Text-to-scene conversion systems map natural language text to formal representations required for visual scenes. The difficulty involved in this mapping is one of the most critical challenges for developing these systems. The current study mapped Persian natural language text as the headmost system to a conceptual scene model. This conceptual scene model is an intermediate semantic representation between natural language and the visual scene and contains descriptions of visual elements of the scene. It will be used to produce meaningful animation based on an input story in this ongoing study. The mapping task was modeled as a sequential labeling problem, and a conditional random field (CRF) model was trained and tested for sequential labeling of scene model elements. To the best of the authors’ knowledge, no dataset for this task exists; thus, the required dataset was collected for this task. The lack of required off-the-shelf natural language processing modules and a significant error rate in the available corpora were important challenges to dataset collection. Some features of the dataset were manually annotated. The results were evaluated using standard text classification metrics, and an average accuracy of 85.7% was obtained, which is satisfactory.

Recently, neural abstractive text summarization (NATS) models based on sequence-to-sequence architecture have drawn a lot of attention. Real-world texts that need to be summarized range from short news with dozens of words to long reports with thousands of words. However, most existing NATS models are not good at summarizing long documents, due to the inherent limitations of their underlying neural architectures. In this paper, we focus on the task of long document summarization (LDS). Based on the inherent section structures of source documents, we divide an abstractive LDS problem into several smaller-sized problems. In this circumstance, how to provide a less-biased target summary as the supervision for each section is vital for the model’s performance. As a preliminary, we formally describe the section-to-summary-sentence (S2SS) alignment for LDS. Based on this, we propose a novel NATS framework for the LDS task. Our framework is built based on the theory of unbalanced optimal transport (UOT), and it is named as UOTSumm. It jointly learns three targets in a unified training objective, including the optimal S2SS alignment, a section-level NATS summarizer, and the number of aligned summary sentences for each section. In this way, UOTSumm directly learns the text alignment from summarization data, without resorting to any biased tool such as ROUGE. UOTSumm can be easily adapted to most existing NATS models. And we implement two versions of UOTSumm, with and without the pretrain-finetune technique. We evaluate UOTSumm on three publicly available LDS benchmarks: PubMed, arXiv, and GovReport. UOTSumm obviously outperforms its counterparts that use ROUGE for the text alignment. When combined with UOTSumm, the performance of two vanilla NATS models improves by a large margin. Besides, UOTSumm achieves better or comparable performance when compared with some recent strong baselines.

Text style transfer (TST) aims at automatically changing a text’s stylistic features, such as formality, sentiment, authorial style, humor, and complexity, while still trying to preserve its content. Although the scientific community has investigated TST since the 1980s, it has recently regained attention by adopting deep unsupervised strategies to address the challenge of training without parallel data. In this manuscript, we investigate how relying on sequence-to-sequence pretraining models affects the performance of TST when the pretraining step leverages pairs of paraphrase data. Furthermore, we propose a new technique to enhance the sequence-to-sequence model by distilling knowledge from masked language models. We evaluate our proposals on three unsupervised style transfer tasks with widely used benchmarks: author imitation, formality transfer, and polarity swap. The evaluation relies on quantitative and qualitative analyses and comparisons with the results of state-of-the-art models. For the author imitation and the formality transfer task, we show that using the proposed techniques improves all measured metrics and leads to state-of-the-art (SOTA) results in content preservation and an overall score in the author imitation domain. In the formality transfer domain, we paired with the SOTA method in the style control metric. Regarding the polarity swap domain, we show that the knowledge distillation component improves all measured metrics. The paraphrase pretraining increases content preservation at the expense of harming style control. Based on the results reached in these domains, we also discuss in the manuscript if the tasks we address have the same nature and should be equally treated as TST tasks.

