Detection of text boundaries between human written and machine generated text portions in partially machine generated texts. Achieved 30% improvement over existing proprietary systems using less than 1% as much training data. The proposed approach achives similar performance evn on very short texts (2-3 sentences)