In this article, we introduce an extended, freely available resource for the Romanian language, named RoLEX. The dataset was developed mainly for speech processing applications, yet its applicability extends beyond this domain. RoLEX includes over 330,000 curated entries with information regarding lemma, morphosyntactic description, syllabification, lexical stress and phonemic transcription. The process of selecting the list of word entries and semi-automatically annotating the complete lexical information associated with each of the entries is thoroughly described.

The dataset’s inherent knowledge is then evaluated in a task of concurrent prediction of syllabification, lexical stress marking and phonemic transcription. The evaluation looked into several dataset design factors, such as the minimum viable number of entries for correct prediction, the optimisation of the minimum number of required entries through expert selection and the augmentation of the input with morphosyntactic information, as well as the influence of each task in the overall accuracy. The best results were obtained when the orthographic form of the entries was augmented with the complete morphosyntactic tags. A word error rate of 3.08% and a character error rate of 1.08% were obtained this way. We show that using a carefully selected subset of entries for training can result in a similar performance to the performance obtained by a larger set of randomly selected entries (twice as many). In terms of prediction complexity, the lexical stress marking posed most problems and accounts for around 60% of the errors in the predicted sequence.

The Portable Extendable Traffic Information Collator (POETIC) is an information extraction system that extracts traffic information from free text occurring in police incident logs and initiates (simulated) broadcasts of traffic bulletins to motorists when appropriate. POETIC is a second stage prototype system; the initial prototype (TIC, see Evans and Hartley 1990) was limited to the practices and requirements of a single police force. In POETIC, the architecture and data representations have been generalised to make the system tailorable to many different police force ‘domains’. In this paper we describe these developments, and report on tests of the system on authentic input data from three police domains.

In machine-learning applications, data selection is of crucial importance if good runtime performance is to be achieved. In a scenario where the test set is accessible when the model is being built, training instances can be selected so they are the most relevant for the test set. Feature Decay Algorithms (FDA) are a technique for data selection that has demonstrated excellent performance in a number of tasks. This method maximizes the diversity of the n-grams in the training set by devaluing those ones that have already been included. We focus on this method to undertake deeper research on how to select better training data instances. We give an overview of FDA and propose improvements in terms of speed and quality. Using German-to-English parallel data, first we create a novel approach that decreases the execution time of FDA when multiple computation units are available. In addition, we obtain improvements on translation quality by extending FDA using information from the parallel corpus that is generally ignored.

In written Chinese, personal pronouns are commonly dropped when they can be inferred from context. This practice is particularly common in informal genres like Short Message Service messages sent via cell phones. Restoring dropped personal pronouns can be a useful preprocessing step for information extraction. Dropped personal pronoun recovery can be divided into two subtasks: (1) detecting dropped personal pronoun slots and (2) determining the identity of the pronoun for each slot. We address a simpler version of restoring dropped personal pronouns wherein only the person numbers are identified. After applying a word segmenter, we used a linear-chain conditional random field to predict which words were at the start of an independent clause. Then, using the independent clause start information, as well as lexical and syntactic information, we applied a conditional random field or a maximum-entropy classifier to predict whether a dropped personal pronoun immediately preceded each word and, if so, the person number of the dropped pronoun. We conducted a series of experiments using a manually annotated corpus of Chinese Short Message Service. Our approaches substantially outperformed a rule-based approach based partially on rules developed by Chung and Gildea (2010, Effects of Empty Categories on Machine Translation. Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP). Association for Computational Linguistics. pp. 636–45). Our approaches also outperformed (though by a considerably smaller margin) a machine-learning approach based closely on work by Yang, Liu, and Xue in (2015, Recovering Dropped Pronouns from Chinese Text Messages. Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics (ACL). Association for Computational Linguistics. pp. 309–13). Features derived from parsing largely did not help our approaches. We conclude that, given independent clause start information, the parse information we used was largely superfluous for identifying dropped personal pronouns.

The specificities of the Arabic language, mainly agglutination and vocalization make the task of POS-tagging more difficult than for Indo-European languages. Consequently, POS-tagging texts with good accuracy remains a challenging problem for Arabic language processing applications. In this work, we consider the task of POS-tagging as an optimization problem modeled as a graph whose nodes correspond to all possible grammatical tags given by a morphological analyzer for words in a sentence and the goal is to find the best path (sequence of tags) in this graph. To resolve this problem, we propose a novel approach based on ant colony. Ant colony-based algorithms are among the most efficient methods to resolve optimization problems modeled as a graph. The collaboration of ants having various knowledge creates a collective intelligence and increases efficiency. We have performed experiments on both vocalized and non-vocalized texts and tested two different tagsets containing fine and coarse grained composite tags. The obtained results showed good accuracy rates and hence, the benefits of swarm intelligence for the POS-tagging problem.