Arabic sentiment analysis models have recently employed compositional paragraph or sentence embedding features to represent the informal Arabic dialectal content. These embeddings are mostly composed via ordered, syntax-aware composition functions and learned within deep neural network architectures. With the differences in the syntactic structure and words’ order among the Arabic dialects, a sentiment analysis system developed for one dialect might not be efficient for the others. Here we present syntax-ignorant, sentiment-specific n-gram embeddings for sentiment analysis of several Arabic dialects. The novelty of the proposed model is illustrated through its features and architecture. In the proposed model, the sentiment is expressed by embeddings, composed via the unordered additive composition function and learned within a shallow neural architecture. To evaluate the generated embeddings, they were compared with the state-of-the art word/paragraph embeddings. This involved investigating their efficiency, as expressive sentiment features, based on the visualisation maps constructed for our n-gram embeddings and word2vec/doc2vec. In addition, using several Eastern/Western Arabic datasets of single-dialect and multi-dialectal contents, the ability of our embeddings to recognise the sentiment was investigated against word/paragraph embeddings-based models. This comparison was performed within both shallow and deep neural network architectures and with two unordered composition functions employed. The results revealed that the introduced syntax-ignorant embeddings could represent single and combinations of different dialects efficiently, as our shallow sentiment analysis model, trained with the proposed n-gram embeddings, could outperform the word2vec/doc2vec models and rival deep neural architectures consuming, remarkably, less training time.

This paper proposes a new method for semantic document analysis: densification, which identifies and ranks Wikipedia pages relevant to a given document. Although there are similarities with established tasks such as wikification and entity linking, the method does not aim for strict disambiguation of named entity mentions. Instead, densification uses existing links to rank additional articles that are relevant to the document, a form of explicit semantic indexing that enables higher-level semantic retrieval procedures that can be beneficial for a wide range of NLP applications. Because a gold standard for densification evaluation does not exist, a study is carried out to investigate the level of agreement achievable by humans, which questions the feasibility of creating an annotated data set. As a result, a semi-supervised approach is employed to develop a two-stage densification system: filtering unlikely candidate links and then ranking the remaining links. In a first evaluation experiment, Wikipedia articles are used to automatically estimate the performance in terms of recall. Results show that the proposed densification approach outperforms several wikification systems. A second experiment measures the impact of integrating the links predicted by the densification system into a semantic question answering (QA) system that relies on Wikipedia links to answer complex questions. Densification enables the QA system to find twice as many additional answers than when using a state-of-the-art wikification system.

The paper presents background and motivation for a processing model that segments discourse into units that are simple, non-nested clauses, prior to the recognition of clause internal phrasal constituents, and experimental results in support of this model. One set of results is derived from a statistical reanalysis of the Swedish empirical data in Strangert, Ejerhed and Huber 1993 concerning the linguistic structure of major prosodic units. The other set of results is derived from experiments in segmenting part of speech annotated Swedish text corpora into clauses, using a new clause segmentation algorithm. The clause segmented corpus data is taken from the Stockholm Umeå Corpus (SUC), 1 M words of Swedish texts from different genres, part of speech annotated by hand, and from the Umeå corpus DAGENS INDUSTRI 1993 (DI93), 5 M words of Swedish financial newspaper text, processed by fully automatic means consisting of tokenizing, lexical analysis, and probabilistic POS tagging. The results of these two experiments show that the proposed clause segmentation algorithm is 96% correct when applied to manually tagged text, and 91% correct when applied to probabilistically tagged text.

Automatic pronunciation of unknown words (i.e., those not in the system dictionary) is a difficult problem in text-to-speech (TTS) synthesis. Currently, many data-driven approaches have been applied to the problem, as a backup strategy for those cases where dictionary matching fails. The difficulty of the problem depends on the complexity of spelling-to-sound mappings according to the particular writing system of the language. Hence, the degree of success achieved varies widely across languages but also across dictionaries, even for the same language with the same method. Further, the sizes of the training and test sets are an important consideration in data-driven approaches. In this paper, we study the variation of letter-to-phoneme transcription accuracy across seven European languages with twelve different lexicons. We also study the relationship between the size of dictionary and the accuracy obtained. The largest dictionaries of each language have been partitioned into ten approximately equal-sized subsets and combined to give ten different-sized test sets. In view of its superior performance in previous work, the transcription method used is pronunciation by analogy (PbA). Best results are obtained for Spanish, generally believed to have a very regular (‘shallow’) orthography, and poorest results for English, a language whose irregular spelling system is legendary. For those languages for which multiple dictionaries were available (i.e., French and English), results were found to vary across dictionaries. For the relationship between dictionary size and transcription accuracy, we find that as dictionary size grows, so performance grows monotonically. However, the performance gain decelerates (tends to saturate) as the dictionary increases in size; the relation can simply be described by a logarithmic regression, one parameter of which (α) can be taken as quantifying the depth of orthography of a language. We find that α for a language is significantly correlated with transcription performance on a small dictionary (approximately 10,000 words) for that language, but less so for asymptotic performance. This may be because our measure of asymptotic performance is unreliable, being extrapolated from the fitted logarithmic regression.

