Content uploaded by Mika Hämäläinen
Author content
All content in this area was uploaded by Mika Hämäläinen on Sep 22, 2021
Content may be subject to copyright.
Seventh international workshop on computational linguistics of Uralic languages, pages 54–61
September 23–24, 2021. ©2021 Association for Computational Linguistics
https://doi.org/10.26615/978-1-954085-82-4_007
The Current State of Finnish NLP
Mika Hämäläinen
Faculty of Arts
University of Helsinki
and Rootroo Ltd
mika.hamalainen@helsinki.fi
Khalid Alnajjar
Faculty of Arts
University of Helsinki
and Rootroo Ltd
khalid.alnajjar@helsinki.fi
Abstract
There are a lot of tools and resources avail-
able for processing Finnish. In this pa-
per, we survey recent papers focusing on
Finnish NLP related to many dierent sub-
categories of NLP such as parsing, gen-
eration, semantics and speech. NLP re-
search is conducted in many dierent re-
search groups in Finland, and it is frequently
the case that NLP tools and models result-
ing from academic research are made avail-
able for others to use on platforms such as
Github.
Tiivistelmä
Suomen kielen koneelliseen käsittelyyn
on tarjolla paljon valmiita työkaluja
ja resursseja. Tässä artikkelissa
tarkastelemme viimeaikoina julkaistuja
tieteellisiä artikkeleita, joissa keski-
tytään suomen kielen kieliteknologiaan.
Tarkastelemme kieliteknologian eri alalu-
okkia, kuten jäsentämistä, tuottamista,
semantiikkaa ja puheetta. kieliteknologista
tutkimusta tehdään Suomessa monissa eri
tutkimusryhmissä, ja usein akateemisen
tutkimuksen tuloksena tuotetut kielite-
knologian työkalut ja mallit julkaistaan
muiden käytettäväksi esimerkiksi Githu-
bissa.
1 Introduction
There is no doubt that, within the Uralic language
family, Finnish is one of the most well-resourced
languages in terms of natural language processing
(NLP). This has, however, not always been the case.
Currently, NLP research conducted for Finnish has
started to fragment into research outputs of several
dierent research groups, and there is no survey pa-
per out there that would describe the current state of
Finnish NLP.
We hope that this survey paper claries the cur-
rent situation and makes it clearer for people work-
ing in the academia outside of Finnish universities
or in the industry and also for students. As it has
been discussed before (Hämäläinen,2021), Finnish
is certainly not a low-resourced language, and our
current survey further proves this point.
It is also important for researchers working on
other smaller Uralic languages to see what has been
done for Finnish in terms of NLP to see what the
possible and meaningful directions are for further
developing the resources needed. Especially since
Uralic language share the same feature of rich mor-
phology, which is something that commonly causes
problems for computers.
2 Finnish NLP
In this section, we present a survey on the current
state of Finnish NLP. We have tried to gather most
of the current research on the topic, but we are cer-
tain that there are some research out there we have
not been able to nd. We have categorized the sur-
veyed research outputs into parsing, generation, se-
mantics and speech.
2.1 Parsing
Starting from morphology, stemming and spell
checking Finnish is well supported in multiple com-
mercial applications such as Microsoft and Google
products. In the open-source world, low-level tasks
such as stemming and spell checking can be con-
ducted with Voikko¹.
Omor (Pirinen,2015)²is currently the most
well supported tool for morphological analysis (in-
¹https://voikko.puimula.org/
²https://github.com/ammie/omor
54
cluding lemmatization) and generation. It is an
FST (nite-state transducer) based tool developed
on HFST (Helsinki nite-state technology) (Lindén
et al.,2013) and it works together with constraint
grammar (CG) based disambiguators and syntac-
tic parsers available in the Giellatekno (Moshagen
et al.,2014) repositories³.
FinnPos⁴(Silfverberg et al.,2016) is another
morphological tagger and lemmatizer tool based on
CRF (conditional random eld). There have been
recently more data driven approaches focusing on
Finnish (Silfverberg and Hulden,2018).
While rule-based tradition has been strong in the
past⁵, there are several machine learning driven de-
pendency parsers for Finnish, such as the statistical
one⁶(Haverinen et al.,2014) and neural one⁷(Kan-
erva et al.,2018) by TurkuNLP.
Out of the aforementioned tools Omor (and the
CG disambigator) and the machine learning based
parsers are available to use through a Python pack-
age named UralicNLP⁸ ⁹ (Hämäläinen,2019).
