Papers reviewed and determined not to be word norm studies. Use the flag icon to report errors or suggest re-inclusion.
18265 papers
In this paper, we propose efficient and less resource-intensive strategies for parsing of code-mixed data. These strategies are not constrained by in-domain annotations, rather they leverage pre-existing monolingual annotated resources for training. We show that these methods can produce significantly better results as compared to an informed baseline. Besides, we also present a data set of 450 Hindi and English code-mixed tweets of Hindi multilingual speakers for evaluation. The data set is manually annotated with Universal Dependencies.
<strong>Berkeley parser model for Korean: Sejong treebank</strong> Jungyeul Park, Jeen-Pyo Hong, and Jeong-Won Cha (2016). Korean Language Resources for Everyone. In Proceedings of the 30th Pacific Asia Conference on Language, Information and Computation (PACLIC 30). Seoul, Korea. [pdf] @inproceedings{park-hong-cha:2016:PACLIC, <br> address = {Seoul, Korea}, <br> author = {Park, Jungyeul and Hong, Jeen-Pyo and Cha, Jeong-Won}, <br> booktitle = {Proceedings of the 30th Pacific Asia Conference on Language, Information and Computation (PACLIC 30)}, <br> pages = {49--58}, <br> title = {{Korean Language Resources for Everyone}}, <br> year = {2016} <br> } It requires Espresso's POS tagging results for input. Espresso is available at https://zenodo.org/record/884606
In this contribution, I review Augustinus's (2015) dissertation on the syntax of verb clusters. While the main theoretical aspects of this book (within the HPSG framework) may not be of interest to everyone, the corpus methodology used and the GrETEL tool that was developed in the course of this project are impressive, The book contains previously unknown descriptive details about possible clustering verbs. Due to its interdisciplinary nature, there should be something of interest here for anyone who is interested in verb clusters.
The use of the senses of vision and audition as interactive means has dominated the field of Human-Computer Interaction (HCI) for decades, even though nature has provided us with many more senses for perceiving and interacting with the world around us. That said, it has become attractive for HCI researchers and designers to harness touch, taste, and smell in interactive tasks and experience design. In this paper, we present research and design insights gained throughout an interdisciplinary collaboration on a six-week multisensory display – Tate Sensorium – exhibited at the Tate Britain art gallery in London, UK. This is a unique and first time case study on how to design art experiences whilst considering all the senses (i.e., vision, sound, touch, smell, and taste), in particular touch, which we exploited by capitalizing on a novel haptic technology, namely, mid-air haptics. We first describe the overall set up of Tate Sensorium and then move on to describing in detail the design process of the mid-air haptic feedback and its integration with sound for the Full Stop painting by John Latham (1961). This was the first time that mid-air haptic technology was used in a museum context over a prolonged period of time and integrated with sound to enhance the experience of visual art. As part of an interdisciplinary team of curators, sensory designers, sound artists, we selected a total of three variations of the mid-air haptic experience (i.e., haptic patterns), which were alternated at dedicated times throughout the six-week exhibition. We collected questionnaire-based feedback from 2500 visitors and conducted 50 interviews to gain quantitative and qualitative insights on visitors’ experiences and emotional reactions. Whilst the questionnaire results are generally very positive with only a small variation of the visitors’ arousal ratings across the three tactile experiences designed for the Full Stop painting, the interview data shed light on the differences in the visitors’ subjective experiences. Our findings suggest multisensory designers and art curators can ensure a balance between surprising experiences versus the possibility of free exploration for visitors. In addition, participants expressed that experiencing art with the combination of mid-air haptic and sound was immersive and provided an up-lifting experience of touching without touch. We are convinced that the insights gained from this large-scale and real-world field exploration of multisensory experience design exploiting a new and emerging technology provide a solid starting point for the HCI community, creative industries, and art curators to think beyond conventional art experiences. Specifically, our work demonstrates how novel mid-air technology can make art more emotionally engaging and stimulating, especially abstract art that is often open to interpretation.