Text normalization is the task of mapping noncanonical language, typical of speech transcription and computer-mediated communication, to a standardized writing. This task is especially important for languages such as Swiss German, with strong regional variation and no written standard. In this paper, we propose a novel solution for normalizing Swiss German WhatsApp messages using the encoder–decoder neural machine translation (NMT) framework. We enhance the performance of a plain character-level NMT model with the integration of a word-level language model and linguistic features in the form of part-of-speech (POS) tags. The two components are intended to improve the performance by addressing two specific issues: the former is intended to improve the fluency of the predicted sequences, whereas the latter aims at resolving cases of word-level ambiguity. Our systematic comparison shows that our proposed solution results in an improvement over a plain NMT system and also over a comparable character-level statistical machine translation system, considered the state of the art in this task till recently. We perform a thorough analysis of the compared systems’ output, showing that our two components produce indeed the intended, complementary improvements.

In this paper, a novel phrase alignment strategy combining linguistic knowledge and cooccurrence measures extracted from bilingual corpora is presented. The algorithm is mainly divided into four steps, namely phrase selection and classification, phrase alignment, one-to-one word alignment and postprocessing. The first stage selects a linguistically-derived set of phrases that convey a unified meaning during translation and are therefore aligned together in parallel texts. These phrases include verb phrases, idiomatic expressions and date expressions. During the second stage, very high precision links between these selected phrases for both languages are produced. The third step performs a statistical word alignment using association measures and link probabilities with the remaining unaligned tokens, and finally the fourth stage takes final decisions on unaligned tokens based on linguistic knowledge. Experiments are reported for an English-Spanish parallel corpus, with a detailed description of the evaluation measure and manual reference used. Results show that phrase cooccurrence measures convey a complementary information to word cooccurrences and a stronger evidence of a correct alignment, successfully introducing linguistic knowledge in a statistical word alignment scheme. Precision, Recall and Alignment Error Rate (AER) results are presented, outperforming state-of-the-art alignment algorithms.

This paper proposes a novel application of topic models to do entity relation detection (ERD). In order to make use of the latent semantics of text, we formulate the task of relation detection as a topic modeling problem. The motivation is to find underlying topics that are indicative of relations between named entities (NEs). Our approach considers pairs of NEs and features associated with them as mini documents, and aims to utilize the underlying topic distributions as indicators for the types of relations that may exist between the NE pair. Our system, ERD-MedLDA, adapts Maximum Entropy Discriminant Latent Dirichlet Allocation (MedLDA) with mixed membership for relation detection. By using supervision, ERD-MedLDA is able to learn topic distributions indicative of relation types. Further, ERD-MedLDA is a topic model that combines the benefits of both, maximum likelihood estimation (MLE) and maximum margin estimation (MME), and the mixed-membership formulation enables the system to incorporate heterogeneous features. We incorporate different features into the system and perform experiments on the ACE 2005 corpus. Our approach achieves better overall performance for precision, recall, and F-measure metrics as compared to baseline SVM-based and LDA-based models. We also find that our system shows better and consistent improvements with the addition of complex informative features as compared to baseline systems.

This paper presents a new type of nonlinear discourse structure found to be very common in free English texts. This structure reflects nonlinear presentation of the information and knowledge conveyed by the texts. It is argued that such nonlinearity is representationally and informationally advantageous because it allows one to create smaller, more compact texts. The paper presents a heuristics-based, relatively domain-independent algorithm for computing this new text structure. The paper discusses good quantitative and qualitative performance of the algorithm, and presents the results of the extensive tests on a large volume of free English texts.