Deep learning approaches are superior in natural language processing due to their ability to extract informative features and patterns from languages. The two most successful neural architectures are LSTM and transformers, used in large pretrained language models such as BERT. While cross-lingual approaches are on the rise, most current natural language processing techniques are designed and applied to English, and less-resourced languages are lagging behind. In morphologically rich languages, information is conveyed through morphology, for example, through affixes modifying stems of words. The existing neural approaches do not explicitly use the information on word morphology. We analyse the effect of adding morphological features to LSTM and BERT models. As a testbed, we use three tasks available in many less-resourced languages: named entity recognition (NER), dependency parsing (DP) and comment filtering (CF). We construct baselines involving LSTM and BERT models, which we adjust by adding additional input in the form of part of speech (POS) tags and universal features. We compare the models across several languages from different language families. Our results suggest that adding morphological features has mixed effects depending on the quality of features and the task. The features improve the performance of LSTM-based models on the NER and DP tasks, while they do not benefit the performance on the CF task. For BERT-based models, the added morphological features only improve the performance on DP when they are of high quality (i.e., manually checked) while not showing any practical improvement when they are predicted. Even for high-quality features, the improvements are less pronounced in language-specific BERT variants compared to massively multilingual BERT models. As in NER and CF datasets manually checked features are not available, we only experiment with predicted features and find that they do not cause any practical improvement in performance.

In almost every science fiction movie you’ll see people conversing with machines. Of course, the rise of intelligent personal assistants means you probably do this yourself already. This posting asks: what’s the difference? Also, recent news on Facebook acquisitions, spoken language translation, and sentiment analysis.

Specialised pre-trained language models are becoming more frequent in Natural language Processing (NLP) since they can potentially outperform models trained on generic texts. BioBERT (Sanh et al., Distilbert, a distilled version of bert: smaller, faster, cheaper and lighter. arXiv preprint arXiv: 1910.01108, 2019) and BioClinicalBERT (Alsentzer et al., Publicly available clinical bert embeddings. In Proceedings of the 2nd Clinical Natural Language Processing Workshop, pp. 72–78, 2019) are two examples of such models that have shown promise in medical NLP tasks. Many of these models are overparametrised and resource-intensive, but thanks to techniques like knowledge distillation, it is possible to create smaller versions that perform almost as well as their larger counterparts. In this work, we specifically focus on development of compact language models for processing clinical texts (i.e. progress notes, discharge summaries, etc). We developed a number of efficient lightweight clinical transformers using knowledge distillation and continual learning, with the number of parameters ranging from $15$ million to $65$ million. These models performed comparably to larger models such as BioBERT and ClinicalBioBERT and significantly outperformed other compact models trained on general or biomedical data. Our extensive evaluation was done across several standard datasets and covered a wide range of clinical text-mining tasks, including natural language inference, relation extraction, named entity recognition and sequence classification. To our knowledge, this is the first comprehensive study specifically focused on creating efficient and compact transformers for clinical NLP tasks. The models and code used in this study can be found on our Huggingface profile at https://huggingface.co/nlpie and Github page at https://github.com/nlpie-research/Lightweight-Clinical-Transformers, respectively, promoting reproducibility of our results.

The Journal of Natural Language Engineering is now in its 25th year. The editorial preface to the first issue emphasised that the focus of the journal was to be on the practical application of natural language processing (NLP) technologies: the time was ripe for a serious publication that helped encourage research ideas to find their way into real products. The commercialisation of NLP technologies had already started by that point, but things have advanced tremendously over the last quarter-century. So, to celebrate the journal’s anniversary, we look at how commercial NLP products have developed over the last 25 years.