As Finnish data is available in several multi-
lingual datasets, there are many multilingual ap-
proaches for parsing (Qi et al.,2020)¹⁰ (Honnibal
et al.,2020)¹¹ and morphology (Aharoni and Gold-
berg,2017;Nicolai and Yarowsky,2019;Silfver-
berg and Tyers,2019;Grönroos et al.,2020).
The fact that spoken Finnish is very dierent
to standard Finnish has drawn some attention in
the past (Jauhiainen,2001) and recently (Partanen
et al.,2019). The latter leading to a Python library
called Murre¹² for automatic normalization of di-
alectal Finnish.
Non-standard data has been an issue in digital
humanities (DH) projects (Mäkelä et al.,2020),
and lately there have been eorts in automati-
cally correcting OCR errors in existing histori-
cal datasets (Kettunen,2015;Drobac and Lindén,
2020;Drobac,2020;Duong et al.,2020).
Named entity recognition has also been under
study with FiNER¹³ and its recently released data
³https://github.com/giellalt/lang-n/tree/main/src/cg3
⁴https://github.com/mpsilfve/FinnPos
⁵See Pirinen,2019b for some comparison between rules
and neural networks
⁶https://turkunlp.org/Finnish-dep-parser/
⁷http://turkunlp.org/Turku-neural-parser-pipeline/
⁸https://github.com/mikahama/uralicNLP
⁹https://github.com/mikahama/uralicNLP/wiki/Dependency-
parsing
¹⁰https://stanfordnlp.github.io/stanza/
¹¹https://spacy.io/
¹²https://github.com/mikahama/murre
¹³https://github.com/Traubert/FiNer-
(Ruokolainen et al.,2019). There is also another re-
cent BERT (Devlin et al.,2019) based approach¹⁴
to the topic (Luoma et al.,2020).
There have been several approaches to language
detection including detection of Finnish from web
corpora (see Jauhiainen et al.,2021). Similarly, na-
tive Finnish has been automatically identied from
learner’s Finnish (Malmasi and Dras,2014).
In summary, parsing has been researched on dif-
ferent levels of language such as syntax, morphol-
ogy, POS and NER tagging, and lemmatization. It
has been mainly focusing on standard well-formed
Finnish, although there are methods for coping with
dialectal Finnish and OCR errors as well.
2.2 Generation
The lowest level of natural language generation is
surface realization (see Reiter,1994), and for that
there are tools such as Omor and Syntax Maker¹⁵
(Hämäläinen and Rueter,2018). The latter uses
Omor for morphological inection while it takes
care of higher level morphosyntax such as case gov-
ernment and agreement.
There is a strong computational creativity fo-
cus in Helsinki and it also shows in Finnish
NLG, as there are several poem generators such
as Keinoleino¹⁶ (Hämäläinen,2018b), Poeticus
(Toivanen et al.,2012) and others (Hämäläinen and
Alnajjar,2019a,b). There is also an interactive
poem generator tool called Runokone (Poem Ma-
chine)¹⁷ (Hämäläinen,2018c).
Recently there have been several approaches to
enhancing existing news headlines (Alnajjar et al.,
2019;Rämö and Leppänen,2021). And some ap-
proaches to generating entire news articles automat-
ically (Kanerva et al.,2019;Haapanen and Leppä-
nen,2020).
Paraphrase generation (Sjöblom et al.,2020) has
also become a researched topic with the availability
of monolingually aligned parallel corpora (Creutz,
2018). There is also an approach to converting stan-
dard Finnish text into dierent dialects (Hämäläi-
nen et al.,2020).
Finnish is a typical language for machine trans-
lation tasks and it is not uncommon to see it fea-
tured in several papers that deal with multiple lan-
guages. However, there are several papers that fo-
rules/blob/master/ner-readme.md
¹⁴https://turkunlp.org/n-ner.html
¹⁵https://github.com/mikahama/syntaxmaker
¹⁶https://github.com/mikahama/keinoleino
¹⁷http://runokone.cs.helsinki./
55
cus on Finnish in particular (Hurskainen and Tiede-
mann,2017;Hämäläinen and Alnajjar,2019c;Piri-
nen,2019a;Tiedemann et al.,2020).
There is also a recent approach to dialog genera-
tion in Finnish (Leino et al.,2020). Also non-native
language learner’s errors have been corrected suc-
cessfully automatically (Creutz and Sjöblom,2019).
To summarize the approaches, there are several
generators for poetry and news that benet from
the available surface realizers. Paraphrasing, di-
alect adaptation, dialog generation and learners’ er-
ror correction are domains with some research with
potential for new discoveries in the future. Ma-
chine translation gets frequently attention from dif-
ferent researchers. There are several more NLG
tasks (see Gatt and Krahmer 2018) that have not
been researched at all in Finnish, which means that
there is a lot of room for more research on this topic.