Usually unsupervised dependency parsers try to optimize the probability of a corpus by revising the dependency model that is assumed to have generated the corpus. In this paper we explore a different view in which a dependency structure is, among other things, a partial order on the nodes in terms of centrality or saliency. Under this assumption we directly model centrality and derive dependency trees from the ordering of words. The result is an approach to unsupervised dependency parsing that is very different from standard ones in that it requires no training data. The input words are ordered by centrality, and a parse is derived from the ranking using a simple deterministic parsing algorithm, relying on the universal dependency rules defined by Naseem et al. (Naseem, T., Chen, H., Barzilay, R., Johnson, M. 2010. Using universal linguistic knowledge to guide grammar induction. In Proceedings of Empirical Methods in Natural Language Processing, Boston, MA, USA, pp. 1234–44.). Our approach is evaluated on data from twelve different languages and is remarkably competitive.

In this paper, we experiment with several techniques to solve the problem of lexical substitution, both in a lexical sample as well as an all-words setting, and compare the benefits of combining multiple lexical resources using both unsupervised and supervised approaches. Overall in the lexical sample setting, the results obtained through the combination of several resources exceed the current state-of-the-art when selecting the best substitute for a given target word, and place second when selecting the top ten substitutes, thus demonstrating the usefulness of the approach. Further, we put forth a novel exploration in all-words lexical substitution and set ground for further explorations of this more generalized setting.

Light verb constructions (LVCs) are verb and noun combinations in which the verb has lost its meaning to some degree and the noun is used in one of its original senses, typically denoting an event or an action. They exhibit special linguistic features, especially when regarded in a multilingual context. In this paper, we focus on the automatic detection of LVCs in raw text in four different languages, namely, English, German, Spanish, and Hungarian. First, we analyze the characteristics of LVCs from a linguistic point of view based on parallel corpus data. Then, we provide a standardized (i.e., language-independent) representation of LVCs that can be used in machine learning experiments. After, we experiment on identifying LVCs in different languages: we exploit language adaptation techniques which demonstrate that data from an additional language can be successfully employed in improving the performance of supervised LVC detection for a given language. As there are several annotated corpora from several domains in the case of English and Hungarian, we also investigate the effect of simple domain adaptation techniques to reduce the gap between domains. Furthermore, we combine domain adaptation techniques with language adaptation techniques for these two languages. Our results show that both out-domain and additional language data can improve performance. We believe that our language adaptation method may have practical implications in several fields of natural language processing, especially in machine translation.

Logic Forms (LF) are simple, first-order logic knowledge representations of natural language sentences. Each noun, verb, adjective, adverb, pronoun, preposition and conjunction generates a predicate. LF systems usually identify the syntactic function by means of syntactic rules but this approach is difficult to apply to languages with a high syntax flexibility and ambiguity, for example, Spanish. In this study, we present a mixed method for the derivation of the LF of sentences in Spanish that allows the combination of hard-coded rules and a classifier inspired on semantic role labeling. Thus, the main novelty of our proposal is the way the classifier is applied to generate the predicates of the verbs, while rules are used to translate the rest of the predicates, which are more straightforward and unambiguous than the verbal ones. The proposed mixed system uses a supervised classifier to integrate syntactic and semantic information in order to help overcome the inherent ambiguity of Spanish syntax. This task is accomplished in a similar way to the semantic role labeling task. We use properties extracted from the AnCora-ES corpus in order to train a classifier. A rule-based system is used in order to obtain the LF from the rest of the phrase. The rules are obtained by exploring the syntactic tree of the phrase and encoding the syntactic production rules. The LF algorithm has been evaluated by using shallow parsing with some straightforward Spanish phrases. The verb argument labeling task achieves 84% precision and the proposed mixed LFi method surpasses 11% a system based only on rules.

The LinGO grammar consists of a specification of a type system and of various typed feature structures which are well-formed according to the type system. The typed feature structures function as grammar rules, lexical rules and lexical entries. There are several variant typed feature structure formalisms, with different computational properties, so in this appendix we very briefly specify the version assumed by the LinGO grammar.

This appendix is necessarily terse, and is only intended to allow a reader who already has a knowledge of typed feature structures to understand the specific formalism used in the LinGO grammar. The definitions given below basically follow Carpenter (1992), with the notion of type constraint from Copestake (1992). For formal details of typed feature structures in general see Carpenter (1992). A detailed account of the specific assumptions made here is given in Copestake (1999) (See Chapter 4 for an introduction, and Chapter 5 for a semi-formal account.)

