Search

This paper explores cost-effective alternatives for resource-constrained environments in the context of language models by investigating methods such as quantization and CPUbased model implementations. The study addresses the computational efficiency of language models during inference and the development of infrastructure for text document processing. The paper discusses related technologies, the CLARIN-PL infrastructure architecture, and implementations of small and large language models. The emphasis is on model formats, data precision, and runtime environments (GPU and CPU). It identifies optimal solutions through extensive experimentation. In addition, the paper advocates for a more comprehensive performance evaluation approach. Instead of reporting only average token throughput, it suggests considering the curve’s shape, which can vary from constant to monotonically increasing or decreasing functions. Evaluating token throughput at various curve points, especially for different output token counts, provides a more informative perspective.

This paper explores cost-effective alternatives for resource-constrained environments in the context of language models by investigating methods such as quantization and CPUbased model implementations. The study addresses the computational efficiency of language models during inference and the development of infrastructure for text document processing. The paper discusses related technologies, the CLARIN-PL infrastructure architecture, and implementations of small and large language models. The emphasis is on model formats, data precision, and runtime environments (GPU and CPU). It identifies optimal solutions through extensive experimentation. In addition, the paper advocates for a more comprehensive performance evaluation approach. Instead of reporting only average token throughput, it suggests considering the curve’s shape, which can vary from constant to monotonically increasing or decreasing functions. Evaluating token throughput at various curve points, especially for different output token counts, provides a more informative perspective. [ABSTRACT FROM AUTHOR]

У статті описано групу Clarin-PL, яка є польським представництвом Європейської дослідницької інфраструктури CLARIN ERIC. Представлено завдання та цілі інфраструктури CLARIN ERIC та групи Clarin-PL. Як приклади подано окремі мовні засоби та ресурси, розроблені групою Clarin-PL. Особливу увагу присвячено тим багатомовним ресурсам, головна роль у побудові яких належить команді Інституту славістики Польської академії наук (ІС ПАН), зокрема це два розширені багатомовні корпуси сучасних текстів Polish-Lithuanian Parallel Corpus «2» i Polish-Ukrainian Parallel Corpus. Схарактеризовано провідну роль ІС ПАН у побудові групою Clarin-PL багатомовних корпусів. Окреслено нові, вже розпочаті та заплановані завдання, пов’язані з побудовою багатомовних ресурсів Clarin-PL.

On New Manually Aligned and Tagged Bilingual Parallel Corpora and Their Applications This article is devoted to the manually aligned and tagged bilingual parallel CLARIN-PL-BIZ corpora of the Baltic and Slavic languages which are currently being developed. The study discusses the essential features of these corpora that make their applications go far beyond typical corpus analysis. Applications of these corpora include the design of cross-language models for the development of machine translation and artificial intelligence. The article also draws attention to the high potential of these resources as a model training base for testing natural language processing tools. O nowych ręcznie zrównoleglonych i znakowanych dwujęzycznych korpusach równoległych oraz ich zastosowaniach W artykule autor opisuje obecnie powstające ręcznie zrównoleglone i znakowane dwujęzyczne korpusy równoległe CLARIN-PL-BIZ języków bałtyckich i słowiańskich. Omawia wyróżniające cechy tych korpusów, które sprawią, że zastosowania tych korpusów znacznie wykroczą poza typowe analizy korpusowe. Wśród zastosowań tych korpusów autor wymienia definiowanie modeli międzyjęzykowych na rzecz rozwoju przekładu maszynowego i rozwoju sztucznej inteligencji. Zwraca również uwagę na wysoki potencjał tych zasobów jako wzorcowej bazy treningowej do testowania narzędzi przetwarzania języka naturalnego.

Copyright of Acta Baltico-Slavica is the property of Polish Academy of Sciences, Institute of Slavic Studies and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Background. This article describes the Clarin-PL consortium, which represents the Polish contribution to the CLARIN ERIC European research infrastructure. The aims and tasks of both CLARIN ERIC and Clarin-PL are presented. Purpose. Presentation of the achievements of researchers from the Institute of Slavic Studies of the Polish Academy of Sciences in the field of creating and developing multilingual corpora, including tagging and parallelizing texts. Methods. The team of the Institute of Slavic Studies of the Polish Academy of Sciences adopted common assumptions for the construction of multilingual corpora of the Slavic and Baltic languages. Namely, the corpora contains selected modern texts that represent all functional styles to the greatest extent. Mutual translations are preferred. Results. The article presents a description of selected multilingual resources created by Clarin-PL and made available online via the Clarin-PL website, which a team from the Institute of Slavic Studies of the Polish Academy of Sciences (IS PAN) played a key role in creating. These resources are two expanded multilingual corpora of parallel contemporary texts: the Polish-Lithuanian Parallel Corpus 2 and the Polish-Ukrainian Parallel Corpus. Due to the fact that IS PAN played a leading role in the development of the multilingual corpora in the Clarin-PL consortium, it was decided to present an outline of corpus linguistics development in IS PAN. Discussion. The European Clarin-ERIC infrastructure is steadily developing. Scattered resources (previously created and newly emerging) are combined into a coherent whole. The Polish Consortium Clarin-PL primarily creates and develops resources and tools for the Polish language. The aim of these works is to provide the recipient with the highest possible quality of corpora compatible with constantly changing standards, allowing for the versatile use of tools. [ABSTRACT FROM AUTHOR]

