Subwords have become very popular, but the BERTa and ERNIEb tokenizers often produce surprising results. Byte pair encoding (BPE) trains a dictionary with a simple information theoretic criterion that sidesteps the need for special treatment of unknown words. BPE is more about training (populating a dictionary of word pieces) than inference (parsing an unknown word into word pieces). The parse at inference time can be ambiguous. Which parse should we use? For example, “electroneutral” can be parsed as electron-eu-tral or electro-neutral, and “bidirectional” can be parsed as bid-ire-ction-al and bi-directional. BERT and ERNIE tend to favor the parse with more word pieces. We propose minimizing the number of word pieces. To justify our proposal, a number of criteria will be considered: sound, meaning, etc. The prefix, bi-, has the desired vowel (unlike bid) and the desired meaning (bi is Latin for two, unlike bid, which is Germanic for offer).

Distributional semantic word representations are at the basis of most modern NLP systems. Their usefulness has been proven across various tasks, particularly as inputs to deep learning models. Beyond that, much work investigated fine-tuning the generic word embeddings to leverage linguistic knowledge from large lexical resources. Some work investigated context-dependent word token embeddings motivated by word sense disambiguation, using sequential context and large lexical resources. More recently, acknowledging the need for an in-context representation of words, some work leveraged information derived from language modelling and large amounts of data to induce contextualised representations. In this paper, we investigate Syntax-Aware word Token Embeddings (SATokE) as a way to explicitly encode specific information derived from the linguistic analysis of a sentence in vectors which are input to a deep learning model. We propose an efficient unsupervised learning algorithm based on tensor factorisation for computing these token embeddings given an arbitrary graph of linguistic structure. Applying this method to syntactic dependency structures, we investigate the usefulness of such token representations as part of deep learning models of text understanding. We encode a sentence either by learning embeddings for its tokens and the relations between them from scratch or by leveraging pre-trained relation embeddings to infer token representations. Given sufficient data, the former is slightly more accurate than the latter, yet both provide more informative token embeddings than standard word representations, even when the word representations have been learned on the same type of context from larger corpora (namely pre-trained dependency-based word embeddings). We use a large set of supervised tasks and two major deep learning families of models for sentence understanding to evaluate our proposal. We empirically demonstrate the superiority of the token representations compared to popular distributional representations of words for various sentence and sentence pair classification tasks.

This paper is a case study of user involvement in the requirements specification for project ISLE: Interactive Spoken Language Education. Developers of Spoken Language Dialogue Systems should involve users from the outset, particularly if the aim is to develop novel solutions for a generic target application area or market. As well as target end-users, SLDS developers should identify and consult ‘meta-level’ domain experts with expertise in human-to-human dialogue in the target domain. In our case, English language teachers and publishers provided generic knowledge of learners' dialogue preferences; other applications have analogous domain language experts. These domain language experts can help to pin down a domain-specific sublanguage which fits the constraints of current speech recognition technology: linguistically-naive end-users may expect unconstrained conversational English, but in practice, dialogue interactions have to be constrained in vocabulary and syntax. User consultation also highlighted a need to consider how to integrate speech input and output with other modes of interaction and processing; in our case the input speech signal is processed by speech recogniser, stress and mispronunciation detectors, and output responses are text and graphics as well as speech. This suggests a need to revisit the definition of ‘dialogue’: other SLDS developers should also consider the merits of multimodality as an adjunct to pure spoken language dialogue, particularly given that current systems are not capable of accurately handling unconstrained English.

Twitter and other social media platforms are often used for sharing interest in products. The identification of purchase decision stages, such as in the AIDA model (Awareness, Interest, Desire, and Action), can enable more personalized e-commerce services and a finer-grained targeting of advertisements than predicting purchase intent only. In this paper, we propose and analyze neural models for identifying the purchase stage of single tweets in a user’s tweet sequence. In particular, we identify three challenges of purchase stage identification: imbalanced label distribution with a high number of non-purchase-stage instances, limited amount of training data, and domain adaptation with no or only little target domain data. Our experiments reveal that the imbalanced label distribution is the main challenge for our models. We address it with ranking loss and perform detailed investigations of the performance of our models on the different output classes. In order to improve the generalization of the models and augment the limited amount of training data, we examine the use of sentiment analysis as a complementary, secondary task in a multitask framework. For applying our models to tweets from another product domain, we consider two scenarios: for the first scenario without any labeled data in the target product domain, we show that learning domain-invariant representations with adversarial training is most promising, while for the second scenario with a small number of labeled target examples, fine-tuning the source model weights performs best. Finally, we conduct several analyses, including extracting attention weights and representative phrases for the different purchase stages. The results suggest that the model is learning features indicative of purchase stages and that the confusion errors are sensible.

