Generating Pseudo Connectives with MLMs for Implicit Discourse Relation Recognition

Jiang, Congcong; Qian, Tieyun; Chen, Zhuang; Tang, Kejian; Zhan, Shaohui; Zhan, Tao

doi:10.1007/978-3-030-89363-7_9

Cited by 5 publications

(4 citation statements)

References 35 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…[45] enriches the training dataset by incorporating the corpus of explicitly related arguments. [46,47] expanded pseudo-labeled samples by training the explicit classifier. [48] utilizing explicit data for one-teacher multi-data multi-task learning.…”

Section: Implicit Discourse Relation Recognitionmentioning

confidence: 99%

Syntactic Harmony: Introducing Shared-private Framework for Implicit Discourse Relation Recognition

Ruixia,

Yanbin,

Zheng

2024

Preprint

View full text Add to dashboard Cite

Implicit discourse relation recognition (IDRR) still faces challenges in achieving high accuracy compared to explicit discourse relation recognition. Many existing approaches attempt to combine explicit and implicit data. However, language differences and variations in category distribution make it challenging to use multitask learning to directly optimize IDRR task. This paper considers explicit data as out-of-domain data and implicit data as in-domain data, proposed a shared-private-based framework with dynamic dependency matching (SPDDM) for IDRR task. This framework explores the deep-level syntactic correlation between explicit and implicit data by encoding dependency grammar information through cross-attention mechanism and confuse it to shared-private framework through dynamic matching . Extensive experiments on the popular PDTB datasets show that our model achieves state-of-the-art performance, and a detailed analysis highlights the efficiency of our proposed framework.

show abstract

Section: Implicit Discourse Relation Recognitionmentioning

confidence: 99%

Syntactic Harmony: Introducing Shared-private Framework for Implicit Discourse Relation Recognition

Ruixia,

Yanbin,

Zheng

2024

Preprint

View full text Add to dashboard Cite

show abstract

“…In discourse parsing, it is well-known that there exists a clear gap between explicit (relations that are marked explicitly with a DM) and implicit (relation between two spans of text exists, but is not marked explicitly with a DM) relation classification, namely, 90% vs. 50% of accuracy (respectively) in 4-way classification (as indicated by Shi and Demberg [60]). To improve discourse relation parsing, several works focused on enhancing their systems for implicit relation classification: removing DMs from explicit relations for implicit relation classification data augmentation [6,58]; framing explicit vs. implicit relation classification as a domain adaptation problem [26,53]; learning sentence representations by exploring automatically collected large-scale datasets [44,61]; multi-task learning [31,43]; automatic explicitation of implicit DMs followed by explicit relation classification [28,29,60].…”

Section: Rocha Et Al / Cross-genre Argument Mining: Automatically Fil...mentioning

confidence: 99%

“…However, we work in a more challenging scenario, where the DM augmentation is performed at the paragraph level following an end-to-end approach (i.e., from raw text to a DM-augmented text). Consequently, our approach differs from prior work in multiple ways: (a) we aim to explicitate all discourse relations at the paragraph level, while prior work focuses on one relation at a time [28,29,60,64] (our models can explore a wider context window and take into account the interdependencies between different discourse relations), and (b) we do not require any additional information, such as prior knowledge regarding discourse units boundaries (e.g., clauses or sentences) [28,29,60,64] or the target discourse relations [64].…”

Section: Rocha Et Al / Cross-genre Argument Mining: Automatically Fil...mentioning

confidence: 99%

Cross-genre argument mining: Can language models automatically fill in missing discourse markers?

Rocha,

Lopes Cardoso,

Belouadi

et al. 2024

AAC

View full text Add to dashboard Cite

Available corpora for Argument Mining differ along several axes, and one of the key differences is the presence (or absence) of discourse markers to signal argumentative content. Exploring effective ways to use discourse markers has received wide attention in various discourse parsing tasks, from which it is well-known that discourse markers are strong indicators of discourse relations. To improve the robustness of Argument Mining systems across different genres, we propose to automatically augment a given text with discourse markers such that all relations are explicitly signaled. Our analysis unveils that popular language models taken out-of-the-box fail on this task; however, when fine-tuned on a new heterogeneous dataset that we construct (including synthetic and real examples), they perform considerably better. We demonstrate the impact of our approach on an Argument Mining downstream task, evaluated on different corpora, showing that language models can be trained to automatically fill in discourse markers across different corpora, improving the performance of a downstream model in some, but not all, cases. Our proposed approach can further be employed as an assistant tool for better discourse understanding.

