<p>Borehole data interpretation involves the comprehensive utilization of multi-source geological information, serving as a critical issue in geoscience information fusion and knowledge representation. Although lithology classification and stratigraphic unit division are correlated tasks, traditional methods predominantly process them independently and rely on single-modal data, making it difficult to fully utilize multi-source information. To address these issues, a Multimodal Multi-task Gated Fusion Network (M3GF-Net) integrating geological semantics and structural attributes is proposed for borehole interval interpretation. Based on the hard parameter sharing mechanism, this method achieves collaborative modeling for both tasks. A pre-trained language model is used to extract semantic features from geological texts, and Bidirectional Long Short-Term Memory (BiLSTM) is further employed to enhance sequential information representation; simultaneously, independent representation learning is conducted for layer depth and layer thickness. By introducing a Gated Multimodal Unit (GMU), the adaptive fusion of multimodal information is achieved to construct a unified representation, which is respectively applied to the lithology classification and stratigraphic unit division tasks. Experimental results on the evaluated Ordos Basin borehole dataset show that M3GF-Net achieves an average Macro-F1 score of 81.08% across the two interpretation tasks. These results indicate that the unified use of geological descriptions and structural attributes can improve borehole record interpretation in the evaluated dataset.</p>