Neural Networks applied to Machine Translation need a finite vocabulary to express textual information as a sequence of discrete tokens. The currently dominant subword vocabularies exploit statistically-discovered common parts of words to achieve the flexibility of character-based vocabularies without delegating the whole learning of word formation to the neural network. However, they trade this for the inability to apply word-level token associations, which limits their use in semantically-rich areas and prevents some transfer learning approaches e.g. cross-lingual pretrained embeddings, and reduces their interpretability. In this work, we propose new hybrid linguistically-grounded vocabulary definition strategies that keep both the advantages of subword vocabularies and the word-level associations, enabling neural networks to profit from the derived benefits. We test the proposed approaches in both morphologically rich and poor languages, showing that, for the former, the quality in the translation of out-of-domain texts is improved with respect to a strong subword baseline.

This paper describes how the use of monadic second-order logic for specifying regular languages can be extended for specifying regular relations, providing a declarative description language for finite state transductions of the sort used in NLP. We discuss issues arising in the integration into an automaton toolkit of an implementation of the conversion from logic formulas to automata. The utility of the logic of regular relations is demonstrated by showing how it can be used to define the family of replacement operators in a way that lends itself to straightforward proofs of correctness.

This paper presents the design and implementation of a finite-state syntactic grammar of Basque that has been used with the objective of extracting information about verb subcategorization instances from newspaper texts. After a partial parser has built basic syntactic units such as noun phrases, prepositional phrases, and sentential complements, a finite-state parser performs syntactic disambiguation, determination of clause boundaries and filtering of the results, in order to obtain a verb occurrence together with its associated syntactic components, either complements or adjuncts. The set of occurrences for each verb is then filtered by statistical measures that distinguish arguments from adjuncts.

Prepositional Phrases (PP) perform a variety of syntactic functions in a conventional sentence, and cause severe problems to computer systems that automatically analyse the sentential syntax. A major issue in this area has been the automatic determination of the syntactic functions of PPs. Most work published so far makes use of the probabilistic approach, and attach PPs to either the antecedent noun or verb phrase. Due to the natural limitation of the probabilistic approach, it is important to evaluate the linguistic behaviour of prepositional phrases and propose qualitative solutions to the problem. In this article, I first provide a detailed account of statistics regarding the frequency of use for (i) types of prepositions, (ii) syntactic categories as realisations of prepositional complements, and (iii) the syntactic functions of prepositional phrases. Statistics reported here all derive from a representative corpus of contemporary British English. I then describe a set of rules that has been implemented in order to label PPs automatically for their syntactic functions. I finally report on the coverage of these rules empirically observed in an experiment which involved a set of naturally occurring PPs as test data.

It is a challenging task to discover information from a large amount of data in an open domain.1 In this paper, an event network framework is proposed to address this challenge. It is in fact an empirical construct for exploring open information, composed of three steps: document event detection, event network construction and event network analysis. First, documents are clustered into document events for reducing the impact of noisy and heterogeneous resources. Secondly, linguistic units (e.g., named entities or entity relations) are extracted from each document event and combined into an event network, which enables content-oriented retrieval. Then, in the final step, techniques such as social network or complex network can be applied to analyze the event network for exploring open information. In the implementation section, we provide examples of exploring open information via event network.

Significant advances have been achieved in bilingual word-level alignment, yet the challenge remains for phrase-level alignment. Moreover, the need for parallel data is a critical drawback for the alignment task. This work proposes a system that alleviates these two problems: a unified phrase representation model using cross-lingual word embeddings as input and an unsupervised training algorithm inspired by recent works on neural machine translation. The system consists of a sequence-to-sequence architecture where a short sequence encoder constructs cross-lingual representations of phrases of any length, then an LSTM network decodes them w.r.t their contexts. After training with comparable corpora and existing key phrase extraction, our encoder provides cross-lingual phrase representations that can be compared without further transformation. Experiments on five data sets show that our method obtains state-of-the-art results on the bilingual phrase alignment task and improves the results of different length phrase alignment by a mean of 8.8 points in MAP.

This study systematically reviews existing approaches to unsupervised grammar induction in terms of their theoretical underpinnings, practical implementations and evaluation. Our motivation is to identify the influence of functional-cognitive schools of grammar on language processing models in computational linguistics. This is an effort to fill any gap between the theoretical school and the computational processing models of grammar induction. Specifically, the review aims to answer the following research questions: Which types of grammar theories have been the subjects of grammar induction? Which methods have been employed to support grammar induction? Which features have been used by these methods for learning? How were these methods evaluated? Finally, in terms of performance, how do these methods compare to one another? Forty-three studies were identified for systematic review out of which 33 described original implementations of grammar induction; three provided surveys and seven focused on theories and experiments related to acquisition and processing of grammar in humans. The data extracted from the 33 implementations were stratified into 7 different aspects of analysis: theory of grammar; output representation; how grammatical productivity is processed; how grammatical productivity is represented; features used for learning; evaluation strategy and implementation methodology. In most of the implementations considered, grammar was treated as a generative-formal system, autonomous and independent of meaning. The parser decoding was done in a non-incremental, head-driven fashion by assuming that all words are available for the parsing model and the output representation of the grammar learnt was hierarchical, typically a dependency or a constituency tree. However, the theoretical and experimental studies considered suggest that a usage-based, incremental, sequential system of grammar is more appropriate than the formal, non-incremental, hierarchical view of grammar. This gap between the theoretical as well as experimental studies on one hand and the computational implementations on the other hand should be addressed to enable further progress in computational grammar induction research.