2.3 Semantics
Vector representations of meaning have become
common place in NLP and Finnish is no excep-
tion with the availability of pretrained word2vec¹⁸
¹⁹ (Laippala and Ginter,2014;Kutuzov et al.,2017)
and fastText²⁰ (Bojanowski et al.,2017) models.
BERT models have also become available as part
of the multilingual BERT model²¹ (Devlin et al.,
2019) or trained separately for Finnish²² ²³ (Kutuzov
et al.,2017;Virtanen et al.,2019). Even Elmo mod-
els have been made available for Finnish²⁴ (Ulčar
and Robnik-Šikonja,2020).
In addition to the standard vector-based repre-
sentations of meaning, there is another statistical
model called SemFi²⁵ (Hämäläinen,2018a). The
model is a relational database that captures seman-
tic relations of words based on their syntactic co-
occurencies.
Before the era of machine learning, there were
two prominent projects for modeling meaning
computationally which have been translated into
Finnish WordNet (Lindén and Carlson,2010) and
FrameNet (Lindén et al.,2019).
With the similar ideology to the hand crafted re-
sources, there have been several dierent linked
¹⁸http://vectors.nlpl.eu/repository/
¹⁹https://bionlp.utu./nnish-internet-parsebank.html
²⁰https://fasttext.cc/docs/en/pretrained-vectors.html
²¹https://github.com/google-research/bert
²²http://vectors.nlpl.eu/repository/
²³https://github.com/TurkuNLP/FinBERT
²⁴https://www.clarin.si/repository/xmlui/handle/11356/1277
²⁵https://github.com/mikahama/uralicNLP/wiki/Semantics-
(SemFi,-SemUr)
data projects in Finland representing semantics
in structured ontologies (Hyvönen et al.,2006;
Nyrkkö,2018;Thomas et al.,2018;Koho et al.,
2019). Many of the linked data projects are avail-
able on the Linked Data Finland website²⁶.
There is a Python library called FinMeter²⁷
(Hämäläinen and Alnajjar,2019b) that has some
higher level semantic tools for Finnish such as
metaphor interpretation, word concreteness analy-
sis and sentiment analysis. Sentiment analysis for
Finnish has also been studied later on²⁸ (Öhman
et al.,2020;Vankka et al.,2019;Lindén et al.,
2020). There is also research on topic modeling
methods (Ginter et al.,2009;Hengchen et al.,2018;
Loukasmäki and Makkonen,2019).
Finnish is well supported by traditional represen-
tations of semantics and latest vector based mod-
els. There is a vast amount of linked data resources
in a variety of domains. Higher-level semantics
such as metaphor interpretation and sentiment anal-
ysis also have received their share of research inter-
est, although there are many more questions related
to pragmatics and gurative language that have not
been researched, such as sarcasm detection, multi-
hop reasoning and fake news detection to name a
few.
2.4 Speech
Apart from Finnish speech being supported by com-
panies, there are some open-source tools that can
synthesize Finnish. Festival²⁹ has a Finnish voice
named Suopuhe³⁰, and eSpeak-ng³¹ can even gen-
erate IPA characters for Finnish.
There are several more modern approaches to
speech recognition (Enarvi et al.,2017;Varjokallio
et al.,2021) and speech synthesis (Raitio et al.,
2008,2014). Although, speech synthesis has not
gained much interest in the recent years.
There are several approaches to analyzing speech
prosody (Virkkunen et al.,2018;Šimko et al.,
2020). There is also some work on detecting dif-
ferent accents in spoken Finnish (Behravan et al.,
2013,2015) and named entity recognition (Porja-
zovski et al.,2020).
In summary, several approaches exist for speech
processing in Finnish relating to recognition, ac-
²⁶https://www.ldf./
²⁷https://github.com/mikahama/nmeter
²⁸a dataset https://github.com/Helsinki-NLP/XED
²⁹https://www.cstr.ed.ac.uk/projects/festival/
³⁰http://urn./urn:nbn::lb-20140730144
³¹https://github.com/espeak-ng/espeak-ng
56
cents and prosody. However, speech synthesis has
received a surprisingly small amount of attention in
the recent past. With the emergence of neural mod-
els, new research on synthesis could reach to poten-
tially interesting new contributions.
