Grammatical Dependencies Research Articles

Integrating graph convolutional networks to enhance prompt learning for biomedical relation extraction

Background and Objective:Biomedical relation extraction aims to reveal the relation between entities in medical texts. Currently, the relation extraction models that have attracted much attention are mainly to fine-tune the pre-trained language models (PLMs) or add template prompt learning, which also limits the ability of the model to deal with grammatical dependencies. Graph convolutional networks (GCNs) can play an important role in processing syntactic dependencies in biomedical texts. Methods:In this work, we propose a biomedical relation extraction model that fuses GCNs enhanced prompt learning to handle limitations in syntactic dependencies and achieve good performance. Specifically, we propose a model that combines prompt learning with GCNs for relation extraction, by integrating the syntactic dependency information analyzed by GCNs into the prompt learning model, by predicting the correspondence with [MASK] tokens labels for relation extraction. Results:Our model achieved F1 scores of 85.57%, 80.15%, 95.10%, and 84.11% in the biomedical relation extraction datasets GAD, ChemProt, PGR, and DDI, respectively, all of which outperform some existing baseline models. Conclusions:In this paper, we propose enhancing prompt learning through GCNs, integrating syntactic information into biomedical relation extraction tasks. Experimental results show that our proposed method achieves excellent performance in the biomedical relation extraction task.

Aspect-level sentiment analysis based on aspect-sentence graph convolution network

Aspect-level sentiment analysis aims to identify the sentiment polarity of aspect words in sentences. The existing research methods only focus on the grammatical dependencies between words, ignoring the semantic connections between aspect words and the dependency types between words, which limits the performance of the aspect-level sentiment analysis model. Therefore, this paper proposes an aspect-sentence Graph Convolutional Networks model (ASGCN) to perceive more comprehensive semantic information. Specifically, the model consists of sentence-focused GCN (SentenceGCN) and aspects-focused GCN (AspectsGCN) sub models. In the SentenceGCN model, this paper proposes a method to calculate the adjacency matrix (As) of syntactic dependency graph, which uses the position encoding mechanism and pays attention to the influence of different dependency types on semantics, so that SentenceGCN can capture the semantic information of the whole sentence more comprehensively. In the AspectsGCN model, this paper also proposes a method to calculate the adjacency matrix (Aa) of aspect words, which models the relational graph as fully connected and gives weight to the edges between aspect words according to the position, so that the AspectsGCN can pay attention to the semantic relation between different aspect words in the sentence. The proposed model outperforms all baseline models with 86.34 % accuracy and 79.96 F1 score, which indicates that there are more advantages in perceiving semantic information in ASGCN.

Grammatical dependencies shape compositional units in sentence production

Five recall-based structural priming experiments tested the predictions about dative structural priming derived from a new theory of structure building in sentence production. When both prime and target sentences contained direct object filler-gap dependencies, repeating a dative verb enhanced dative priming (the lexical boost). In contrast, the lexical boost was not observed when only target sentences contained object filler-gap dependencies. Additionally, the lexical boost was not observed when prime and target sentences contained object filler-gap dependencies but had mismatching tenses/aspects. In contrast, when neither prime nor target sentences contained object filler-gap dependencies, the lexical boost was observed despite prime and target sentences having different tenses/aspects. These findings confirm the unique set of predictions of the proposed theory, which posits that the size of compositional units is affected by the dependency structures of sentences.

Assessing spoken lexical and lexicogrammatical proficiency using features of word, bigram, and dependency bigram use

AbstractThe measurement of second language (L2) productive lexical proficiency has driven a great deal of research over the past two decades. Research has indicated that more proficient speakers and writers tend to use a wider range of words and that more proficient writers tend to use words that are more sophisticated (less frequent in reference corpora). Research over the past 15 years has also demonstrated that the way words are used in context (i.e., collocation use) is also an important indicator of both written and spoken proficiency. In this study, we extend recent research that has modeled writing proficiency using collocation indices based on grammatical dependencies (e.g., verb–direct object) to spoken contexts. In particular, we model speaking proficiency scores from a large corpus of oral proficiency interview responses using a range of well‐known indices of productive proficiency and newly developed grammatical dependency indices. The results indicated that all index types demonstrated small to moderate correlations with speaking proficiency individually but explained a large proportion of the variance when used in a multivariate model that included dependency collocation indices.