We argue that coherence relations (relations between propositions, such as Concession or Purpose) are signalled more frequently and by more means than is generally believed. We examine how coherence relations in text are indicated by all possible textual signals, and whether every relation is signalled. To that end, we conducted a corpus study on the RST Discourse Treebank, a corpus of newspaper articles annotated for rhetorical (or coherence) relations. Results from our corpus study show that most relations in text (over 90%) are signalled and also that most signalled relations (over 80%) are indicated not only by discourse markers (and, but, if, since), but also by a wide variety of signals other than discourse markers, such as reference, lexical, semantic, syntactic and graphical features. These findings suggest that signalling of coherence relations is much more sophisticated than previously thought.
This case study of a team in an international workplace investigates processes of language socialization in a transient multilingual setting. Using interview and observational data, the analysis shows how social and linguistic norms are negotiated, with the newcomer positioned as a catalyst for changing language practices toward more English, with the ultimate aim of creating a “global mindset” in the organization. Language socialization in a transient multilingual setting is shown to focus on and assign positive value to new linguistic norms that experienced members are socialized into in a process that hinges on new members functioning as tools for management to bring about the desired change. The article shows that while the newcomer is used as a catalyst for increased use of English and for the creation of a “global mindset,” she is at the same time socialized into the existing Danish egalitarian workplace culture. The analysis also highlights the vulnerable position of the catalyst. While her status as an expert English speaker with the right “global mindset” may confer power in an internationalizing workplace, the position of catalyst is also linked with feelings of exclusion.
Several recent studies have demonstrated that people intuitively make consistent matches between classical music and specific wines. It is not clear, however, what governs such crossmodal mappings. Here, we assess the role of emotion—specifically different dimensional aspects of valence, arousal, and dominance—in mediating such mappings. Participants matched three different red wines to three different pieces of classical music. Subsequently, they made emotion ratings separately for each wine and each musical selection. The results revealed that certain wine–music pairings were rated as being significantly better matches than others. More importantly, there was evidence that the participants’ dominance and arousal ratings for the wines and the music predicted their matching rating for each wine–music pairing. These results therefore support the view that wine–music associations are not arbitrary but can be explained, at least in part, by common emotional associations.
We present a new, efficient frame-semantic parser that labels semantic arguments to FrameNet predicates. Built using an extension to the segmental RNN that emphasizes recall, our basic system achieves competitive performance without any calls to a syntactic parser. We then introduce a method that uses phrase-syntactic annotations from the Penn Treebank during training only, through a multitask objective; no parsing is required at training or test time. This "syntactic scaffold" offers a cheaper alternative to traditional syntactic pipelining, and achieves state-of-the-art performance.
Unstructured text data produced on the internet grows rapidly, and sentiment analysis for short texts becomes a challenge because of the limit of the contextual information they usually contain. Learning good vector representations for sentences is a challenging task and an ongoing research area. Moreover, learning long-term dependencies with gradient descent is difficult in neural network language model because of the vanishing gradients problem. Natural Language Processing (NLP) systems traditionally treat words as discrete atomic symbols; the model can leverage small amounts of information regarding the relationship between the individual symbols. In this paper, we propose ConvLstm, neural network architecture that employs Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) on top of pre-trained word vectors. In our experiments, ConvLstm exploit LSTM as a substitute of pooling layer in CNN to reduce the loss of detailed local information and capture long term dependencies in sequence of sentences. We validate the proposed model on two sentiment datasets IMDB, and Stanford Sentiment Treebank (SSTb). Empirical results show that ConvLstm achieved comparable performances with less parameters on sentiment analysis tasks.
This paper presents a novel neural machine translation model which jointly learns translation and source-side latent graph representations of sentences. Unlike existing pipelined approaches using syntactic parsers, our end-to-end model learns a latent graph parser as part of the encoder of an attention-based neural machine translation model, and thus the parser is optimized according to the translation objective. In experiments, we first show that our model compares favorably with state-of-the-art sequential and pipelined syntax-based NMT models. We also show that the performance of our model can be further improved by pretraining it with a small amount of treebank annotations. Our final ensemble model significantly outperforms the previous best models on the standard Englishto-Japanese translation dataset.