3 Discussion and Conclusions
In this survey, we have gathered research conducted
on dierent aspects of NLP. We have included links
to models and code implementations for most of the
research papers. It has been a pleasant thing to no-
tice that not only Finnish NLP research exists but
also it is often not conducted in a closed fashion, but
the actual research outputs have been made openly
available for a wider community of people even out-
side of academia. This is crucial for any language
that is relatively small, like Finnish. If Finnish aca-
demics did not release their research, there would
not be many other people in the world that would
produce high-quality tools for Finnish.
Digital extinction is something that many endan-
gered languages are facing right now (see Kornai
2013). Therefore, it is important to ensure that
NLP resources become openly available for endan-
gered Uralic languages as well. Availability itself is
not enough, however, as the resources need to be
easy to nd and use. Despite the fact that we have
open NLP tools for Finnish, we are still far a way
from a world where machines use our language u-
ently. Finnair’s in-ight entertainment system still
announces happliy: *saavumme kohteeseen Helsinki
(*we arrive in destination Helsinki) instead of ex-
pressing it correctly, saavumme Helsinkiin (we ar-
rive in Helsinki), Google Doc’s spell checker does
not recognize mostly any inectional form with
a possessive sux and predictive text in mobile
keyboards suggest overly formal normative Finnish
only.
While Finnish NLP has come far in terms of aca-
demic research and tools built as a result, we as a na-
tion are still far away from having Finnish language
technology fully integrated into the systems we use
every day. Many of the problems have been solved
already, it is just the matter of the industry nding
out about the NLP tools that are out there.
We have limited our survey to NLP tools and
methods only. We know that there are a plethora
of language resources available for Finnish as well.
Based on our experiences, many corpora are well
hidden and digging them up is a time consuming
eort worthy of a separate survey paper. Unfor-
tunately the Finnish practice of describing data on
Metashare³² is very unhelpful in this respect because
the metadata descriptions in the service hardly ever
contain information about where to access the data,
how to cite it and who the real authors are.
References
Roee Aharoni and Yoav Goldberg. 2017. Morpholog-
ical inection generation with hard monotonic atten-
tion. In Proceedings of the 55th Annual Meeting of the
Association for Computational Linguistics (Volume 1:
Long Papers), pages 2004–2015, Vancouver, Canada.
Association for Computational Linguistics.
Khalid Alnajjar, Leo Leppänen, Hannu Toivonen, et al.
2019. No time like the present: methods for generat-
ing colourful and factual multilingual news headlines.
In Proceedings of the 10th International Conference on
Computational Creativity. Association for Computa-
tional Creativity.
Hamid Behravan, Ville Hautamäki, and Tomi Kinnunen.
2013. Foreign accent detection from spoken nnish
using i-vectors. In INTERSPEECH, volume 2013,
page 14th.
Hamid Behravan, Ville Hautamäki, and Tomi Kinnunen.
2015. Factors aecting i-vector based foreign accent
recognition: A case study in spoken nnish. Speech
Communication, 66:118–129.
Piotr Bojanowski, Edouard Grave, Armand Joulin, and
Tomas Mikolov. 2017. Enriching word vectors with
subword information. Transactions of the Association
for Computational Linguistics, 5:135–146.
Mathias Creutz. 2018. Open subtitles paraphrase cor-
pus for six languages. In Proceedings of the Eleventh
International Conference on Language Resources and
Evaluation (LREC 2018).
Mathias Creutz and Eetu Eetu Sjöblom. 2019. To-
ward automatic improvement of language produced
by non-native language learners. In Proceedings of
the 8th Workshop on Natural Language Processing for
Computer Assisted Language Learning (NLP4CALL
2019), September 30, Turku Finland, 164, pages 20–
30. Linköping University Electronic Press.
Jacob Devlin, Ming-Wei Chang, Kenton Lee, and
Kristina Toutanova. 2019. BERT: Pre-training of
deep bidirectional transformers for language under-
standing. In Proceedings of the 2019 Conference of
the North American Chapter of the Association for
Computational Linguistics: Human Language Tech-
nologies, Volume 1 (Long and Short Papers), pages
4171–4186, Minneapolis, Minnesota. Association for
Computational Linguistics.
³²https://metashare.csc./
57
Senka Drobac. 2020. OCR and post-correction of histori-
cal newspapers and journals. Ph.D. thesis, University
of Helsinki, Finland.
Senka Drobac and Krister Lindén. 2020. Optical
character recognition with neural networks and post-
correction with nite state methods. International
Journal on Document Analysis and Recognition (IJ-
DAR), 23(4):279–295.
Quan Duong, Mika Hämäläinen, and Simon Hengchen.
