Jointly Modeling WSD and SRL with Markov Logic

Proceedings of the 23rd International Conference on Computational Linguistics (Coling 2010), pages 161–169, Beijing, August 2010 Jointly Modeling WSD and SRL with Markov Logic Wanxiang Che and Ting Liu Research Center for Information Retrieval MOE-Microsoft Key Laboratory of Natural Language Processing and Speech School of Computer Science and Technology {car, tliu}@ir.hit.edu.cn Abstract Semantic role labeling (SRL) and word sense disambiguation (WSD) are two fun- damental tasks in natural language pro- cessing to find a sentence-level seman- tic representation. To date, they have mostly been modeled in isolation. How- ever, this approach neglects logical con- straints between them. We therefore ex- ploit some pipeline systems which verify the automatic all word sense disambigua- tion could help the semantic role label- ing and vice versa. We further propose a Markov logic model that jointly labels se- mantic roles and disambiguates all word senses. By evaluating our model on the OntoNotes 3.0 data, we show that this joint approach leads to a higher perfor- mance for word sense disambiguation and semantic role labeling than those pipeline approaches. 1 Introduction Semantic role labeling (SRL) and word sense dis- ambiguation (WSD) are two fundamental tasks in natural language processing to find a sentence- level semantic representation. Semantic role la- beling aims at identifying the relations between predicates in a sentence and their associated ar- guments. Word sense disambiguation is the pro- cess of identifying the correct meaning, or sense of a word in a given context. For example, for the sentence in Figure 1, we can find out that the predicate token “hitting” at position 3 has sense “cause to move by striking” and the sense label is “hit.01”. The argument headed by the token “cat” at position 1 with sense “feline mammal” (cat.01) is referring to the player (A0), and the argument headed by the token “ball” at position 5 with sense Figure 1: A sample of word sense disambiguation and semantic role labeling. “round object that is hit in games” (ball.01) is re- ferring to the game object (A1) being hit. Normally, semantic role labeling and word sense disambiguation are regarded as two inde- pendent tasks, i.e., the word sense information is rarely used in a semantic role labeling system and vice versa. A few researchers have used se- mantic roles to help the verb sense disambigua- tion (Dang and Palmer, 2005). More people used predicate senses in semantic role labeling (Hajiˇc et al., 2009; Surdeanu et al., 2008). However, both of the pipeline methods ignore possible dependen- cies between the word senses and semantic roles, and can result in the error propagation problem. The same problem also appears in other natural language processing tasks. In order to make different natural language pro- cessing tasks be able to help each other, jointly modeling methods become popular recently, such as joint Chinese word segmentation and part-of- speech tagging (Kruengkrai et al., 2009; Zhang and Clark, 2008; Jiang et al., 2008), joint lemma- tization and part-of-speech prediction (Toutanova and Cherry, 2009), joint morphological segmenta- tion …

Content truncated to 500 words