Word embeddings that can capture semantic and syntactic information from contexts have been extensively used for various natural language processing tasks. However, existing methods for learning contextbased word embeddings typically fail to capture sufficient sentiment information. This may result in words with similar vector representations having an opposite sentiment polarity (e.g., good and bad), thus degrading sentiment analysis performance. Therefore, this study proposes a word vector refinement model that can be applied to any pre-trained word vectors (e.g., Word2vec and GloVe). The refinement model is based on adjusting the vector representations of words such that they can be closer to both semantically and sentimentally similar words and further away from sentimentally dissimilar words. Experimental results show that the proposed method can improve conventional word embeddings and outperform previously proposed sentiment embeddings for both binary and fine-grained classification on Stanford Sentiment Treebank (SST).
Discourse parsing is an integral part of understanding information flow and argumentative structure in documents. Most previous research has focused on inducing and evaluating models from the English RST Discourse Treebank. However, discourse treebanks for other languages exist, including Spanish, German, Basque, Dutch and Brazilian Portuguese. The treebanks share the same underlying linguistic theory, but differ slightly in the way documents are annotated. In this paper, we present (a) a new discourse parser which is simpler, yet competitive (significantly better on 2/3 metrics) to state of the art for English, (b) a harmonization of discourse treebanks across languages, enabling us to present (c) what to the best of our knowledge are the first experiments on crosslingual discourse parsing.
Recent papers have shown that neural networks obtain state-of-the-art performance on several different sequence tagging tasks. One appealing property of such systems is their generality, as excellent performance can be achieved with a unified architecture and without task-specific feature engineering. However, it is unclear if such systems can be used for tasks without large amounts of training data. In this paper we explore the problem of transfer learning for neural sequence taggers, where a source task with plentiful annotations (e.g., POS tagging on Penn Treebank) is used to improve performance on a target task with fewer available annotations (e.g., POS tagging for microblogs). We examine the effects of transfer learning for deep hierarchical recurrent networks across domains, applications, and languages, and show that significant improvement can often be obtained. These improvements lead to improvements over the current state-of-the-art on several well-studied tasks.
The Web has accumulated a rich source of information, such as text, image, rating, etc, which represent different aspects of user preferences. However, the heterogeneous nature of this information makes it difficult for recommender systems to leverage in a unified framework to boost the performance. Recently, the rapid development of representation learning techniques provides an approach to this problem. By translating the various information sources into a unified representation space, it becomes possible to integrate heterogeneous information for informed recommendation.
Virtual reality (VR) has been proposed as a methodological tool to study the basic science of psychology and other fields. One key advantage of VR is that sharing of virtual content can lead to more robust replication and representative sampling. A database of standardized content will help fulfill this vision. There are two objectives to this study. First, we seek to establish and allow public access to a database of immersive VR video clips that can act as a potential resource for studies on emotion induction using virtual reality. Second, given the large sample size of participants needed to get reliable valence and arousal ratings for our video, we were able to explore the possible links between the head movements of the observer and the emotions he or she feels while viewing immersive VR. To accomplish our goals, we sourced for and tested 73 immersive VR clips which participants rated on valence and arousal dimensions using self-assessment manikins. We also tracked participants' rotational head movements as they watched the clips, allowing us to correlate head movements and affect. Based on past research, we predicted relationships between the standard deviation of head yaw and valence and arousal ratings. Results showed that the stimuli varied reasonably well along the dimensions of valence and arousal, with a slight underrepresentation of clips that are of negative valence and highly arousing. The standard deviation of yaw positively correlated with valence, while a significant positive relationship was found between head pitch and arousal. The immersive VR clips tested are available online as supplemental material.