2020. An unsupervised method for ocr post-
correction and spelling normalisation for nnish.
arXiv preprint arXiv:2011.03502.
Seppo Enarvi, Peter Smit, Sami Virpioja, and Mikko
Kurimo. 2017. Automatic speech recognition with
very large conversational nnish and estonian vocab-
ularies. IEEE/ACM Transactions on Audio, Speech,
and Language Processing, 25(11):2085–2097.
Albert Gatt and Emiel Krahmer. 2018. Survey of the
state of the art in natural language generation: Core
tasks, applications and evaluation. Journal of Arti-
cial Intelligence Research, 61:65–170.
Filip Ginter, Hanna Suominen, Sampo Pyysalo, and
Tapio Salakoski. 2009. Combining hidden markov
models and latent semantic analysis for topic seg-
mentation and labeling: Method and clinical appli-
cation. International journal of medical informatics,
78(12):e1–e6.
Stig-Arne Grönroos, Sami Virpioja, and Mikko Kurimo.
2020. Morfessor em+ prune: Improved subword seg-
mentation with expectation maximization and prun-
ing. In Proceedings of The 12th Language Resources
and Evaluation Conference, pages 3944–3953.
Lauri Haapanen and Leo Leppänen. 2020. Recycling a
genre for news automation: The production of valtteri
the election bot. AILA Review, 33(1):67–85.
Mika Hämäläinen. 2018a. Extracting a semantic
database with syntactic relations for nnish to boost
resources for endangered uralic languages. The Pro-
ceedings of Logic and Engineering of Natural Lan-
guage Semantics 15 (LENLS15).
Mika Hämäläinen. 2018b. Harnessing nlg to cre-
ate nnish poetry automatically. In Proceedings of
the ninth international conference on computational
creativity. Association for Computational Creativity
(ACC).
Mika Hämäläinen. 2018c. Poem machine-a co-creative
nlg web application for poem writing. In The 11th In-
ternational Conference on Natural Language Genera-
tion Proceedings of the Conference. The Association
for Computational Linguistics.
Mika Hämäläinen and Khalid Alnajjar. 2019a. Gen-
erating modern poetry automatically in nnish. In
2019 Conference on Empirical Methods in Natural
Language Processing and 9th International Joint Con-
ference on Natural Language Processing Proceedings
of the Conference. The Association for Computational
Linguistics.
Mika Hämäläinen and Khalid Alnajjar. 2019b. Let’s
face it. nnish poetry generation with aesthetics and
framing. In Proceedings of the 12th International Con-
ference on Natural Language Generation, pages 290–
300.
Mika Hämäläinen and Khalid Alnajjar. 2019c. A
template based approach for training nmt for low-
resource uralic languages-a pilot with nnish. In Pro-
ceedings of the 2019 2nd International Conference
on Algorithms, Computing and Articial Intelligence,
pages 520–525.
Mika Hämäläinen, Niko Partanen, Khalid Alnajjar, Jack
Rueter, and Thierry Poibeau. 2020. Automatic di-
alect adaptation in nnish and its eect on perceived
creativity. In 11th International Conference on Com-
putational Creativity (ICCC’20). Association for Com-
putational Creativity.
Mika Hämäläinen and Jack Rueter. 2018. Development
of an open source natural language generation tool for
Finnish. In Proceedings of the Fourth International
Workshop on Computational Linguistics of Uralic Lan-
guages, pages 51–58, Helsinki, Finland. Association
for Computational Linguistics.
Katri Haverinen, Jenna Nyblom, Timo Viljanen,
Veronika Laippala, Samuel Kohonen, Anna Mis-
silä, Stina Ojala, Tapio Salakoski, and Filip Ginter.
2014. Building the essential resources for Finnish:
the Turku Dependency Treebank.Language Re-
sources and Evaluation, 48:493–531. Open access.
Simon Hengchen, Antti Olavi Kanner, Jani Pekka Mar-
janen, and Eetu Mäkelä. 2018. Comparing topic
model stability between nnish, swedish, english and
french. In Digital Humanities in the Nordic Countries.
Matthew Honnibal, Ines Montani, Soe Van Lan-
deghem, and Adriane Boyd. 2020. spaCy: Industrial-
strength Natural Language Processing in Python.
Arvi Hurskainen and Jörg Tiedemann. 2017. Rule-based
machine translation from english to nnish. In Pro-
ceedings of the Second Conference on Machine Trans-
lation, pages 323–329.