This study examines the ability of a semantic space model to represent the meaning of noun compounds such as ‘information gathering’ or ‘heart disease.’ For a semantic space model to compute the meaning and the attributional similarity (or semantic relatedness) for unfamiliar noun compounds that do not occur in a corpus, the vector for a noun compound must be computed from the vectors of its constituent words using vector composition algorithms. Six composition algorithms (i.e., centroid, multiplication, circular convolution, predication, comparison, and dilation) are compared in terms of the quality of the computation of the attributional similarity for English and Japanese noun compounds. To evaluate the performance of the computation of the similarity, this study uses three tasks (i.e., related word ranking, similarity correlation, and semantic classification), and two types of semantic spaces (i.e., latent semantic analysis-based and positive pointwise mutual information-based spaces). The result of these tasks is that the dilation algorithm is generally most effective in computing the similarity of noun compounds, while the multiplication algorithm is best suited specifically for the positive pointwise mutual information-based space. In addition, the comparison algorithm works better for unfamiliar noun compounds that do not occur in the corpus. These findings indicate that in general a semantic space model, and in particular the dilation, multiplication, and comparison algorithms have sufficient ability to compute the attributional similarity for noun compounds.

For complex tasks such as parse selection, the creation of labelled training sets can be extremely costly. Resource-efficient schemes for creating informative labelled material must therefore be considered. We investigate the relationship between two broad strategies for reducing the amount of manual labelling necessary to train accurate parse selection models: ensemble models and active learning. We show that popular active learning methods for reducing annotation costs can be outperformed by instead using a model class which uses the available labelled data more efficiently. For this, we use a simple type of ensemble model called the Logarithmic Opinion Pool (LOP). We furthermore show that LOPs themselves can benefit from active learning. As predicted by a theoretical explanation of the predictive power of LOPs, a detailed analysis of active learning using LOPs shows that component model diversity is a strong predictor of successful LOP performance. Other contributions include a novel active learning method, a justification of our simulation studies using timing information, and cross-domain verification of our main ideas using text classification.

Nowadays, automated essay evaluation (AEE) systems play an important role in evaluating essays and have been successfully used in large-scale writing assessments. However, existing AEE systems mostly focus on grammar or shallow content measurements rather than higher-order traits such as ideas. This paper proposes a new formulation of graph-based features for concept maps using word embeddings to evaluate the quality of ideas for Chinese compositions. The concept map derived from the student’s composition is composed of the concepts appearing in the essay and the co-occurrence relationship between the concepts. By utilizing real compositions written by eighth-grade students from a large-scale assessment, the scoring accuracy of the computer evaluation system (named AECC-I: Automated Evaluation for Chinese Compositions—Ideas) is higher than the baselines. The results indicate that the proposed method deepens the construct-relevant coverage of automatic ideas evaluation in compositions and that it can provide constructive feedback for students.

Since the release of ChatGPT at the end of November 2022, generative AI has been talked about endlessly in both the technical press and the mainstream media. Large language model technology has been heralded as many things: the disruption of the search engine, the end of the student essay, the bringer of disinformation … but what does it mean for commercial providers of earlier iterations of natural language generation technology? We look at how the major players in the space are responding, and where things might go in the future.

The ACL-2019 Business meeting ended with a discussion of reviewing. Conferences are experiencing a success catastrophe. They are becoming bigger and bigger, which is not only a sign of success but also a challenge (for reviewing and more). Various proposals for reducing submissions were discussed at the Business meeting. IMHO, the problem is not so much too many submissions, but rather, random reviewing. We cannot afford to do reviewing as badly as we do (because that leads to even more submissions). Negative feedback loops are effective. The reviewing process will improve over time if reviewers teach authors how to write better submissions, and authors teach reviewers how to write more constructive reviews. If you have received a not-ok (unhelpful/offensive) review, please help program committees improve by sharing your not-ok reviews on social media.

Reading comprehension (RC) tests involve reading a short passage of text and answering a series of questions pertaining to that text. We present a methodology for evaluation of the application of modern natural language technologies to the task of responding to RC tests. Our work is based on ABCs (Abduction Based Comprehension system), an automated system for taking tests requiring short answer phrases as responses. A central goal of ABCs is to serve as a testbed for understanding the role that various linguistic components play in responding to reading comprehension questions. The heart of ABCs is an abductive inference engine that provides three key capabilities: (1) first-order logical representation of relations between entities and events in the text and rules to perform inference over such relations, (2) graceful degradation due to the inclusion of abduction in the reasoning engine, which avoids the brittleness that can be problematic in knowledge representation and reasoning systems and (3) system transparency such that the types of abductive inferences made over an entire corpus provide cues as to where the system is performing poorly and indications as to where existing knowledge is inaccurate or new knowledge is required. ABCs, with certain sub-components not yet automated, finds the correct answer phrase nearly 35 percent of the time using a strict evaluation metric and 45 percent of the time using a looser inexact metric on held out evaluation data. Performance varied for the different question types, ranging from over 50 percent on who questions to over 10 percent on what questions. We present analysis of the roles of individual components and analysis of the impact of various characteristics of the abductive proof procedure on overall system performance.