Positive emotional perceptions and healthy emotional intelligence (EI) are important for social functioning. In this study, we investigated whether loving kindness meditation (LKM) combined with anodal transcranial direct current stimulation (tDCS) would facilitate improvements in EI and changes in affective experience of visual stimuli. LKM has been shown to increase positive emotional experiences and we hypothesized that tDCS could enhance these effects. Eighty-seven undergraduates were randomly assigned to 30 minutes of LKM or a relaxation control recording with anodal tDCS applied to the left dorsolateral prefrontal cortex (left dlPFC) or right temporoparietal junction (right TPJ) at 0.1 or 2.0 milliamps. The primary outcomes were self-reported affect ratings of images from the International Affective Picture System and EI as measured by the Mayer, Salovey and Caruso Emotional Intelligence Test. Results indicated no effects of training on EI, and no main effects of LKM, electrode placement, or tDCS current strength on affect ratings. There was a significant interaction of electrode placement by meditation condition (p = 0.001), such that those assigned to LKM and right TPJ tDCS, regardless of current strength, rated neutral and positive images more positively after training. Results suggest that LKM may enhance positive affective experience.
We systematically explore regularizing neural networks by penalizing low\nentropy output distributions. We show that penalizing low entropy output\ndistributions, which has been shown to improve exploration in reinforcement\nlearning, acts as a strong regularizer in supervised learning. Furthermore, we\nconnect a maximum entropy based confidence penalty to label smoothing through\nthe direction of the KL divergence. We exhaustively evaluate the proposed\nconfidence penalty and label smoothing on 6 common benchmarks: image\nclassification (MNIST and Cifar-10), language modeling (Penn Treebank), machine\ntranslation (WMT'14 English-to-German), and speech recognition (TIMIT and WSJ).\nWe find that both label smoothing and the confidence penalty improve\nstate-of-the-art models across benchmarks without modifying existing\nhyperparameters, suggesting the wide applicability of these regularizers.\n
Large-scale distributed training requires significant communication bandwidth for gradient exchange that limits the scalability of multi-node training, and requires expensive high-bandwidth network infrastructure. The situation gets even worse with distributed training on mobile devices (federated learning), which suffers from higher latency, lower throughput, and intermittent poor connections. In this paper, we find 99.9% of the gradient exchange in distributed SGD is redundant, and propose Deep Gradient Compression (DGC) to greatly reduce the communication bandwidth. To preserve accuracy during compression, DGC employs four methods: momentum correction, local gradient clipping, momentum factor masking, and warm-up training. We have applied Deep Gradient Compression to image classification, speech recognition, and language modeling with multiple datasets including Cifar10, ImageNet, Penn Treebank, and Librispeech Corpus. On these scenarios, Deep Gradient Compression achieves a gradient compression ratio from 270x to 600x without losing accuracy, cutting the gradient size of ResNet-50 from 97MB to 0.35MB, and for DeepSpeech from 488MB to 0.74MB. Deep gradient compression enables large-scale distributed training on inexpensive commodity 1Gbps Ethernet and facilitates distributed training on mobile. Code is available at: https://github.com/synxlin/deep-gradient-compression.
With the advent of word embeddings, lexicons are no longer fully utilized for sentiment analysis although they still provide important features in the traditional setting. This paper introduces a novel approach to sentiment analysis that integrates lexicon embeddings and an attention mechanism into Convolutional Neural Networks. Our approach performs separate convolutions for word and lexicon embeddings and provides a global view of the document using attention. Our models are experimented on both the SemEval'16 Task 4 dataset and the Stanford Sentiment Treebank and show comparative or better results against the existing state-of-the-art systems. Our analysis shows that lexicon embeddings allow building high-performing models with much smaller word embeddings, and the attention mechanism effectively dims out noisy words for sentiment analysis.
