Effective classroom communication is critical for students' academic performance. This study investigates the semantic similarity of adjacent teacher-to-student, teacher-to-teacher, and student-to-student utterances using natural language pro-cessing (NLP) tools. It explores their relationship with quality ratings of Instructional Dialogue obtained from Classroom Atmosphere Scoring System (CLASS) observations and the quality ratings of Teacher¿½fs Use of Student Contribution from Mathematical Quality of Instruction (MQI) observations. Fo-cusing on the cohesiveness of classroom language, the study analyzes transcripts from elementary math classrooms, associ-ating each with corresponding CLASS scores. Linear regres-sion models identified that the semantic similarity between teacher-to-student utterances was a significant predictor of the CLASS and MQI scores. However, the models explain little variance in the observational scores. The study underscores the complexity of classroom discourse and proposes future analyses. Additionally, it explores the utility of NLP tools in measuring teacher practices, emphasizing the need for ad-vancements in audio and textual models trained on speech-to-text transcriptions for accurate spoken language analysis. The findings prompt reflection on the practical significance of ob-served associations and highlight the importance of consider-ing the evolving landscape of educational technology in sup-porting teacher practice.