An open stylometric system based on multilevel text analysis Stylometric techniques are usually applied to a limited number of typical tasks, such as authorship attribution, genre analysis, or gender studies. However, they could be applied to several tasks beyond this canonical set, if only stylometric tools were more accessible to users from different areas of the humanities and social sciences. This paper presents a general idea, followed by a fully functional prototype of an open stylometric system that facilitates its wide use through to two aspects: technical and research flexibility. The system relies on a server installation combined with a web-based user interface. This frees the user from the necessity of installing any additional software. At the same time, the system offers a variety of ways in which the input texts can be analysed: they include not only the usual lexical level, but also deep-level linguistic features. This enables a range of possible applications, from typical stylometric tasks to the semantic analysis of text documents. The internal architecture of the system relies on several well-known software packages: a collection of language tools (for text pre-processing), Stylo (for stylometric analysis) and Cluto (for text clustering). The paper presents: (1) The idea behind the system from the user’s perspective. (2) The architecture of the system, with a focus on data processing. (3) Features for text description. (4) The use of analytical systems such as Stylo and Cluto. The presentation is illustrated with example applications. Otwarty system stylometryczny wykorzystujący wielopoziomową analizę języka Zastosowania metod stylometrycznych na ogół ograniczają się do kilku typowych problemów badawczych, takich jak atrybucja autorska, styl gatunków literackich czy studia nad zróżnicowaniem stylistycznym kobiet i mężczyzn. Z pewnością dałoby się je z powodzeniem zastosować również do wielu innych problemów klasyfikacji tekstów, gdyby tylko owe metody oraz odpowiednie narzędzia były bardziej dostępne dla uczonych reprezentujących różne dyscypliny nauk humanistycznych i społecznych. Artykuł niniejszy omawia założenia teoretyczne oraz w pełni funkcjonalny prototyp otwartego systemu stylometrycznego, którego szerokie zastosowanie umożliwią dwie jego cechy: elastyczność techniczna oraz dostosowywalność do różnych pytań badawczych. System opiera się na instalacji serwerowej sprzęgniętej z sieciowym interfejsem użytkownika. Uwalnia to użytkownika od konieczności instalowania jakichkolwiek dodatkowych programów. Jednocześnie system oferuje wiele sposobów analizowania tekstów nie tylko na poziomie leksykalnym, lecz także poprzez cechy językowe niskiego poziomu. Daje to możliwość stosowania systemu na wiele różnych sposobów, od typowych testów stylometrycznych do analizy semantycznej dokumentów. Wewnętrzna architektura systemu składa się z wielu elementów znanych ze swej funkcjonalności, w tym z pakietu Stylo przeznaczonego do analiz stylometrycznych oraz pakietu Cluto służącego do zaawansowanej analizy skupień. Artykuł omawia: (1) Koncepcję całego systemu, postrzeganą z punktu widzenia użytkownika, (2) Architekturę systemu oraz jego elementy odpowiedzialne za przetwarzanie tekstu, (3) Cechy językowe służące do opisu dokumentów, (4) Zastosowanie modułów analizy danych, takich jak Stylo czy Cluto. W artykule zostały też przedstawione przykładowe zastosowania systemu.

Stylometric techniques are usually applied to a limited number of typical tasks, such as authorship attribution, genre analysis, or gender studies. However, they could be applied to several tasks beyond this canonical set, if only stylometric tools were more accessible to users from different areas of the humanities and social sciences. This paper presents a general idea, followed by a fully functional prototype of an open stylometric system that facilitates its wide use through to two aspects: technical and research flexibility. The system relies on a server installation combined with a web-based user interface. This frees the user from the necessity of installing any additional software. At the same time, the system offers a variety of ways in which the input texts can be analysed: they include not only the usual lexical level, but also deep-level linguistic features. This enables a range of possible applications, from typical stylometric tasks to the semantic analysis of text documents. The internal architecture of the system relies on several well-known software packages: a collection of language tools (for text pre-processing), Stylo (for stylometric analysis) and Cluto (for text clustering). The paper presents: (1) The idea behind the system from the user's perspective. (2) The architecture of the system, with a focus on data processing. (3) Features for text description. (4) The use of analytical systems such as Stylo and Cluto. The presentation is illustrated with example applications. [ABSTRACT FROM AUTHOR]