We investigate the effective memory depth of RNN models by using them for $n$-gram language model (LM) smoothing. Experiments on a small corpus (UPenn Treebank, one million words of training data and 10k vocabulary) have found the LSTM cell with dropout to be the best model for encoding the $n$-gram state when compared with feed-forward and vanilla RNN models. When preserving the sentence independence assumption the LSTM $n$-gram matches the LSTM LM performance for $n=9$ and slightly outperforms it for $n=13$. When allowing dependencies across sentence boundaries, the LSTM $13$-gram almost matches the perplexity of the unlimited history LSTM LM. LSTM $n$-gram smoothing also has the desirable property of improving with increasing $n$-gram order, unlike the Katz or Kneser-Ney back-off estimators. Using multinomial distributions as targets in training instead of the usual one-hot target is only slightly beneficial for low $n$-gram orders. Experiments on the One Billion Words benchmark show that the results hold at larger scale: while LSTM smoothing for short $n$-gram contexts does not provide significant advantages over classic N-gram models, it becomes effective with long contexts ($n > 5$); depending on the task and amount of data it can match fully recurrent LSTM models at about $n=13$. This may have implications when modeling short-format text, e.g. voice search/query LMs. Building LSTM $n$-gram LMs may be appealing for some practical situations: the state in a $n$-gram LM can be succinctly represented with $(n-1)*4$ bytes storing the identity of the words in the context and batches of $n$-gram contexts can be processed in parallel. On the downside, the $n$-gram context encoding computed by the LSTM is discarded, making the model more expensive than a regular recurrent LSTM LM.
This paper describes and compares two straightforward approaches for dependency parsing with partial annotations (PA). The first approach is based on a forest-based training objective for two CRF parsers, i.e., a biaffine neural network graph-based parser (Biaffine) and a traditional log-linear graph-based parser (LLGPar). The second approach is based on the idea of constrained decoding for three parsers, i.e., a traditional linear graph-based parser (LGPar), a globally normalized neural network transition-based parser (GN3Par) and a traditional linear transition-based parser (LTPar). For the test phase, constrained decoding is also used for completing partial trees. We conduct experiments on Penn Treebank under three different settings for simulating PA, i.e., random, most uncertain, and divergent outputs from the five parsers. The results show that LLGPar is most effective in directly learning from PA, and other parsers can achieve best performance when PAs are completed into full trees by LLGPar.
Recurrent neural networks (RNNs), such as long short-term memory networks (LSTMs), serve as a fundamental building block for many sequence learning tasks, including machine translation, language modeling, and question answering. In this paper, we consider the specific problem of word-level language modeling and investigate strategies for regularizing and optimizing LSTM-based models. We propose the weight-dropped LSTM which uses DropConnect on hidden-to-hidden weights as a form of recurrent regularization. Further, we introduce NT-ASGD, a variant of the averaged stochastic gradient method, wherein the averaging trigger is determined using a non-monotonic condition as opposed to being tuned by the user. Using these and other regularization strategies, we achieve state-of-the-art word level perplexities on two data sets: 57.3 on Penn Treebank and 65.8 on WikiText-2. In exploring the effectiveness of a neural cache in conjunction with our proposed model, we achieve an even lower state-of-the-art perplexity of 52.8 on Penn Treebank and 52.0 on WikiText-2.
This paper proposes a reinforcing method that refines the output layers of existing Recurrent Neural Network (RNN) language models. We refer to our proposed method as Input-to-Output Gate (IOG). IOG has an extremely simple structure, and thus, can be easily combined with any RNN language models. Our experiments on the Penn Treebank and WikiText-2 datasets demonstrate that IOG consistently boosts the performance of several different types of current topline RNN language models.
Because of the importance of the information conveyed by the clinical documents and owing to the large quantity of raw texts produced in the healthcare system, it became a determinant challenge, in the NLP research field, to arrange the extraction and the management of meaningful data, starting from real text occurrences. In this paper we approach a corpus of 5000 medical diagnoses with sophisticated linguistic and computational devices, which are able to access the semantic dimension of words and sentences contained in it. Our morphosemantic method is grounded on a list of neoclassical formative elements pertaining to the medical domain which has been used for the automatic creation and population of medical lexical resources. The outcomes of this work are automatically built electronic dictionaries and thesauri and an annotated corpus for the NLP in the medical domain.
