Create accountLogin

Recent searches

No recent searches

Search options

Only available when logged in.
livellosegreto.it is one of the many independent Mastodon servers you can use to participate in the fediverse.
Livello Segreto è il social etico che ha rispetto di te e del tuo tempo.

Administered by:

Server stats:

1.2K
active users

livellosegreto.it: AboutProfiles directoryPrivacy policy

Mastodon: AboutGet the appKeyboard shortcutsView source codev4.3.7

#corpus

0 posts0 participants0 posts today
Public
h o ʍ l e t t

→ Une #intelligence #artificielle libre est-elle possible ?
linuxfr.org/news/une-intellige

« Posons-nous un instant la question : qu’est-ce que le #code #source d’un réseau de #neurones ? […] La #GPL fournit une définition : le code source est la forme de l’œuvre privilégiée pour effectuer des #modifications. Dans cette acception, le code source d’un réseau de neurones serait l’#algorithme d’entraînement, le réseau de neurones de départ et le #corpus sur lequel le réseau a été entraîné »

linuxfr.orgUne intelligence artificielle libre est-elle possible ? - LinuxFr.orgL’actualité du logiciel libre et des sujets voisins (DIY, Open Hardware, Open Data, les Communs, etc.), sur un site francophone contributif géré par une équipe bénévole par et pour des libristes enthousiastes
#libre
Public *
Mathieu Goux

#ESR #Recrutement #ingenieur #TAL #Corpus #JeRecrute

Le CRISCO, mon laboratoire de recherche à Caen, embauche un.e ingénieur.e d'études pour une durée de 7 mois, à partir de décembre ou de janvier prochain ! Vous trouverez les détails ci-dessous. Profil TAL / Linguistique de Corpus.

Merci de diffuser largement !

Poke @f_moncomble @MarCandea @fay @LeoVarnet @nasiviru

crisco.unicaen.fr/le-crisco-re

CRISCO - Centre de recherches inter-langues sur la signification en contexte - UR 4255 · Le CRISCO recherche un.e ingénieur.e d'études (décembre 2024 ou janvier 2025) · CRISCO - Centre de recherches inter-langues sur la signification en contexte - UR 4255Ingénieur.e d’études (CDD) en linguistique et TAL au CRISCO (Caen) Laboratoire CRISCO, Université de Caen, Normandie Septembre 2024 Dans le cadre du projet AUTOMATED (12/2023-11/2024), porté par Professeur Pierre Larrivée et financé par la Région Normandie (Réseau d’Intérêts Normands) nous recherchons un.e ingénieur.e d’études pour la période de sept mois.
Public
Boiling Steam

Building a Large Japanese Web Corpus for Large Language Models: arxiv.org/abs/2404.17733 #llm #japanese #dataset #corpus #training

arXiv.orgBuilding a Large Japanese Web Corpus for Large Language ModelsOpen Japanese large language models (LLMs) have been trained on the Japanese portions of corpora such as CC-100, mC4, and OSCAR. However, these corpora were not created for the quality of Japanese texts. This study builds a large Japanese web corpus by extracting and refining text from the Common Crawl archive (21 snapshots of approximately 63.4 billion pages crawled between 2020 and 2023). This corpus consists of approximately 312.1 billion characters (approximately 173 million pages), which is the largest of all available training corpora for Japanese LLMs, surpassing CC-100 (approximately 25.8 billion characters), mC4 (approximately 239.7 billion characters) and OSCAR 23.10 (approximately 74 billion characters). To confirm the quality of the corpus, we performed continual pre-training on Llama 2 7B, 13B, 70B, Mistral 7B v0.1, and Mixtral 8x7B Instruct as base LLMs and gained consistent (6.6-8.1 points) improvements on Japanese benchmark datasets. We also demonstrate that the improvement on Llama 2 13B brought from the presented corpus was the largest among those from other existing corpora.
Public *
Sascha Wolfer

🚨 New preprint 🚨
"Does corpus size influence normalised frequencies?"

doi.org/10.31219/osf.io/tr8de

It may sound like a silly question, but many #corpus linguistic measures are influenced by corpus size. So we asked ourselves: Does this also hold for normalised #frequencies, a measure that is meant to correct raw frequencies for the size of the underlying corpus?

We approached this by checking the association between lists of normalised frequencies for samples of different sizes.

Figure 1 of the preprint "Does corpus size influence normalised frequencies?" showing the relationship between sample sizes and Tau B. As samples get larger, associations between frequency lists increase. This holds for all five languages under investigation (English, Finnish, French, German, Vietnamese).
#linguistics
Public *
editora con carrito

Os cuento un poco de intrahistoria del #Corpus Lingüístico del #Eroski en #catalán, #gallego, #euskera y #castellano con los textos de la #RevistaConsumer. Gracias a @fracofavor por compartir el Corpus en este toot:

mastodon.gal/@fracofavor/11188

Cito a mi compañero Iker Merchán, que entonces era el director de la web Consumer.es:

«Un día apareció alguien contando que había sacado a lo bruto los contenidos de la web y había hecho el corpus. Llevaba más de un año con el proyecto...»

+

mastodon.galhélio (-e) 🐗 (@fracofavor@mastodon.gal)Curiosidade sobre Eroski: na súa web levan anos traducindo os seus artigos ao euskera, ao galego e ao catalán e nalgún momento alguén tivo a idea de facer un corpus con todos eses textos, así que agora Eroski ten O ÚNICO CORPUS PARALELO DAS CATRO LINGUAS MÁIS FALADAS DO ESTADO https://corpus.consumer.es/corpus/aurkezpena
Replied in thread
Public
Hugo

@theklan part of of those frequency lists are based on the bible. For each of the 1001 #languages #Unilex scrapped various *open* online resources: wikipedias, bibles translations, wordpress blogs. It is used in all our android phones for text autocomplete. Sometimes, only the bible was available so the frequency list reflects it. I would prefer larger, more diverse raw data but they did with what they got. I do not know any better hyperlingual open #corpus. cc @alexture
github.com/unicode-org/unilex

GitHubGitHub - unicode-org/unilex: Lexical data at UnicodeLexical data at Unicode. Contribute to unicode-org/unilex development by creating an account on GitHub.
Public
Ártemis López

New profile, new #introduction

I'm Ártemis, aka #queerterpreter. This is my academic/languagery account, or that's what I'm going for. I'm a #sociolinguist and a PhD Candidate researching non-binary Spanish. I'm also an ES<>EN #translator and #interpreter. My diss leans heavily on #corpus linguistics, specifically corpora scraped from the birdsite. The data is safe but I'm in mourning.

My more personal acct is at @queerterpreter@queer.party, feel free to follow both or just one (or none) :Schwerified:

SearchLive feeds
About