Negation Word And Sentence Detection Towards Specific Prefix And Suffix

As research on semantic representation of text deepens,dealing with more complex language phenomena has become an important focus for researchers in the field of natural language processing.One of the key concerns is how to help language models better understand complex text information such as negation.Negation identification can be widely applied to various task scenarios,such as machine translation,sentiment analysis,and intelligent question answering.This dissertation analyzes the corpus and related experimental results,and finds that existing models are prone to errors in identifying negation word with specific prefix and suffix1(such as "im-" and "un-" in English negation words like "impossible" and "unhappy").Based on this,the dissertation proposes a task focused on negation word and sentence detection towards specific prefix and suffix,which aims to improve the ability of model to recognize affix negation and understand corresponding negation text information.Specifically,this dissertation conducts research in the following areas:Firstly,for the challenge of existing models encountering difficulties in accurately recognizing negation affixes,this dissertation constructs a corpus of negation words and non-negation words towards specific prefix and suffix.Two identification methods for specific negation prefix and suffix were designed based on this corpus:one based on single-word information and the other one based on bilingual information.The method based on single-word information for identifying negation word with specific prefix and suffix primarily focuses on extracting intraword information,and is designed to recognize multiple-character,individual word,and phrases.On the other hand,the method based on bilingual information combines prompt learning and joint learning to extract understanding information for the same word in different language models,thereby optimizing the prediction results of model.Experimental results demonstrate that both exploration methods can successfully improve the performance of identifying negation word with specific prefix and suffix,thus validating the effectiveness of the proposed methods.Secondly,this paper focus on the challenge of insufficient domain-specific corpus with negation words containing specific prefix and suffix,this dissertation proposes a few-shot learning method based on input-augmentation prompt and a pretrained token-replaced detection model.This approach consists of three steps.Firstly,a novel input expansion template is designed based on the negation relationship between the prefix/suffix and the remaining portion of the word,which is employed to enhance the input information.Secondly,a task model framework is developed by integrating a token-replaced detection model and selecting appropriate label descriptors.Finally,the label descriptor outputs the word information and classification recognition is based on this output.Experimental evaluation demonstrates the impressive effectiveness of this approach to recognize specific prefix and suffix negation sense in words over the other baselines.Finally,in view of the challenge of facilitating existing models’ comprehension of negation information,this dissertation proposes a negation sentence recognition method based on negation prefix and suffix information,starting from the task of negation sentence recognition.This approach comprises three main steps.Firstly,an auxiliary task is performed using a pre-trained model to identify the negation information present in the negation prefix and suffix of the sentence.Secondly,a main task framework for recognizing negation sentence is developed using the same pretrained model.Finally,the representation of the negation prefix and suffix recognition information obtained from the auxiliary task is integrated into the entire sentence feature representation hidden layer of the main task,and the resulting fused representation is utilized for negation sentence recognition.Experimental results demonstrate that this method can effectively improve the performance of negation sentence recognition.In summary,this dissertation addresses the task of identifying negation word with specific prefix and suffix by constructing manually annotated samples and conducting research on supervised learning and few-shot learning methods.Furthermore,this dissertation applies the task of identifying negation word with specific prefix and suffix to the task of negation sentence recognition,effectively improving the ability of existing model to recognize negation information.