We examine how Chinese and American oil companies use the gain- and loss-framed BUILDING source domain to legitimize their business in Corporate Social Responsibility (CSR) reports. Gain and loss frames can create legitimacy because they can ethically position an issue. We will focus on oil companies in China and the U.S. because different socio-cultural contexts in these two countries can potentially lead to different legitimation strategies in CSR reports, which can shed light on differences in Chinese and American CSR. All of the oil companies in our data are on the Fortune 500 list (2020). The results showed that Chinese oil companies used BUILDING metaphors more frequently than American oil companies. The most frequent keyword in Chinese CSRs “build” highlights environmental achievements in compliance with governments’ policies. American CSRs often used the metaphorical verb “support” to show their alignment with environmental policies and the interests of different stakeholders. The BUILDING source domain was used more often as gain frames in both Chinese and American CSR reports to show how oil companies create benefits for different stakeholders.

Metaphors are ubiquitous in natural language, and detecting them requires contextual reasoning about whether a semantic incongruence actually exists. Most existing work addresses this problem using pre-trained contextualized models. Despite their success, these models require a large amount of labeled data and are not linguistically-based. In this paper, we proposed a ContrAstive pre-Trained modEl (CATE) for metaphor detection with semi-supervised learning. Our model first uses a pre-trained model to obtain a contextual representation of target words and employs a contrastive objective to promote an increased distance between target words’ literal and metaphorical senses based on linguistic theories. Furthermore, we propose a simple strategy to collect large-scale candidate instances from the general corpus and generalize the model via self-training. Extensive experiments show that CATE achieves better performance against state-of-the-art baselines on several benchmark datasets.