Because of the importance of the information conveyed by the clinical documents and owing to the large quantity of raw texts produced in the healthcare system, it became a determinant challenge, in the NLP research field, to arrange the extraction and the management of meaningful data, starting from real text occurrences. In this paper we approach a corpus of 5000 medical diagnoses with sophisticated linguistic and computational devices, which are able to access the semantic dimension of words and sentences contained in it. Our morphosemantic method is grounded on a list of neoclassical formative elements pertaining to the medical domain which has been used for the automatic creation and population of medical lexical resources. The outcomes of this work are automatically built electronic dictionaries and thesauri and an annotated corpus for the NLP in the medical domain.
A number of firms in northern Europe and especially in Denmark are owned by private foundations similarly to what would have been the case if the Ford Foundation had owned a majority of the shares in Ford Motor Company. Foundation-owned companies appear to perform surprisingly well in terms of profitability and growth, despite lacking governance mechanisms such as profit incentives or takeover threats. Given their non-profit ownership, they might be expected to behave more responsibly towards stakeholders, such as employees or customers (Hansmann, 1980), but so far there has been little empirical evidence to support this hypothesis. This paper presents new research on the reputation and responsibility of foundation-owned companies. In a panel of large Danish companies 2001–11 we find that foundation-owned firms have better reputations and are regarded as more socially responsible in corporate image ratings. Secondary evidence on labour market behaviour is consistent with these findings. Using matched employer–employee data we show that foundation-owned companies are more stable employers, pay their employees better, and keep them for longer. Altogether, the evidence indicates that foundation-ownership is associated with more responsible business behaviour towards employees.
We present a novel recurrent neural network (RNN) based model that combines the remembering ability of unitary RNNs with the ability of gated RNNs to effectively forget redundant/irrelevant information in its memory. We achieve this by extending unitary RNNs with a gating mechanism. Our model is able to outperform LSTMs, GRUs and Unitary RNNs on several long-term dependency benchmark tasks. We empirically both show the orthogonal/unitary RNNs lack the ability to forget and also the ability of GORU to simultaneously remember long term dependencies while forgetting irrelevant information. This plays an important role in recurrent neural networks. We provide competitive results along with an analysis of our model on many natural sequential tasks including the bAbI Question Answering, TIMIT speech spectrum prediction, Penn TreeBank, and synthetic tasks that involve long-term dependencies such as algorithmic, parenthesis, denoising and copying tasks.
Adversarial training (AT) is a powerful regularization method for neural networks, aiming to achieve robustness to input perturbations. Yet, the specific effects of the robustness obtained from AT are still unclear in the context of natural language processing. In this paper, we propose and analyze a neural POS tagging model that exploits AT. In our experiments on the Penn Treebank WSJ corpus and the Universal Dependencies (UD) dataset (27 languages), we find that AT not only improves the overall tagging accuracy, but also 1) prevents over-fitting well in low resource languages and 2) boosts tagging accuracy for rare / unseen words. We also demonstrate that 3) the improved tagging performance by AT contributes to the downstream task of dependency parsing, and that 4) AT helps the model to learn cleaner word representations. 5) The proposed AT model is generally effective in different sequence labeling tasks. These positive results motivate further use of AT for natural language tasks.