Prosodic cues enhance infants’ sensitivity to nonadjacent regularities

In language, grammatical dependencies often hold between items that are not immediately adjacent to each other. Acquiring these nonadjacent dependencies is crucial for learning grammar. However, there are potentially infinitely many dependencies in the language input. How does the infant brain solve this computational learning problem? Here, we demonstrate that while rudimentary sensitivity to nonadjacent regularities may be present relatively early, robust and reliable learning can only be achieved when convergent statistical and perceptual, specifically prosodic cues, are both present, helping the infant brain detect the building blocks that form a nonadjacent dependency. This study contributes to our understanding of the neural foundations of rule learning that pave the way for language acquisition.

Does Gendered Language Exist in a Foreign Language Context? A Study in Written Discourse of Saudi Male and Female EFL Learners

This investigation aims to verify the hypothesis that gendered language exists in foreign language usage as manifested in the EFL learners’ writing output in a Saudi university. The motivation behind this is to, ultimately, check and weed out elements of gendered language that are early on embedded amongst the genders which, with the passage of time, lead to various biases. Though it may be utopian to think of eradicating linguistic bias, yet this study hopes to contribute meaningfully to curb it and substitute the temptation of allowing it to seep into communication, by educating and training EFL users to substitute these with gender-free language and helping ensure greater gender equality. The writing output of 42 EFL learners was analyzed using an electronic parsing tool called Stanford Parser (v. 3.7.0), and all components classified as grammatical dependencies. Results showed that differences existed in male and female writings in the use of noun in subject position which occurred more frequently in female text at 69.15 mean dependency occurrence. This has long been held as a marked feature of female language use. The same dependency stands at a much lower 51.28 mean occurrence in males. Further, female writing has long been associated with a great deal of use of ‘empty’ modifiers, such as ‘very’ which act as modifiers to adverbs and adjectives. The current study upheld this contention as well. Lastly, backchanneling occurred more in female output in EFL than their male counterparts.

A syntax‒aware deep‒learning model for biomedical semantic role labelling

A deep learning model for biomedical semantic role labeling was build. Semantic role labeling is a useful task that enables the computer to comprehend the key facts expressed in each sentence, and is a necessary first step in the resolution of several other semantic-related tasks, such as event extraction, entity extraction, and Q-A systems... Semantic role labeling is a domain-dependent task. In the biomedical field, semantics are transmitted via more intricate grammatical structures and dependencies in addition to being built on a predicate argument frameset that differs greatly from that of the general domain. To effectively account for these unique characteristics, three types of information were integrated into this deep learning model: Context knowledge obtained from a pre-trained language model trained on a substantial corpus of biomedical texts, dependencies derived from the dependency parse trees and sentence structure obtained from constituency parse trees. To handle grammatical information that is naturally represented as graphs, the Graph Attention Network which is well-known for its remarkable graph learning capabilities, was used. To further boost the model effectiveness, predicate indicator embedding was additionally included in the proposed model. According to experimental findings, the two above-indicated forms of syntactic information along with the predicate indicator embedding, could boost F1 by up to 20%.

Grammatical resumption and dependency processing in Southern Jordanian Arabic

Abstract This study investigates the sensitivity of grammatical resumption to islands in wh-interrogative and relative clause dependencies in Southern Jordanian Arabic (JA). An offline acceptability judgment task and an eye-tracking reading experiment were conducted. The results reveal that resumption in southern JA exhibits sensitivity to strong islands, such as adjunct islands, in both dependencies. The findings also suggest that the southern JA parser posits a resumptive pronoun (RP) inside islands that allow resumption. However, the parser does not predict an RP inside islands that disallow resumption. Furthermore, quantitative data show that wh-interrogative and relative clause dependencies pattern similarly in their sensitivity to islands.