In this paper, we present a novel approach to partial parsing that produces dependency links between words of a sentence. The partial parser called a lightweight dependency analyzer uses information encoded in supertags and hence can produce constituency-based as well as dependency-based analyses. The lightweight dependency analyzer has been used for text chunking, including noun and verb group chunking. We also present a proposal for a general framework for parser evaluation that is applicable for evaluating both constituency-based and dependency-based, partial and complete parsers. The performance results of the lightweight dependency analyzer on Wall Street Journal and Brown corpus using the proposed evaluation metrics are discussed.

In this paper, we study the problem of structural analysis of Web documents aiming at extracting the sectional hierarchy of a document. In general, a document can be represented as a hierarchy of sections and subsections with corresponding headings and subheadings. We developed two machine learning models: heading extraction model and hierarchy extraction model. Heading extraction was formulated as a classification problem whereas a tree-based learning approach was employed in hierarchy extraction. For this purpose, we developed an incremental learning algorithm based on support vector machines and perceptrons. The models were evaluated in detail with respect to the performance of the heading and hierarchy extraction tasks. For comparison, a baseline rule-based approach was used that relies on heuristics and HTML document object model tree processing. The machine learning approach, which is a fully automatic approach, outperformed the rule-based approach. We also analyzed the effect of document structuring on automatic summarization in the context of Web search. The results of the task-based evaluation on TREC queries showed that structured summaries are superior to unstructured summaries both in terms of accuracy and user ratings, and enable the users to determine the relevancy of search results more accurately than search engine snippets.

In this paper, we introduce a method to represent phrase structure grammars for building a large annotated corpus of Korean syntactic trees. Korean is different from English in word order and word compositions. As a result of our study, it turned out that the differences are significant enough to induce meaningful changes in the tree annotation scheme for Korean with respect to the schemes for English. A tree annotation scheme defines the grammar formalism to be assumed, categories to be used, and rules to determine correct parses for unsettled issues in parse construction. Korean is partially free in word order and the essential components such as subjects and objects of a sentence can be omitted with greater freedom than in English. We propose a restricted representation of phrase structure grammar to handle the characteristics of Korean more efficiently. The proposed representation is shown by means of an extensive experiment to gain improvements in parsing time as well as grammar size. We also describe the system named Teb that is a software environment set up with a goal to build a tree annotated corpus of Korean containing more than one million units.

Semantically tagging a corpus is useful for many intermediate NLP tasks such as: acquisition of word argument structures in sublanguages; acquisition of syntactic disambiguation cues; terminology learning; etc. The general idea is that semantic tags allow the generalization of observed word patterns, and facilitate the discovery of recurrent sublanguage phenomena and selectional rules of various types. Yet, as opposed to POS tags in morphology, there is no consensus in the literature about the type and granularity of the semantic tags to be used. In this paper, we argue that an appropriate selection of semantic tags should be domain-dependent. We propose a method by which we select from WordNet an inventory of semantic tags that are ‘optimal’ for a given corpus, according to a scoring function defined as a linear combination of general and corpus-dependent performance factors. We believe that an optimal selection of a category inventory is a necessary premise for obtaining better results in all lexically learning algorithms that are based on, or concerned with, semantic categorization of words. Furthermore, an adequate inventory (one which intuitively ‘fits’ with the semantics of a domain, e.g. phenomenon for Natural Science, or part, piece for a technical handbook) may facilitate the manual annotation of large corpora.

The syntactic structure of a nominal compound must be analyzed first for its semantic interpretation. In addition, the syntactic analysis of nominal compounds is very useful for NLP application such as information extraction, since a nominal compound often has a similar linguistic structure with a simple sentence, as well as representing concrete and compound meaning of an object with several nouns combined. In this paper, we present a novel model for structural analysis of nominal compounds using linguistic and statistical knowledge which is coupled based on lexical information. That is, the syntactic relations defined between nouns (complement-predicate and modifier-head relation) are obtained from large corpora and again used to analyze the structures of nominal compounds and identify the underlying relations between nouns. Experiments show that the model gives good results, and can be effectively used for application systems which do not require deep semantic information.