This paper explores the semantic representation of verbs in computer systems and its impact on lexical selection in machine translation (MT). It argues that lexical selection must consider both the interpretation of the sentence and selection restrictions on verb arguments. A novel representation scheme is proposed, which defines verbs by concepts in different conceptual domains. This approach is compared to transfer-based MT and is seen as compatible with knowledge-based MT (KBMT) approaches. The paper presents examples and experimental results showing that this scheme allows correct lexical selection even without exact matches. The paper discusses the limitations of direct transfer methods, which rely on exhaustive lists of verb pairs in bilingual dictionaries. These methods are inadequate for handling new verb usages, as they cannot account for the fluid nature of verb meanings. The paper presents a multi-domain approach, where verbs are defined by concepts in different conceptual domains. This allows for a more accurate representation of verb meanings and facilitates lexical selection. The paper also discusses the challenges of translating English verbs into Chinese serial verb compounds. It shows that Chinese requires a serial verb construction, which is not directly available in English. The paper presents a decision tree for translating English change-of-state verbs into Chinese, which considers both conceptual similarity and selection restrictions. The paper presents an implementation of the proposed approach, called UNICON, which uses a shared semantic domain for both English and Chinese verbs. The system uses a conceptual hierarchy to measure similarity between verbs and select the most appropriate target verb. The system is tested on a subset of the Brown corpus, showing improved performance compared to existing systems. The paper concludes that a multi-domain approach to verb representation is essential for accurate lexical selection in MT. This approach allows for a more nuanced understanding of verb meanings and facilitates the handling of new and metaphorical usages. The proposed method is seen as a promising alternative to traditional transfer-based approaches.This paper explores the semantic representation of verbs in computer systems and its impact on lexical selection in machine translation (MT). It argues that lexical selection must consider both the interpretation of the sentence and selection restrictions on verb arguments. A novel representation scheme is proposed, which defines verbs by concepts in different conceptual domains. This approach is compared to transfer-based MT and is seen as compatible with knowledge-based MT (KBMT) approaches. The paper presents examples and experimental results showing that this scheme allows correct lexical selection even without exact matches. The paper discusses the limitations of direct transfer methods, which rely on exhaustive lists of verb pairs in bilingual dictionaries. These methods are inadequate for handling new verb usages, as they cannot account for the fluid nature of verb meanings. The paper presents a multi-domain approach, where verbs are defined by concepts in different conceptual domains. This allows for a more accurate representation of verb meanings and facilitates lexical selection. The paper also discusses the challenges of translating English verbs into Chinese serial verb compounds. It shows that Chinese requires a serial verb construction, which is not directly available in English. The paper presents a decision tree for translating English change-of-state verbs into Chinese, which considers both conceptual similarity and selection restrictions. The paper presents an implementation of the proposed approach, called UNICON, which uses a shared semantic domain for both English and Chinese verbs. The system uses a conceptual hierarchy to measure similarity between verbs and select the most appropriate target verb. The system is tested on a subset of the Brown corpus, showing improved performance compared to existing systems. The paper concludes that a multi-domain approach to verb representation is essential for accurate lexical selection in MT. This approach allows for a more nuanced understanding of verb meanings and facilitates the handling of new and metaphorical usages. The proposed method is seen as a promising alternative to traditional transfer-based approaches.