Examining Language Interactivity Through Classifier Masked Primes in Chinese-English Bilinguals

One universal characteristic of human languages is grammatical agreement dependencies. In Mandarin Chinese, classifiers, or measure words, are an obligatory grammatical class used to characterize a quantified noun. The classifier is inserted in between the numeral and the noun (i.e., 一张票 yī zhāng piào, one ticket). English lacks classifiers, so their use in Chinese may lead to cross-language influence for bilinguals and second language learners. Previous research has investigated language interactivity in bilinguals, commonly through the masked priming lexical decision task (LDT), a psycholinguistic task where participants decide if a sequence of characters form a real word or not. This study investigated Chinese-English bilinguals at the University of Florida (n= 24). The experimental group consisted of Chinese-dominant bilinguals who have been in an immersive English environment for at least one semester. The control group included monolingual or dominant English bilingual speakers with another language other than Chinese. Participants completed two LDTs in a within–subjects design. Part 1 featured Chinese (L1) classifier masked primes in an overt English (L2) LDT. Part 2 featured L1 translated masked primes sharing a common classifier with target words in the overt L2 LDT. The results indicated no interaction effect between group and congruency conditions in either part (p = 0.334; p = 0.889). These findings reveal no effect of classifier primes in Chinese-English bilinguals, suggesting minimal, or the absence of cross-language influence in this context.

Preverbal infants' sensitivity to grammatical dependencies.

During their first months of life, infants can already distinguish function words (e.g., pronouns and determiners) from content words (e.g., verbs and nouns). Little research has explored preverbal infants' sensitivity to the relationships between these word categories. This preregistered study examines whether French-learning 8- and 11-month-olds track the grammatical dependencies between determiners and nouns as well as pronouns and verbs. Using the Visual Fixation Procedure, infants were presented with lists containing either grammatical (e.g., tu manges "you eat", des biberons "some bottles") or ungrammatical (e.g., des manges "some eat", tu biberons "you bottle") phrases. In Experiment 1 (N=59), the lists involved common nouns and verbs, while in Experiment 2 (N=28), only common verbs were used. Eleven-month-olds showed a clear preference for correct over incorrect co-occurrences in both experiments, while 8-month-olds showed a trend in the same direction. These results suggest that before their first birthday, infants' storage and access of words and word sequences are sufficiently sophisticated to include the means to track categorical dependencies. This early sensitivity to co-occurrence patterns may be greatly beneficial for constraining lexical access and later on for learning novel words' syntactic and semantic properties.

Deep Learning Classification Model for English Translation Styles Introducing Attention Mechanism

Both short-distance association knowledge and long-distance interaction knowledge in the knowledge base contain rich semantics. When learning entity and relation representation in a knowledge base, if we can learn short-distance association knowledge and long-distance interaction knowledge at the same time, we can learn the representation method with rich semantic information and keep the original structure of the knowledge base. Among the knowledge contained in a large number of records, some knowledge reflects individual characteristics and can be called local knowledge; others reflect group characteristics and can be called global knowledge. Using different ways to learn local and global knowledge in the deep learning model will better reflect the difference between the two kinds of knowledge at the model level, and make the model have the ability to understand both individual characteristics and overall characteristics. Through layer-by-layer forward propagation and error back propagation algorithms, the entire network is gradually optimized in an “end-to-end” manner. This “end-to-end” approach leaves some means of introducing prior knowledge flexibly into the model. Although it can reduce the burden on researchers, this “data-driven” approach brings the shortcomings of poor interpretability of learning results and weak generalization ability. Combining the specific prior knowledge implicit in the data with the deep learning algorithm can optimize the algorithm in a targeted manner and avoid blind searching in the solution space, so as to obtain a model with better performance and wider use. To this end, this paper investigates combining prior knowledge with deep learning to design efficient algorithms to address the classification of English translation styles. This paper combines local knowledge with global knowledge and deep learning methods and proposes a memory neural network method combining local knowledge and global knowledge. By recording the local knowledge in the local memory module and simultaneously recording the global knowledge in the global memory module, the method effectively learns the latent information in a large number of records. This paper combines short-distance association knowledge with long-distance interaction knowledge and a distributed representation learning method based on deep learning and proposes a deep learning method combining short-distance association knowledge and long-distance interaction knowledge. On the IWSLT English translation task, experiments show that the method significantly improves translation quality, confirming that grammatical dependencies enhance attention by supplementing dependent grammatical information, resulting in more effective and richer context vectors that more accurately represent contextual situations. Additional experimental analysis showed that the model underwent careful parameter selection and analysis. By mining valuable long-distance interactive knowledge in the knowledge base and using it in the distributed representation learning of the knowledge base, while constraining the short-distance related knowledge and constraining the long-distance interactive knowledge, the learned knowledge can be used to effectively complete the knowledge base distributed representation for discovering new relations.

