This paper proposes the novel task of video generation conditioned on a SINGLE semantic label map, which provides good balance between flexibility and quality in process. Different from typical end-to-end approaches, model both scene content dynamics single step, we propose to decompose this difficult into two sub-problems. As current image methods do better than terms detail, synthesize high by only generating first frame. Then animate based its meaning obtain temporally coherent video,...

A taxonomy is a semantic hierarchy, consisting of concepts linked by is-a relations. While large number taxonomies have been constructed from human-compiled resources (e.g., Wikipedia), learning text corpora has received growing interest and essential for long-tailed domain-specific knowledge acquisition. In this paper, we overview recent advances on construction free texts, reorganizing relevant subtasks into complete framework. We also evaluation discuss challenges future research.

Taolin Zhang, Zerui Cai, Chengyu Wang, Minghui Qiu, Bite Yang, Xiaofeng He. Proceedings of the 59th Annual Meeting Association for Computational Linguistics and 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

Distant supervision assumes that any sentence containing the same entity pairs reflects identical relationships. Previous works of distantly supervised relation extraction (DSRE) task generally focus on sentence-level or bag-level de-noising techniques independently, neglecting explicit interaction with cross levels. In this paper, we propose a hierarchical contrastive learning Framework for Distantly Supervised (HiCLRE) to reduce noisy sentences, which integrate global structural...

Pre-trained Language Models (PLMs) have achieved remarkable performance for various language understanding tasks in IR systems, which require the fine-tuning process based on labeled training data. For low-resource scenarios, prompt-based learning PLMs exploits prompts as task guidance and turns downstream into masked problems effective few-shot fine-tuning. In most existing approaches, high of heavily relies handcrafted verbalizers, may limit application such approaches real-world...

Recently, researchers have applied the word-character lattice framework to integrated word information, which has become very popular for Chinese named entity recognition (NER). However, prior approaches fuse information by different variants of encoders such as Lattice LSTM or Flat-Lattice Transformer, but are still not data-efficient indeed fully grasp depth interaction cross-granularity and important from lexicon. In this paper, we go beyond typical structure propose a novel...

Fine-tuning large pre-trained language models on various downstream tasks with whole parameters is prohibitively expensive. Hence, Parameter-efficient fine-tuning has attracted attention that only optimizes a few task-specific the frozen model. In this work, we focus prefix tuning, which continuous vectors (i.e. pseudo tokens) inserted into Transformer layers. Based observation learned syntax and semantics representation varies lot at different layers, argue adaptive will be further tailored...

Two facile and diversified protocols for the chem‐selective synthesis of 1,2‐dihydropyridine‐fused coumarins pyridocoumarins have been developed. Readily accessible substrates, mild reaction conditions good functional group tolerance make these procedures attractive construction various dihydropyridine‐fused pyridocoumarins, which may enrich develop heterocyclic chemistry medicinal chemistry.

Recent studies have shown that prompts improve the performance of large pre-trained language models for few-shot text classification. Yet, it is unclear how prompting knowledge can be transferred across similar NLP tasks purpose mutual reinforcement. Based on continuous prompt embeddings, we propose TransPrompt, a transferable framework learning tasks. In employ multi-task meta-knowledge acquisition procedure to train meta-learner captures cross-task knowledge. Two de-biasing techniques are...

Haojie Pan, Chengyu Wang, Minghui Qiu, Yichang Zhang, Yaliang Li, Jun Huang. Proceedings of the 59th Annual Meeting Association for Computational Linguistics and 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

Few-shot Named Entity Recognition (NER) aims to identify named entities with very little annotated data. Previous methods solve this problem based on token-wise classification, which ignores the information of entity boundaries, and inevitably performance is affected by massive non-entity tokens. To end, we propose a seminal span-based prototypical network (SpanProto) that tackles few-shot NER via two-stage approach, including span extraction mention classification. In stage, transform...

Spoken Language Understanding (SLU) aims to interpret the meanings of human speeches in order support various human-machine interaction systems. A key technique for SLU is Automatic Speech Recognition (ASR), which transcribes speech signals into text contents. As output texts modern ASR systems unavoidably contain errors, mainstream models either trained or tested on transcribed by would not be sufficiently error robust. We present ARoBERT, an Robust BERT model, can fine-tuned solve a...

Prompt-based fine-tuning has boosted the performance of Pre-trained Language Models (PLMs) on few-shot text classification by employing task-specific prompts. Yet, PLMs are unfamiliar with prompt-style expressions during pre-training, which limits learning downstream tasks.It would be desirable if models can acquire some prompting knowledge before adapting to specific NLP tasks. We present Unified Prompt Tuning (UPT) framework, leading better for BERT-style explicitly capturing semantics...

Knowledge-Enhanced Pre-trained Language Models (KEPLMs) are pre-trained models with relation triples injecting from knowledge graphs to improve language understanding abilities.Experiments show that our model outperforms other KEPLMs significantly over zero-shot probing tasks and multiple knowledge-aware tasks. To guarantee effective injection, previous studies integrate encoders for representing retrieved graphs. The operations retrieval encoding bring significant computational burdens,...

The automatic construction of large-scale knowledge graphs has received much attention from both academia and industry in the past few years. Notable graph systems include Google Knowledge Graph, DBPedia, YAGO, NELL, Probase many others. organizes information a structured way by explicitly describing relations among entities. Since entity identification relation extraction are highly depending on language itself, data sources largely determine processed, extracted, ultimately how formed,...

Chengyu Wang, Minghui Qiu, Taolin Zhang, Tingting Liu, Lei Li, Jianing Ming Jun Huang, Wei Lin. Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing: System Demonstrations. 2022.

Image-text retrieval is a challenging cross-modal task that arouses much attention. While the traditional methods cannot break down barriers between different modalities, Vision-Language Pre-trained (VLP) models greatly improve image-text performance based on massive pairs. Nonetheless, VLP-based are still prone to produce results be aligned with entities. Recent efforts try fix this problem at pre-training stage, which not only expensive but also unpractical due unavailable of full...