Non-native speakers of English far outnumber native speakers; English is the main language of books, newspapers, airports, air-traffic control, international business, academic conferences, science, technology, diplomacy, sports, international competitions, pop music, and advertising (British Council 2014). Online education in the form of massive online open courses is also primarily in English—even teaching English. This creates enormous amounts of text written by non-native speakers, which in turn generates a need for grammar correction and analysis. Even aside from massive online open courses, the number of English learners in Asia alone is in the tens of millions. In this paper, we provide a survey of the two main areas of existing work on non-native text analysis, prefaced by an overview of common datasets used by researchers, comparing their attributes and potential uses. Then, an introduction to native language identification follows: determining the native language of an author based on text in the second language. This section is subdivided into various techniques and a shared task on this classification problem. Next, we discuss non-native grammatical error correction—finding and modifying text to fix errors or to make it sound more fluent. Again, we discuss different methods before investigating a relevant shared task. Lastly, we end with conclusions and potential future directions. While this survey primarily focuses on detecting and correcting non-native English text, many approaches are general and can be used across any language pairing.

We aim at finding the minimal set of fragments that achieves maximal parse accuracy in Data Oriented Parsing (DOP). Experiments with the Penn Wall Street Journal (WSJ) treebank show that counts of almost arbitrary fragments within parse trees are important, leading to improved parse accuracy over previous models tested on this treebank. We isolate a number of dependency relations which previous models neglect but which contribute to higher accuracy. We show that the history of statistical parsing models displays a tendency towards using more and larger fragments from training data.

This study aims to improve the performance of identifying grammatical functions between an adnoun clause and a noun phrase in Korean. The key task is to determine the relation between the two constituents in terms of such functional categories as subject, object, adverbial and appositive. The problem is mainly caused by the fact that functional morphemes, which are considered to be crucial for identifying the relation, are omitted in the noun phrases. To tackle this problem, we propose to employ the Support Vector Machines (SVM) in determining the grammatical functions. Through an experiment with a tagged corpus for training SVMs, we found the proposed model to be more useful than both the Maximum Entropy Model (MEM) and the backed-off model.

In this paper, we describe a system for coreference resolution and emphasize the role of evaluation for its design. The goal of the system is to group referring expressions (identified beforehand in narrative texts) into sets of coreferring expressions that correspond to discourse entities. Several knowledge sources are distinguished, such as referential compatibility between a referring expression and a discourse entity, activation factors for discourse entities, size of working memory, or meta-rules for the creation of discourse entities. For each of them, the theoretical analysis of its relevance is compared to scores obtained through evaluation. After looping through all knowledge sources, an optimal behavior is chosen, then evaluated on test data. The paper also discusses evaluation measures as well as data annotation, and compares the present approach to others in the field.

In dialogue systems it is important to label the dialogue turns with dialogue-related meaning. Each turn is usually divided into segments and these segments are labelled with dialogue acts (DAs). A DA is a representation of the functional role of the segment. Each segment is labelled with one DA, representing its role in the ongoing discourse. The sequence of DAs given a dialogue turn is used by the dialogue manager to understand the turn. Probabilistic models that perform DA labelling can be used on segmented or unsegmented turns. The last option is more likely for a practical dialogue system, but it provides poorer results. In that case, a hypothesis for the number of segments can be provided to improve the results. We propose some methods to estimate the probability of the number of segments based on the transcription of the turn. The new labelling model includes the estimation of the probability of the number of segments in the turn. We tested this new approach with two different dialogue corpora: SwitchBoard and Dihana. The results show that this inclusion significantly improves the labelling accuracy.

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.

In this paper, we propose an enhanced version of the vanilla transformer for data-to-text generation and then use it as the generator of a conditional generative adversarial model to improve the semantic quality and diversity of output sentences. Specifically, by adding a diagonal mask matrix to the attention scores of the encoder and using the history of the attention weights in the decoder, this enhanced version of the vanilla transformer prevents semantic defects in the output text. Also, using this enhanced transformer along with a triplet network, respectively, as the generator and discriminator of conditional generative adversarial network, diversity and semantic quality of sentences are guaranteed. To prove the effectiveness of the proposed model, called conditional generative adversarial with enhanced transformer (CGA-ET), we performed experiments on three different datasets and observed that our proposed model is able to achieve better results than the baselines models in terms of BLEU, METEOR, NIST, ROUGE-L, CIDEr, BERTScore, and SER automatic evaluation metrics as well as human evaluation.