The Extended Merge Hypothesis and the Fundamental Principle of Grammar

This paper discusses the main minimalist theory within the Minimalist Program, something I dub the (Weak) Merge Hypothesis (MH). (1) The (Weak) Merge Hypothesis (MH): Merge is a central G operation. I suggest that we extend (1) by adding to it a general principle that I dub the Fundamental Principle of Grammar (FPG). (2) The Fundamental Principle of Grammar (FPG): α and β can be grammatically related. (G-related) only if α and β have merged. Adding (1) and (2) gives us the Strong Merge Hypothesis. (3) The Strong Merge Hypothesis (SMH): All grammatical dependencies are mediated by Merge. SMH has some interesting consequences which the rest of the paper briefly reviews. Highlights include the Movement Theory of Construal, The Periscope Property on selection, as well as preserving the standard results from the Weak Merge Hypothesis.

Applying phraseological complexity measures to L2 French

Abstract This study partially replicates Paquot’s (2018, 2019) study of phraseological complexity in L2 English by investigating how phraseological complexity compares across proficiency levels as well as how phraseological complexity measures relate to lexical, syntactic and morphological complexity measures in a corpus of L2 French argumentative essays. Phraseological complexity is operationalized as the diversity (root type-token ratio; RTTR) and sophistication (pointwise mutual information; PMI) of three types of grammatical dependencies: adjectival modifiers, adverbial modifiers and direct objects. Results reveal a significant increase in the mean PMI of direct objects and the RTTR of adjectival modifiers across proficiency levels. In addition to phraseological sophistication, important predictors of proficiency include measures of lexical diversity, lexical sophistication, syntactic (phrasal) complexity and morphological complexity. The results provide cross-linguistic validation for the results of Paquot (2018, 2019) and further highlight the importance of including phraseological measures in the current repertoire of L2 complexity measures.

Eighteen-month-old infants represent nonlocal syntactic dependencies

Significance As mature speakers of a language, we are able to produce and understand an indefinite number of sentences. This ability comes from a powerful cognitive system—syntax—whose properties reveal the types of computations that human minds can engage in. One core property is the capacity to encode abstract grammatical dependencies that can hold at a distance. When does this property emerge in development? We identify a nonlocal dependency that 18-mo-old infants, but not younger infants, represent syntactically. These findings suggest that the second year of life is a period of active syntactic development and that even before they regularly produce full sentences of their own, infants demonstrate some of the core computational capacities that syntax relies on.

The Drama in the Sentence: Sequence as a Crucial Challenge for Literary Translation from and to Korean

In translation, carefully-crafted sentences are exposed to myriad dangers. This is because translators tend to prioritize syntactical fidelity at the expense of sequence, that is, the order of elements insofar as this relates to calculated progression, gradual disclosure of information, and cumulative development of meaning. But if sequence is turned around for the sake of fluency (conforming to the target language’s ostensibly “natural” word order), the reader’s experience changes as well. Through a set of examples drawn from English translations of Korean fiction, this article demonstrates that the common disregard for sequence is tantamount to a neglect of drama and suspense, of narrative perspectivation, of rhetorical sophistication and cognitive effect. But we also see that by favoring functional equivalence over imitation of grammatical dependencies, it is perfectly possible to allow the reader to process all information at a pace that is analogous to that of the original. Our findings provide insights that are of significance for other language pairings as well.

Translating Collocations from English into Arabic and vice versa: An Empirical Study

Collocations are defined as ''the frequent co-occurrence of lexical items that naturally share the characteristics of semantic and grammatical dependencies"(Ibrahim, 2003: iii). In translation, collocations are considered a factor that makes translation more effective and powerful. However, translating collocations is an everlasting struggle for most students of translation. The present study aims at investigating the challenges that Sudanese EFL university students encounter when rendering English collocations into their Arabic equivalences and vice versa as well as the reasons behind these challenges. To this end, 26 Sudanese EFL students, between 20-30 years old, studying at Nahda College in Sudan, were selected. A diagnostic test composed of two questions is used as a tool for data collection. Frequencies, percentages, mean, and standard deviation are used to analyse the collected data. The results of this study manifests that Sudanese EFL university students encounter difficulties in translating collocations from English into Arabic and vice versa; the causes of these difficulties are due to students’ unawareness of the linguistic and cultural differences between the two languages as well as their heavy reliance on literal translation strategy.

Desirable Difficulties in Language Learning? How Talker Variability Impacts Artificial Grammar Learning.