Note that the LinGO grammar uses a very restricted formalism. For instance, it does not utilize disjunctive feature structures, negation, implication, inequalities, defaults, set-valued features, extensionality or relational constraints. Constraint resolution does not require that every type be made maximally specific, and the type inference system is essentially non-recursive. The recursive power necessary in grammars is explicitly encoded via rules, which are expressed as typed feature structures, but interpreted as phrase structure rules.

Computational power needs have greatly increased during the last years, and this is also the case in the Natural Language Processing (NLP) area, where thousands of documents must be processed, i.e., linguistically analyzed, in a reasonable time frame. These computing needs have implied a radical change in the computing architectures and big-scale text processing techniques used in NLP. In this paper, we present a scalable architecture for distributed language processing. The architecture uses Storm to combine diverse NLP modules into a processing chain, which carries out the linguistic analysis of documents. Scalability requires designing solutions that are able to run distributed programs in parallel and across large machine clusters. Using the architecture presented here, it is possible to integrate a set of third-party NLP modules into a unique processing chain which can be deployed onto a distributed environment, i.e., a cluster of machines, so allowing the language-processing modules run in parallel. No restrictions are placed a priori on the NLP modules apart of being able to consume and produce linguistic annotations following a given format. We show the feasibility of our approach by integrating two linguistic processing chains for English and Spanish. Moreover, we provide several scripts that allow building from scratch a whole distributed architecture that can be then easily installed and deployed onto a cluster of machines. The scripts and the NLP modules used in the paper are publicly available and distributed under free licenses. In the paper, we also describe a series of experiments carried out in the context of the NewsReader project with the goal of testing how the system behaves in different scenarios.

Machine learning techniques have been implemented to extract instances of semantic relations using diverse features based on linguistic knowledge, such as tokens, lemmas, PoS-tags, or dependency paths. However, there has been little work aiming to know which of these features works better in the relation extraction task, and less in languages other than English. In this paper, various features representing different levels of linguistic knowledge are systematically evaluated for biographical relation extraction. The effectiveness of these features was measured by training several supervised classifiers that only differ in the type of linguistic knowledge used to define their features. The experiments performed in this paper show that some basic linguistic knowledge (provided by lemmas and their combination in bigrams) behaves better than other complex features, such as those based on syntactic analysis. Furthermore, some feature combinations using different levels of analysis are proposed in order (i) to avoid feature overlapping as well as (ii) to evaluate the use of computationally inexpensive and widespread tools such as tokenization and lemmatization. This paper also describes two new freely available corpora for biographical relation extraction in Portuguese and Spanish, built by means of a distant-supervision strategy. Experiments were performed with five semantic relations and two languages, using these corpora.

Benchmarks can be a useful step toward the goals of the field (when the benchmark is on the critical path), as demonstrated by the GLUE benchmark, and deep nets such as BERT and ERNIE. The case for other benchmarks such as MUSE and WN18RR is less well established. Hopefully, these benchmarks are on a critical path toward progress on bilingual lexicon induction (BLI) and knowledge graph completion (KGC). Many KGC algorithms have been proposed such as Trans[DEHRM], but it remains to be seen how this work improves WordNet coverage. Given how much work is based on these benchmarks, the literature should have more to say than it does about the connection between benchmarks and goals. Is optimizing P@10 on WN18RR likely to produce more complete knowledge graphs? Is MUSE likely to improve Machine Translation?

Deep neural networks as an end-to-end approach lack robustness from an application point of view, as it is very difficult to fix an obvious problem without retraining the model, for example, when a model consistently predicts positive when seeing the word “terrible.” Meanwhile, it is less stressed that the commonly used attention mechanism is likely to “over-fit” by being overly sparse, so that some key positions in the input sequence could be overlooked by the network. To address these problems, we proposed a lexicon-enhanced attention LSTM model in 2019, named ATLX. In this paper, we describe extended experiments and analysis of the ATLX model. And, we also try to further improve the aspect-based sentiment analysis system by combining a vector-based sentiment domain adaptation method.