Most multi-agent systems rely on fairly fixed collaboration patterns: roles, interaction paths, and task decomposition are set in advance. This works poorly for open-ended, literature-grounded research synthesis, where uncertainty changes what needs to be explored, challenged, and verified. We introduce DyCo, a framework for dynamic cognitive role coordination that combines cognitive role partitioning, dynamic teaming, and Exploration–Verification Alternation (EVA) to balance idea generation with validation. We use the Myers–Briggs Type Indicator (MBTI) only as a structured way to specify role priors under controlled conditions, not as a claim about human personality. On DeepResearch Bench, a diverse DyCo configuration reaches a RACE score of 42.75, outperforming the strongest multi-agent baseline by 4.6% (p<0.01, d=0.42), and a budget-matched non-MBTI control shows that this gain is not explained by additional computation. Extended matched-role controls on all 50 evaluation tasks indicate that most of the improvement comes from rich, structured role priors combined with verification-oriented coordination rather than the MBTI taxonomy itself, and the diverse configuration is not significantly better than the strongest homogeneous one. We further present DyCo-Efficient, which removes dynamic teaming and preserves most of the observed quality at roughly one thirteenth of the token cost, positioning Full DyCo as an optional high-budget configuration.

Implicit Discourse Relation Recognition (IDRR) is an important task to classify the discourse relation sense between argument pairs without an explicit connective. Recently, prompt learning methods have demonstrated success in IDRR. However, prior work primarily transform IDRR into a connective-cloze task based on the masked language model (MLM), which limits the predicted connective to one single token. Also, they fail to fully exploit critical semantic features shared among various forms of templates. In this paper, we propose NCPrompt, an NSP-based prompt learning and Contrastive learning method for IDRR. Specifically, we transform the IDRR task into a next sentence prediction (NSP) task, which can allow various-length answer connectives and enlarge the construction of the verbalizer for prompt-learning methods. Also, we notice that various prompt templates naturally constitute positive samples applied for self-supervised contrastive learning. And the usage of NSP naturally creates hard negative samples by introducing different candidate connectives between the same example. To our knowledge, we are the first to combine self-supervised contrastive learning with prompt learning to obtain high-quality semantic representations. Experiments on the PDTB 3.0 corpus have demonstrated the effectiveness and superiority of our model.

Prior studies use one attention mechanism to improve contextual semantic representation learning for implicit discourse relation recognition (IDRR). However, diverse relation senses may benefit from different attention mechanisms. We also argue that some linguistic relation in between two words can be further exploited for IDRR. This paper proposes a Multi-Attentive Neural Fusion (MANF) model to encode and fuse both semantic connection and linguistic evidence for IDRR. In MANF, we design a Dual Attention Network (DAN) to learn and fuse two kinds of attentive representation for arguments as its semantic connection. We also propose an Offset Matrix Network (OMN) to encode the linguistic relations of word-pairs as linguistic evidence. Our MANF model achieves the state-of-the-art results on the PDTB 3.0 corpus.

Recently, prompt-tuning has attracted growing interests in event argument extraction (EAE). However, the existing prompt-tuning methods have not achieved satisfactory performance due to the lack of consideration of entity information. In this paper, we propose a bi-directional iterative prompt-tuning method for EAE, where the EAE task is treated as a cloze-style task to take full advantage of entity information and pre-trained language models (PLMs). Furthermore, our method explores event argument interactions by introducing the argument roles of contextual entities into prompt construction. Since template and verbalizer are two crucial components in a cloze-style prompt, we propose to utilize the role label semantic knowledge to construct a semantic verbalizer and design three kind of templates for the EAE task. Experiments on the ACE 2005 English dataset with standard and low-resource settings show that the proposed method significantly outperforms the peer state-of-the-art methods.