BACKGROUND: Blunted facial affect is a common negative symptom of schizophrenia. Additionally, assessing the trustworthiness of faces is a social cognitive ability that is impaired in schizophrenia. Currently available pharmacological agents are ineffective at improving either of these symptoms, despite their clinical significance. The hypothalamic neuropeptide oxytocin has multiple prosocial effects when administered intranasally to healthy individuals and shows promise in decreasing negative symptoms and enhancing social cognition in schizophrenia. Although two small studies have investigated oxytocin's effects on ratings of facial trustworthiness in schizophrenia, its effects on facial expressivity have not been investigated in any population. METHOD: We investigated the effects of oxytocin on facial emotional expressivity while participants performed a facial trustworthiness rating task in 33 individuals with schizophrenia and 35 age-matched healthy controls using a double-blind, placebo-controlled, cross-over design. Participants rated the trustworthiness of presented faces interspersed with emotionally evocative photographs while being video-recorded. Participants' facial expressivity in these videos was quantified by blind raters using a well-validated manualized approach (i.e. the Facial Expression Coding System; FACES). RESULTS: While oxytocin administration did not affect ratings of facial trustworthiness, it significantly increased facial expressivity in individuals with schizophrenia (Z = -2.33, p = 0.02) and at trend level in healthy controls (Z = -1.87, p = 0.06). CONCLUSIONS: These results demonstrate that oxytocin administration can increase facial expressivity in response to emotional stimuli and suggest that oxytocin may have the potential to serve as a treatment for blunted facial affect in schizophrenia.
Motivated form-meaning mappings are pervasive in signlanguages, and iconicity has recently been shown to facilitatesign learning from early on. This study investigated the role oficonicity for language acquisition in Turkish Sign Language(TID). Participants were 43 signing children (aged 10 to 45months) of deaf parents. Sign production ability was recordedusing the adapted version of MacArthur Bates CommunicativeDevelopmental Inventory (CDI) consisting of 500 items forTID. Iconicity and familiarity ratings for a subset of 104 signswere available. Our results revealed that the iconicity of a signwas positively correlated with the percentage of childrenproducing a sign and that iconicity significantly predicted thepercentage of children producing a sign, independent offamiliarity or phonological complexity. Our results areconsistent with previous findings on sign language acquisitionand provide further support for the facilitating effect of iconicform-meaning mappings in sign learning.
This study explores the translators’ agency and translation norms in institutional translation, namely at the European Central Bank (E.C.B.), illustrated by metaphors translated from English into Romanian. Two methods are used in the study: (1) quantitative and qualitative analysis of a corpus of texts produced at the E.C.B. – around 2 million tokens each for the English and the Romanian language version sub-corpora and (2) ethnographic observation. The strategies employed for translating metaphorical expressions belonging to the conceptual metaphor THE ECONOMY IS A MACHINE are used to illustrate the choices translators make. The findings of the study show that the translators use a number of strategies, from retaining the metaphor and the metaphorical expression to completely demetaphorising the expressions. Their choices are influenced by internal requirements of the institution (e.g. using translation memory tools) as well as by the linguistic norms and language conventions of Romanian – the target language, the formal style of the documents, and the expectations of the target audience. Moreover, different expressions in English are translated using the same expression in Romanian, which shows that the Romanian economic vocabulary is still developing and this fact also influences translation choices.
This paper describes the system for our participation of team Wanghao-ftd-SJTU in the CoNLL 2017 Shared Task: Multilingual Parsing from Raw Text to Universal Dependencies. In this work, we design a system based on UDPipe 1 for universal dependency parsing, where transitionbased models are trained for different treebanks. Our system directly takes raw texts as input, performing several intermediate steps like tokenizing and tagging, and finally generates the corresponding dependency trees. For the special surprise languages for this task, we adopt a delexicalized strategy and predict based on transfer learning from other related languages. In the final evaluation of the shared task, our system achieves a result of 66.53% in macro-averaged LAS F1-score.
This paper investigates interactions in parser performance for the two official standards for written Norwegian: Bokmål and Nynorsk. We demonstrate that while applying models across standards yields poor performance, combining the training data for both standards yields better results than previously achieved for each of them in isolation. This has immediate practical value for processing Norwegian, as it means that a single parsing pipeline is sufficient to cover both varieties, with no loss in accuracy. Based on the Norwegian Universal Dependencies treebank we present results for multiple taggers and parsers, experimenting with different ways of varying the training data given to the learners, including the use of machine translation.