Contending with talker variability has been found to lead to processing costs but also benefits by focusing learners on invariant properties of the signal, indicating that talker variability acts as a desirable difficulty. That is, talker variability may lead to initial costs followed by long-term benefits for retention and generalization. Adult participants learned an artificial grammar affording learning of multiple components in two experiments varying in difficulty. They learned from one, two, or eight talkers and were tested at three time points. The eight-talker condition did not impact learning. The two-talker condition negatively impacted some aspects of learning, but only under more difficult conditions. Generalization of the grammatical dependency was difficult. Thus, we discovered that high and limited talker variability can differentially impact artificial grammar learning. However, talker variability did not act as a desirable difficulty in the current paradigm as the few evidenced costs were not related to long-term benefits.

A 3-phase approach based on sequential mining and dependency parsing for enhancing hypernym patterns performance

AbstractPatterns have been extensively used to extract hypernym relations from texts. The most popular patterns are Hearst’s patterns, formulated as regular expressions mainly based on lexical information. Experiences have reported good precision and low recall for such patterns. Thus, several approaches have been developed for improving recall. While these approaches perform better in terms of recall, it remains quite difficult to further increase recall without degrading precision. In this paper, we propose a novel 3-phase approach based on sequential pattern mining to improve pattern-based approaches in terms of both precision and recall by (i) using a rich pattern representation based on grammatical dependencies (ii) discovering new hypernym patterns, and (iii) extending hypernym patterns with anti-hypernym patterns to prune wrong extracted hypernym relations. The results obtained by performing experiments on three corpora confirm that using our approach, we are able to learn sequential patterns and combine them to outperform existing hypernym patterns in terms of precision and recall. The comparison to unsupervised distributional baselines for hypernym detection shows that, as expected, our approach yields much better performance. When compared to supervised distributional baselines for hypernym detection, our approach can be shown to be complementary and much less loosely coupled with training datasets and corpora.

An enhanced lexicon-based approach for sentiment analysis: a case study on illegal immigration

PurposeResearch on sentiment analysis were mostly conducted on product and services, resulting in scarcity of studies focusing on social issues, which may require different mechanisms due to the nature of the issue itself. This paper aims to address this gap by developing an enhanced lexicon-based approach.Design/methodology/approachAn enhanced lexicon-based approach was employed using General Inquirer, incorporated with multi-level grammatical dependencies and the role of verb. Data on illegal immigration were gathered from Twitter for a period of three months, resulting in 694,141 tweets. Of these, 2,500 tweets were segregated into two datasets for evaluation purposes after filtering and pre-processing.FindingsThe enhanced approach outperformed ten online sentiment analysis tools with an overall accuracy of 81.4 and 82.3% for dataset 1 and 2, respectively as opposed to ten other sentiment analysis tools.Originality/valueThe study is novel in the sense that data pertaining to a social issue were used instead of products and services, which require different mechanism due to the nature of the issue itself.

Toddlers track hierarchical structure dependence

ABSTRACT Previous research suggests that toddlers can rely on distributional cues in the input to track adjacent and nonadjacent grammatical dependencies. It remains unclear whether toddlers understand the hierarchical phrase structures that determine the corresponding grammatical dependencies. We addressed this question by testing toddlers on two different phrase structures in French that govern distinct patterns of grammatical gender feature agreement. We first show that the two structures are in fact extremely infrequent in children’s input. Then we report on a preferential looking experiment in which French-learning 30-month-olds were presented with French sentences in the two structures, and the grammaticality of the feature agreement was manipulated. Crucially, the contrasting structures contained the same sequentially ordered feature-bearing words in our design (e.g., correct agreement: La banane FEMININE dans le chapeau MASCLINE, elle FEMININE … ‘The banana in the hat, it …’; incorrect agreement: *La banane FEMININE et le chapeau MASCLINE, elles FEMININE … ‘The banana and the hat, they …’). Thus, children must go beyond those words and distinguish the structures hierarchically in order to determine the grammaticality of the different agreement patterns. We predicted that if toddlers rely solely on input support to acquire the structures and the agreement, they should show no discrimination of feature grammaticality. If, however, they can rely on mechanisms beyond linear input, such as the principle of structure dependence in the universal grammar (UG), they should show a grammaticality effect. The results confirmed the latter prediction, demonstrating the possible influence of the learner’s internal system, suggesting that UG knowledge might guide early language development.