Eero Hyvönen, Tuukka Ruotsalo, Thomas Häggström,
Mirva Salminen, Miikka Junnila, Mikko Virkkilä,
Mikko Haaramo, Eetu Mäkelä, Tomi Kauppinen, and
Kim Viljanen. 2006. Culturesampo–nnish culture
on the semantic web: The vision and rst results. In
Developments in Articial Intelligence and the Seman-
tic Web-Proceedings of the 12th Finnish AI Conference
STeP, pages 26–27.
Mika Hämäläinen. 2019. UralicNLP: An NLP library
for Uralic languages.Journal of Open Source Soft-
ware, 4(37):1345.
58
Mika Hämäläinen. 2021. Endangered languages are not
low-resourced! In Mika Hämäläinen, Niko Partanen,
and Khalid Alnajjar, editors, Multilingual Facilitation.
Rootroo Ltd.
Tommi Jauhiainen. 2001. Using existing written
language analyzers in understanding natural spoken
Finnish. In Proceedings of the 13th Nordic Confer-
ence of Computational Linguistics (NODALIDA 2001),
Uppsala, Sweden. Department of Linguistics, Upp-
sala University, Sweden.
Tommi Jauhiainen, Heidi Jauhiainen, and Krister
Lindén. 2021. Suomalais-ugrilaiset kielet ja internet
-projekti 2013-2019. In Mika Hämäläinen, Niko Par-
tanen, and Khalid Alnajjar, editors, Multilingual Fa-
cilitation. Rootroo Ltd.
Jenna Kanerva, Filip Ginter, Niko Miekka, Akseli
Leino, and Tapio Salakoski. 2018. Turku neural
parser pipeline: An end-to-end system for the conll
2018 shared task. In Proceedings of the CoNLL 2018
Shared Task: Multilingual Parsing from Raw Text to
Universal Dependencies. Association for Computa-
tional Linguistics.
Jenna Kanerva, Samuel Rönnqvist, Riina Kekki, Tapio
Salakoski, and Filip Ginter. 2019. Template-free
data-to-text generation of nnish sports news. In Pro-
ceedings of the 22nd Nordic Conference on Computa-
tional Linguistics, pages 242–252.
Kimmo Kettunen. 2015. Keep, change or delete? setting
up a low resource ocr post-correction framework for
a digitized old nnish newspaper collection. In Italian
Research Conference on Digital Libraries, pages 95–
103. Springer.
Mikko Koho, Esko Ikkala, Petri Leskinen, Minna Tam-
per, Jouni Tuominen, and Eero Hyvönen. 2019.
Warsampo knowledge graph: Finland in the sec-
ond world war as linked open data. Semantic Web,
(Preprint):1–14.
András Kornai. 2013. Digital language death. PloS one,
8(10):e77056.
Andrei Kutuzov, Murhaf Fares, Stephan Oepen, and
Erik Velldal. 2017. Word vectors, reuse, and replica-
bility: Towards a community repository of large-text
resources. In Proceedings of the 58th Conference on
Simulation and Modelling, pages 271–276. Linköping
University Electronic Press.
Veronika Laippala and Filip Ginter. 2014. Syntactic n-
gram collection from a large-scale corpus of inter-
net nnish. In Human Language Technologies-The
Baltic Perspective: Proceedings of the Sixth Interna-
tional Conference Baltic HLT, volume 268, page 184.
Katri Leino, Juho Leinonen, Mittul Singh, Sami Virpi-
oja, and Mikko Kurimo. 2020. Finchat: Corpus and
evaluation setup for nnish chat conversations on ev-
eryday topics. Proc. Interspeech 2020, pages 429–
433.
Krister Lindén, Erik Axelson, Senka Drobac, Sam Hard-
wick, Juha Kuokkala, Jyrki Niemi, Tommi A Piri-
nen, and Miikka Silfverberg. 2013. HFST a system
for creating NLP tools. In International Workshop on
Systems and Frameworks for Computational Morphol-
ogy, pages 53–71. Springer.
Krister Lindén and Lauri Carlson. 2010. Finnwordnet–
nnish wordnet by translation. LexicoNordica–Nordic
Journal of Lexicography, 17:119–140.
Krister Lindén, Heidi Haltia, Antti Laine, Juha Luukko-
nen, Jussi Piitulainen, and Niina Väisänen. 2019.
Finntransframe: translating frames in the nn-
framenet project. Language Resources and Evalua-
tion, 53(1):141–171.
Krister Lindén, Tommi Jauhiainen, and Sam Hardwick.
2020. Finnsentiment–a nnish social media cor-
pus for sentiment polarity annotation. arXiv preprint
arXiv:2012.02613.
Petri Loukasmäki and Kimmo Makkonen. 2019.
