investigated the impact of extracting different types of multiword expressions (MWEs) in improving the accuracy of a data-driven dependency parser for a morphologically rich language (Turkish). We showed that in the training stage, the unification of MWEs of a certain type, namely compound verb and noun formations, has a negative effect on parsing accuracy by increasing the lexical sparsity. Our results gave a statistically significant improvement by using a variant of the treebank excluding this MWE type in the training stage. Our extrinsic evaluation of an ideal MWE recognizer (for only extracting MWEs of type named entities, duplications, numbers, dates and some predefined list of compound prepositions) showed that the preprocessing of the test data would improve the labeled parsing accuracy by 1.5%.
This chapter discusses the inspection of the relevance of language to power and social diversity. It explores the elastic impact of power on the social life of living languages. Sociolinguists and historical linguists have demonstrated that linguistic evolution has been shaped by many forces, including political circumstances that are not egalitarian. Einar Haugen promoted the study of language within its ecological context. Some critics of quantitative variationist sociolinguistics have noted that scholars who classify speakers based on pre-ordained social categories, like race, may miss important nuances in linguistic behavior that defy easy circumstantial classification. Hymes affirmed that communicative events demand a high degree of communicative competence as related to language usage throughout the world. William Labov's study of the social stratification of English speakers in New York City is illustrative of urban linguistic stratification. The majority of speech communities throughout the world set the indigenous standard linguistic norms.
We show how punctuation can be used to improve unsupervised dependency parsing. Our linguistic analysis confirms the strong connection between English punctuation and phrase boundaries in the Penn Treebank. However, approaches that naively include punctuation marks in the grammar (as if they were words) do not perform well with Klein and Manning’s Dependency Model with Valence (DMV). Instead, we split a sentence at punctuation and impose parsing restrictions over its fragments. Our grammar inducer is trained on the Wall Street Journal (WSJ) and achieves 59.5 % accuracy out-of-domain (Brown sentences with 100 or fewer words), more than 6 % higher than the previous best results. Further evaluation, using the 2006/7 CoNLL sets, reveals that punctuation aids grammar induction in 17 of 18 languages, for an overall average net gain of 1.3%. Some of this improvement is from training, but more than half is from parsing with induced constraints, in inference. Punctuation-aware decoding works with existing (even already-trained) parsing models and always increased accuracy in our experiments. 1
Resolving coordination ambiguity is a classic hard problem. This paper looks at coordination disambiguation in complex noun phrases (NPs). Parsers trained on the Penn Treebank are reporting impressive numbers these days, but they don’t do very well on this problem (79%). We explore systems trained using three types of corpora: (1) annotated (e.g. the Penn Treebank), (2) bitexts (e.g. Europarl), and (3) unannotated monolingual (e.g. Google N-grams). Size matters: (1) is a million words, (2) is potentially billions of words and (3) is potentially trillions of words. The unannotated monolingual data is helpful when the ambiguity can be resolved through associations among the lexical items. The bilingual data is helpful when the ambiguity can be resolved by the order of words in the translation. We train separate classifiers with monolingual and bilingual features and iteratively improve them via co-training. The co-trained classifier achieves close to 96 % accuracy on Treebank data and makes 20 % fewer errors than a supervised system trained with Treebank annotations. 1