This paper explores the effectiveness of parallel stylometric document embeddings in solving the authorship attribution task by testing a novel approach on literary texts in 7 different languages, totaling in 7051 unique 10,000-token chunks from 700 PoS and lemma annotated documents. We used these documents to produce four document embedding models using Stylo R package (word-based, lemma-based, PoS-trigrams-based, and PoS-mask-based) and one document embedding model using mBERT for each of the seven languages. We created further derivations of these embeddings in the form of average, product, minimum, maximum, and $l^2$ norm of these document embedding matrices and tested them both including and excluding the mBERT-based document embeddings for each language. Finally, we trained several perceptrons on the portions of the dataset in order to procure adequate weights for a weighted combination approach. We tested standalone (two baselines) and composite embeddings for classification accuracy, precision, recall, weighted-average, and macro-averaged $F_1$-score, compared them with one another and have found that for each language most of our composition methods outperform the baselines (with a couple of methods outperforming all baselines for all languages), with or without mBERT inputs, which are found to have no significant positive impact on the results of our methods.

Parallel natural language processing systems were previously successfully tested on the tasks of part-of-speech tagging and authorship attribution through mini-language modeling, for which they achieved significantly better results than independent methods in the cases of seven European languages. The aim of this paper is to present the advantages of using composite language models in the processing and evaluation of texts written in arbitrary highly inflective and morphology-rich natural language, particularly Serbian. A perplexity-based dataset, the main asset for the methodology assessment, was created using a series of generative pre-trained transformers trained on different representations of the Serbian language corpus and a set of sentences classified into three groups (expert translations, corrupted translations, and machine translations). The paper describes a comparative analysis of calculated perplexities in order to measure the classification capability of different models on two binary classification tasks. In the course of the experiment, we tested three standalone language models (baseline) and two composite language models (which are based on perplexities outputted by all three standalone models). The presented results single out a complex stacked classifier using a multitude of features extracted from perplexity vectors as the optimal architecture of composite language models for both tasks.

In a setting where multiple automatic annotation approaches coexist and advance separately but none completely solve a specific problem, the key might be in their combination and integration. This paper outlines a scalable architecture for Part-of-Speech tagging using multiple standalone annotation systems as feature generators for a stacked classifier. It also explores automatic resource expansion via dataset augmentation and bidirectional training in order to increase the number of taggers and to maximize the impact of the composite system, which is especially viable for low-resource languages. We demonstrate the approach on a preannotated dataset for Serbian using nested cross-validation to test and compare standalone and composite taggers. Based on the results, we conclude that given a limited training dataset, there is a payoff from cutting a percentage of the initial training set and using it to fine-tune a machine-learning-based stacked classifier, especially if it is trained bidirectionally. Moreover, we found a measurable impact on the usage of multiple tagsets to scale-up the architecture further through transfer learning methods.

This study presents the Sentiment Analysis of the Serbian old novels from the 1840-1920 period, employing the Mistral Large Language Model (LLM) to pioneer zero and few-shot learning techniques. The main approach innovates by devising research prompts that include guidance text for zero-shot classification and examples for fewshot learning, enabling the LLM to classify sentiments into positive, negative, or objective categories. This methodology aims to streamline sentiment analysis by limiting responses, thereby enhancing classification precision. Python, along with the Hugging Face Transformers and LangChain libraries, serves as our technological backbone, facilitating the creation and refinement of research prompts tailored for sentence-level sentiment analysis. The results of sentiment analysis in both scenarios, zero-shot and few-shot, have indicated that the zero-shot approach outperforms, achieving an accuracy of 68.2%.