Eduskunnan täysistunnon puheenaiheet 1999-2014:
miten käsitellä lda-aihemalleja? Politiikka, 61(2).
Jouni Luoma, Miika Oinonen, Maria Pyykönen,
Veronika Laippala, and Sampo Pyysalo. 2020. A
broad-coverage corpus for nnish named entity
recognition. In Proceedings of The 12th Lan-
guage Resources and Evaluation Conference, pages
4615–4624.
Eetu Mäkelä, Krista Lagus, Leo Lahti, Tanja Säily,
Mikko Tolonen, Mika Hämäläinen, Samuli
Kaislaniemi, Terttu Nevalainen, et al. 2020. Wran-
gling with non-standard data. In Proceedings of
the Digital Humanities in the Nordic Countries 5th
Conference Riga, Latvia, October 21-23, 2020.
CEUR-WS. org.
Shervin Malmasi and Mark Dras. 2014. Finnish native
language identication. In Proceedings of the Aus-
tralasian Language Technology Association Workshop
2014, pages 139–144.
Sjur Moshagen, Jack Rueter, Tommi Pirinen, Trond
Trosterud, and Francis M. Tyers. 2014. Open-
source infrastructures for collaborative work on
under-resourced languages. The LREC 2014 Work-
shop “CCURL 2014 - Collaboration and Computing
for Under-Resourced Languages in the Linked Open
Data Era”.
Garrett Nicolai and David Yarowsky. 2019. Learning
morphosyntactic analyzers from the Bible via iterative
annotation projection across 26 languages. In Pro-
ceedings of the 57th Annual Meeting of the Associa-
tion for Computational Linguistics, pages 1765–1774,
Florence, Italy. Association for Computational Lin-
guistics.
Seppo Nyrkkö. 2018. Building a nnish som-based on-
tology concept tagger and harvester. In Proceedings of
the Fourth International Workshop on Computational
Linguistics of Uralic Languages, pages 18–25.
59
Emily Öhman, Marc Pàmies, Kaisla Kajava, and Jörg
Tiedemann. 2020. Xed: A multilingual dataset for
sentiment analysis and emotion detection. In Proceed-
ings of the 28th International Conference on Compu-
tational Linguistics, pages 6542–6552.
Niko Partanen, Mika Hämäläinen, and Khalid Alnajjar.
2019. Dialect text normalization to normative stan-
dard Finnish. In Proceedings of the 5th Workshop
on Noisy User-generated Text (W-NUT 2019), pages
141–146, Hong Kong, China. Association for Com-
putational Linguistics.
Tommi Pirinen. 2019a. Apertium-n-eng–rule-based
shallow machine translation for WMT 2019 shared
task. In Proceedings of the Fourth Conference on
Machine Translation (Volume 2: Shared Task Papers,
Day 1), pages 335–341, Florence, Italy. Association
for Computational Linguistics.
Tommi A Pirinen. 2015. Development and use of com-
putational morphology of nnish in the open source
and open science era: Notes on experiences with
omor development. SKY Journal of Linguistics,
28:381–393.
Tommi A Pirinen. 2019b. Neural and rule-based nnish
nlp models—expectations, experiments and experi-
ences. In Proceedings of the Fifth International Work-
shop on Computational Linguistics for Uralic Lan-
guages, pages 104–114.
Dejan Porjazovski, Juho Leinonen, and Mikko Kurimo.
2020. Named entity recognition for spoken nnish.
In Proceedings of the 2nd International Workshop on
AI for Smart TV Content Production, Access and De-
livery, pages 25–29.
Peng Qi, Yuhao Zhang, Yuhui Zhang, Jason Bolton, and
Christopher D. Manning. 2020. Stanza: A python
natural language processing toolkit for many human
languages. In Proceedings of the 58th Annual Meet-
ing of the Association for Computational Linguistics:
System Demonstrations, pages 101–108, Online. As-
sociation for Computational Linguistics.
Tuomo Raitio, Heng Lu, John Kane, Antti Suni, Martti
Vainio, Simon King, and Paavo Alku. 2014. Voice
source modelling using deep neural networks for sta-
tistical parametric speech synthesis. In 2014 22nd
European Signal Processing Conference (EUSIPCO),
pages 2290–2294. IEEE.
Tuomo Raitio, Antti Suni, Martti Vainio, and Paavo
Alku. 2008. Hmm-based nnish text-to-speech sys-
tem utilizing glottal inverse ltering. In Interspeech
2008, Brisbane, Australia, September 22-26, 2008.
