Context Limitations Make Neural Language Models More Human-Like