Taesuk Hong, Youbin Ahn, Dongkyu Lee, Joongbo Shin, Seungpil Won, Janghoon Han, Stanley Jungkyu Choi, Jungyun Seo. Proceedings of the 25th Annual Meeting of the Special Interest Group on Discourse and Dialogue. 2024.

Janghoon Han, Taesuk Hong, Byoungjae Kim, Youngjoong Ko, Jungyun Seo. Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2021.

Recently, chatbot research for conversation has been actively conducted in the field of natural language processing. Among the dialogue systems, a retrieval-based system has emerged as an important one. Because, it showed better performance than the response generation models and it actually seemed applicable. The response selection task is used to develop a retrieval-based system. Thus, we did research to improve the performance of retrieval-based dialogue systems for response selection. We used Google's bidirectional encoder representations from transformers (BERT) natural-language processor with the robustly optimized BERT pre-training model (RoBERTa) as the base model to improve performance. The Recall@N of our model was improved by 1.7–4.2% using two methods: First, instead of using a simple feed-forward network at the end of the model, we employed a randomly wired neural network that contained multiple wiring paths, thus achieving better performance than that of feed-forward networks. The experimental results of this study demonstrated that feed-forward networks can be replaced by randomly wired neural networks. Second, we calculated the correlation between dialog history (i.e., context) and the last utterance of the context, reflecting this correlation as an attention network that yielded an answer prediction. Using first and second method achieved higher performance than using only the first method in dialog response selection tasks.

As research on utilizing human knowledge in natural language processing has attracted considerable attention in recent years, knowledge graph (KG) completion has come into the spotlight.Recently, a new knowledge graph completion method using a pre-trained language model, such as KG-BERT, was presented and showed high performance.However, its scores in ranking metrics such as Hits@k are still behind state-of-the-art models.We claim that there are two main reasons: 1) failure in sufficiently learning relational information in knowledge graphs, and 2) difficulty in picking out the correct answer from lexically similar candidates.In this paper, we propose an effective multi-task learning method to overcome the limitations of previous works.By combining relation prediction and relevance ranking tasks with our target link prediction, the proposed model can learn more relational properties in KGs and properly perform even when lexical similarity occurs.Experimental results show that we not only largely improve the ranking performances compared to KG-BERT but also achieve the state-of-the-art performances in Mean Rank and Hits@10 on the WN18RR dataset.

Korean named-entity recognition (NER) systems have been developed mainly on the morphological-level, and they are commonly based on a pipeline framework that identifies named-entities (NEs) following the morphological analysis. However, this framework can mean that the performance of NER systems is degraded, because errors from the morphological analysis propagate into NER systems. This paper proposes a novel syllable-level NER system, which does not require a morphological analysis and can achieve a similar or better performance compared with the morphological-level NER systems. In addition, because the proposed system does not require a morphological analysis step, its processing speed is about 1.9 times faster than those of the previous morphological-level NER systems.

Speech act classification is an essential part of a dialogue system because it is very important to catch user's intention. The previous approaches on speech act classification were focused on obtaining high performances in a single-domain, but they did not deal with a feature interference problem that frequently rises in a multi-domain. In this paper, we propose a two-step system for speech act classification in a multi-domain. In a first step, the proposed system detects a dialogue domain associated with user's utterance. In the second step, the proposed system determines the speech act of his/her utterance based on the statistical information of the detected domain. Owing to this architecture, the proposed system show ed higher precision of 5.5% than the baseline system based on the mixed statistical information.

A (page or web) snippet is document excerpts allowing a user to understand if a document is indeed relevant without accessing it. This paper proposes an effective snippet generation method. The pseudo relevance feedback technique and text summarization techniques are applied to salient sentences extraction for generating good quality snippets. In the experimental results, the proposed method showed much better performance than other methods including Google and Naver.

A wide range of supervised learning algorithms has been applied to Text Categorization. However, the supervised learning approaches have some problems. One of them is that they require a large, often prohibitive, number of labeled training documents for accurate learning. Generally, acquiring class labels for training data is costly, while gathering a large quantity of unlabeled data is cheap. We here propose a new automatic text categorization method for learning from only unlabeled data using a bootstrapping framework and a feature projection technique. From results of our experiments, our method showed reasonably comparable performance compared with a supervised method. If our method is used in a text categorization task, building text categorization systems will become significantly faster and less expensive.