Miia Rämö and Leo Leppänen. 2021. Using contextual
and cross-lingual word embeddings to improve vari-
ety in template-based nlg for automated journalism.
In Proceedings of the EACL Hackashop on News Me-
dia Content Analysis and Automated Report Genera-
tion, pages 62–70.
Ehud Reiter. 1994. Has a consensus nl generation archi-
tecture appeared, and is it psycholinguistically plau-
sible? In Proceedings of the Seventh International
Workshop on Natural Language Generation.
Teemu Ruokolainen, Pekka Kauppinen, Miikka Silfver-
berg, and Krister Lindén. 2019. A nnish news cor-
pus for named entity recognition. Language Resources
and Evaluation, pages 1–26.
Miikka Silfverberg and Mans Hulden. 2018. Initial ex-
periments in data-driven morphological analysis for
Finnish. In Proceedings of the Fourth International
Workshop on Computational Linguistics of Uralic Lan-
guages, pages 98–105, Helsinki, Finland. Association
for Computational Linguistics.
Miikka Silfverberg, Teemu Ruokolainen, Krister
Lindén, and Mikko Kurimo. 2016. Finnpos: an
open-source morphological tagging and lemmati-
zation toolkit for nnish. Language Resources and
Evaluation, 50(4):863–878.
Miikka Silfverberg and Francis Tyers. 2019. Data-
driven morphological analysis for uralic languages. In
Proceedings of the Fifth International Workshop on
Computational Linguistics for Uralic Languages, pages
1–14, Tartu, Estonia. Association for Computational
Linguistics.
Eetu Sjöblom, Mathias Creutz, and Yves Scherrer. 2020.
Paraphrase generation and evaluation on colloquial-
style sentences. In Proceedings of The 12th Language
Resources and Evaluation Conference, pages 1814–
1822.
Suzanne Elizabeth Thomas, Anna Pia Frederike Wess-
man, Esko Ikkala, Jouni Antero Tuominen, Mikko
Koho, Eero Antero Hyvönen, and Ville Rohiola.
2018. (co-) creating a sustainable platform for n-
land’s archaeological chance nds: The story of sualt.
In Digital Heritage and Archaeology in Practice. Uni-
versity press of Florida.
Jörg Tiedemann, Tommi Nieminen, Mikko Aulamo,
Jenna Kanerva, Akseli Leino, Filip Ginter, and Niko
Papula. 2020. The FISKMÖ project: Resources
and tools for Finnish-Swedish machine translation
and cross-linguistic research. In Proceedings of the
12th Language Resources and Evaluation Conference,
pages 3808–3815, Marseille, France. European Lan-
guage Resources Association.
Jukka Toivanen, Hannu Toivonen, Alessandro Valitutti,
Oskar Gross, et al. 2012. Corpus-based generation of
content and form in poetry. In Proceedings of the third
international conference on computational creativity.
University College Dublin.
Matej Ulčar and Marko Robnik-Šikonja. 2020. High
quality elmo embeddings for seven less-resourced
languages³³ In Proceedings of The 12th Language
³³We don’t agree with the title of the paper declaring Finnish
as a ”less-resourced” language. As we have seen in this paper
Finnish does have a bunch of resources!
60
Resources and Evaluation Conference, pages 4731–
4738.
Jouko Vankka, Heikki Myllykoski, Tuomas Peltonen,
and Ken Riippa. 2019. Sentiment analysis of nnish
customer reviews. In 2019 Sixth International Con-
ference on Social Networks Analysis, Management and
Security (SNAMS), pages 344–350.
Matti Varjokallio, Sami Virpioja, and Mikko Kurimo.
2021. Morphologically motivated word classes
for very large vocabulary speech recognition of
nnish and estonian. Computer Speech & Language,
66:101141.
Päivi Johanna Virkkunen, Juraj Simko, Heini Henriikka
Kallio, Martti Tapani Vainio, et al. 2018. Prosodic
features of nnish compound words. In Proceedings
of the 9th International Conference on Speech Prosody
2018. International Speech Communications Associ-
ation.
Antti Virtanen, Jenna Kanerva, Rami Ilo, Jouni Luoma,
Juhani Luotolahti, Tapio Salakoski, Filip Ginter, and
Sampo Pyysalo. 2019. Multilingual is not enough:
Bert for nnish. arXiv preprint arXiv:1912.07076.
Juraj Šimko, Martti Vainio, and Antti Suni. 2020. Anal-
ysis of speech prosody using WaveNet embeddings:
The Lombard eect. In Proc. 10th International Con-
ference on Speech Prosody 2020, pages 910–914.
61