show abstract

“…Pitler and Nenkova (2009) train a classifier with connectives in the text as the only features and find it could achieve over 90% accuracy on explicit relation recognition. Similarly, many attempts have been made using connectives to improve the recognition performance on implicit relations, including pipeline methods (Zhou et al, 2010;Jiang et al 2021), multi-task training (Kishimoto et al, 2020;Long and Webber, 2022), adversarial training (Qin et al, 2017), joint training (Liu and Strube, 2023), and prompt learning (Zhou et al, 2022;Xiang et al, 2023). Our work differs from them in both motivation and application scenarios.…”

Section: Related Workmentioning

confidence: 99%

Theoretical Studies of Natural Gas Hydrates and H-bonded Clusters and Crystals

Liu¹

View full text Add to dashboard Cite

In this thesis H-bonded systems (natural gas hydrates, water clusters, and crystal ice) are studied by density functional theory (DFT) computations.Natural gas hydrates (NGHs) play an important role in energy and environmental fields: NGHs are considered as a promising backup energy resource in the near-future due to their tremendous carbon content; improper exploration of NGHs could induce geological disasters and aggravate the greenhouse effect. In addition, many technologies based on gas hydrates are being applied and developed. The thermodynamic stabilities of various water cavities in different clathrate crystalline phases occupied by hydrocarbon gas molecules are studied by dispersion-corrected hybrid functionals. The Raman spectra of C-C and C-H stretching vibrations of hydrocarbon molecules in various water cavities in the solid state are derived. The trends of C-H stretching vibrational frequencies are found to follow the prediction by the "loose cage ─ tight cage" model. In addition, the trends and origins of 13 C NMR chemical shifts of hydrocarbon molecules in various NGHs are presented. These theoretical results will enlarge the database of C-C and C-H stretching vibrational frequencies and 13 C NMR parameters of hydrocarbon molecules in NGHs, and provide valuable information to help identify the types of clathrate phases and varieties of guest molecules included in NGHs samples taken from natural sites.The behavior of water clusters may help to understand the properties of its liquid and solid states. The thermodynamic stabilities and IR spectra of a small-, medium-, and large-sized water cluster are studied in this work. After full optimization of (H 2 O) 20,54,100 using the hybrid functional B3LYP, the electronic energies, zero-point energies, internal energies, enthalpies, entropies, and Gibbs free energies of the water clusters are computed. The OH stretching vibrational IR spectra of (H 2 O) 20,54,100 are also presented and split into sub-spectra for different H-bond types based on the specific contributions from each group. It is found that the OH stretching vibrational frequencies of water are sensitive to the conformations of the H-bonds and the vibrations of the H-bonds belonging to different types are located in separated regions in the IR spectra. Thus, the spectroscopic fingerprints will reflect the H-bond topology of the water molecules in a water cluster.Ice XI has been suggested to be involved in the process of planetary formation as a considerable electric field might be formed from the ferroelectric ice XI in space. IR and Raman spectroscopic technology can be directly used to identify the occurrence of ferroelectric ice XI in laboratory or extraterrestrial settings. Due to the difficulty for DFT to describe non-covalent systems, the performance of 16 different DFT methods applied on the ice Ih, VIII, IX, and XI crystal phases are assessed. Based on the computational accuracy and cost, the II IR and Raman spectra of ice Ih and XI are derived and compared. The librational vib...

show abstract

Generating Pseudo Connectives with MLMs for Implicit Discourse Relation Recognition

Cited by 5 publications

References 35 publications

Syntactic Harmony: Introducing Shared-private Framework for Implicit Discourse Relation Recognition

Syntactic Harmony: Introducing Shared-private Framework for Implicit Discourse Relation Recognition

Cross-genre argument mining: Can language models automatically fill in missing discourse markers?

Theoretical Studies of Natural Gas Hydrates and H-bonded Clusters and Crystals

Contact Info

Product

Resources

About