A speech act is a linguistic action intended by a speaker. Speech act classification is an essential part of a dialogue understanding system because the speech act of an utterance is closely tied with the user's intention in the utterance. We propose a neural network model for Korean speech act classification. In addition, we propose a method that extracts morphological features from surface utterances and selects effective ones among the morphological features. Using the feature selection method, the proposed neural network can partially increase precision and decrease training time. In the experiment, the proposed neural network showed better results than other models using comparatively high-level linguistic features. Based on the experimental result, we believe that the proposed neural network model is suitable for real field applications because it is easy to expand the neural network model into other domains. Moreover, we found that neural networks can be useful in speech act classification if we can convert surface sentences into vectors with fixed dimensions by using an effective feature selection method.

This study aims to improve the performance of identifying grammatical functions between an adnoun clause and a noun phrase in Korean. The key task is to determine the relation between the two constituents in terms of such functional categories as subject, object, adverbial and appositive. The problem is mainly caused by the fact that functional morphemes, which are considered to be crucial for identifying the relation, are omitted in the noun phrases. To tackle this problem, we propose to employ the Support Vector Machines (SVM) in determining the grammatical functions. Through an experiment with a tagged corpus for training SVMs, we found the proposed model to be more useful than both the Maximum Entropy Model (MEM) and the backed-off model.

Referring expressions in multimodal dialogues have different aspects compared to those in language-only dialogues. They often refer to the items signified by either a gesture or visual means. In this article we classify referring expressions into two types (i.e., a deictic reference and an anaphoric reference), and propose two general methods to resolve these referring expressions. One method is a simple mapping algorithm that can find items referred with/without pointing gestures on a screen. The other is the centering algorithm with a dual cache model, to which Walker's centering algorithm is extended for a multimodal dialogue system. The extended algorithm is appropriate for resolving various anaphoric references in a multimodal dialogue. In the experiments, the proposed system correctly resolved 376 out of 405 referring expressions in 40 dialogues (0.54 referring expressions per utterance) showing 92.84 percent correctness.

We propose a statistical dialogue analysis model to determine discourse structures as well as speech acts using maximum entropy model. The model can automatically acquire probabilistic discourse knowledge from a discourse tagged corpus to resolve ambiguities. We propose the idea of tagging discourse segment boundaries to represent the structural information of discourse. Using this representation we can effectively combine speech act analysis and discourse structure analysis in one framework.

This paper presents a system for automatic verb sense disambiguation using a small corpus and a Machine-Readable Dictionary (MRD) in Korean. The system learns a set of typical uses listed in the MRD usage examples for each of the senses of a polysemous verb in the MRD definitions using verb-object co-occurrences acquired from the corpus. This paper concentrates on the problem of data sparseness in two ways. First, by extending word similarity measures from direct co-occurrences to co-occurrences of co-occurring words, we compute the word similarities using non co-occurring words but co-occurring clusters. Secondly, we acquire IS-A relations of nouns from the MRD definitions. It is possible to roughly cluster the nouns by the identification of the IS-A relationship. Using these methods, two words may be considered similar even if they do not share any word elements. Experiments show that this method can learn from a very small training corpus, achieving over an 86% correct disambiguation performance without any restriction on a word's senses.

An active database system reacts to a set of external events such as a timer interrupt or access to a particular object in the database. A trigger is a general mechanism for active data management, both in the context of a centralized system or a distributed system (including that of autonomous and cooperating agents). It consists of three parts; event specification, integrity constraint specification, and action specification. The event specification in a trigger is a set of events which will cause the condition in the constraint specification to be checked. If the condition is true, the actions in the action specification will be initiated. In this paper, we develop a framework for supporting triggers in object-oriented database systems. The framework consists of a categorization for each of the three components of a trigger. The framework is first cast in the context of the relational model of data, and is then extended to account for object-oriented concepts that constitute an object-oriented data model, including nested objects, methods, and inheritance hierarchy.

In this paper, we present a computational method for transforming a syntactic graph, which represents all syntactic interpretations of a sentence, into a semantic graph which filters out certain interpretations, but also incorporates any remaining ambiguities. We argue that the resulting ambiguous graph, supported by an exclusion matrix, is a useful data structure for question answering and other semantic processing. Our research is based on the principle that ambiguity is an inherent aspect of natural language communication.

This report discusses the implementation of a knowledge base for a library information system. It is done using a typed logic programming language—LOGIN—where type inheritance is built in. The knowledge base is structured in a hierarchical taxonomy of library object classes where each class is represented in a FRAME style knowledge structure and inherits the properties of its parents, and where infrastructural inference rules have been established through typed Horn clauses. Also in this document, some programming techniques aimed at using the power of inheritance as taxonomic inference are discussed.