Discourse-annotated corpora are an important resource for the community, but they are often annotated according to different frameworks. This makes comparison of the annotations difficult, thereby also preventing researchers from searching the corpora in a unified way, or using all annotated data jointly to train computational systems. Several theoretical proposals have recently been made for mapping the relational labels of different frameworks to each other, but these proposals have so far not been validated against existing annotations. The two largest discourse relation annotated resources, the Penn Discourse Treebank and the Rhetorical Structure Theory Discourse Treebank, have however been annotated on the same text, allowing for a direct comparison of the annotation layers. We propose a method for automatically aligning the discourse segments, and then evaluate existing mapping proposals by comparing the empirically observed against the proposed mappings. Our analysis highlights the influence of segmentation on subsequent discourse relation labeling, and shows that while agreement between frameworks is reasonable for explicit relations, agreement on implicit relations is low. We identify several sources of systematic discrepancies between the two annotation schemes and discuss consequences of these discrepancies for future annotation and for the training of automatic discourse relation labellers.
We present a simple LSTM-based transition-based dependency parser. Our model is composed of a single LSTM hidden layer replacing the hidden layer in the usual feed-forward network architecture. We also propose a new initialization method that uses the pre-trained weights from a feed-forward neural network to initialize our LSTM-based model. We also show that using dropout on the input layer has a positive effect on performance. Our final parser achieves a 93.06% unlabeled and 91.01% labeled attachment score on the Penn Treebank. We additionally replace LSTMs with GRUs and Elman units in our model and explore the effectiveness of our initialization method on individual gates constituting all three types of RNN units.
We explore the concept of hybrid grammars, which formalize and generalize a range of existing frameworks for dealing with discontinuous syntactic structures. Covered are both discontinuous phrase structures and non-projective dependency structures. Technically, hybrid grammars are related to synchronous grammars, where one grammar component generates linear structures and another generates hierarchical structures. By coupling lexical elements of both components together, discontinuous structures result. Several types of hybrid grammars are characterized. We also discuss grammar induction from treebanks. The main advantage over existing frameworks is the ability of hybrid grammars to separate discontinuity of the desired structures from time complexity of parsing. This permits exploration of a large variety of parsing algorithms for discontinuous structures, with different properties. This is confirmed by the reported experimental results, which show a wide variety of running time, accuracy, and frequency of parse failures.
We tackle the prediction of instructor intervention in student posts from discussion forums in Massive Open Online Courses (MOOCs). Our key finding is that using automatically obtained discourse relations improves the prediction of when instructors intervene in student discussions, when compared with a state-of-the-art, feature-rich baseline. Our supervised classifier makes use of an automatic discourse parser which outputs Penn Discourse Treebank (PDTB) tags that represent in-post discourse features. We show PDTB relation-based features increase the robustness of the classifier and complement baseline features in recalling more diverse instructor intervention patterns. In comprehensive experiments over 14 MOOC offerings from several disciplines, the PDTB discourse features improve performance on average. The resultant models are less dependent on domain-specific vocabulary, allowing them to better generalize to new courses.
Representing multiple compositions of human language has been a difficult task due to the complex hierarchical and compositional nature of language. Hierarchical structures are one of the architectures which can be used to capture such compositionalities. In this paper, we introduce temporal hierarchies to the Neural Language Model (NLM) with the help of a Deep Gated Recurrent Neural Network with adaptive timescales to help represent multiple compositions of language. We demonstrate that by representing multiple compositions of language in a deep recurrent neural network architecture, we can improve the performance of Language Models without complex hierarchical architectures. We report the performance of the proposed model using the popular Penn Treebank (PTB) dataset. The results show that by using the multiple timescale concept in an NLM, we can achieve better perplexities compared to the existing baselines.
Unsupervised dependency parsing aims to learn a dependency parser from unannotated sentences. Existing work focuses on either learning generative models using the expectation-maximization algorithm and its variants, or learning discriminative models using the discriminative clustering algorithm. In this paper, we propose a new learning strategy that learns a generative model and a discriminative model jointly based on the dual decomposition method. Our method is simple and general, yet effective to capture the advantages of both models and improve their learning results. We tested our method on the UD treebank and achieved a state-ofthe-art performance on thirty languages.