What language was used most for written poetry in the 1890s in Norway? What was the gender distribution of published poets in that decade? To what degree does the lyrical subject ”reveal” itself, explicitly or implicitly, in the poems? This article presents answers to these questions in relation to corpus of poetry written by Norwegian and Danish poets, published between 1890 and 1899. The corpus contains 3,440 poems from 81 books, encoded in TEI XML, and annotated with author and publication metadata, as well as lyrical features such as rhyme schemes and presence of the lyrical subject. The objective of constructing this corpus is to fill a resource gap for research on Norwegian poetry. It provides empirical data for investigating historical claims about the literary period, and comparing findings from both close and distant readings of the data. It is structured in a well-known format (TEI) that enables testing and development of automatic poetry analysis tools, as well as the addition of further annotations. We have published the data in a public repository, making it easily findable, accessible, interoperable and resuable (FAIR).
during a historical period that existing literary studies claim to be influential for modern Norwegian poetry. Previous studies has mainly focused on a handful of canonized terms and poets, while this corpus will enable comprehensive investigations of the genre as a whole during the 1890s. The second goal is to explore new analytical methods and fill a gap of Norwegianspecific digital tools for literary studies.
We describe this gap in section2, and the data selection process and the curation of the corpus metadata in section3. The process of extracting the poems and encoding them in TEI is summarized in section4. Section 5 presents some key features and distributions of the corpus metadata that help us answer our overarching questions about the 1890s poetry scene. Section 6 describes the annotation and distribution of lyrical features in the text data. We conclude the corpus construction work in sectio7nand outline planned future work.
De Sisto et al. (2024) present an overview of existing digital tools and data for automatic poetry analysis, focusing on resources for European language7s]. A[mong the 57 described resources, none were developed for any of the Scandinavian languages. The Corpus of Czech Verse and The Diachronic Spanish Sonnet Corpus (DISCO) are most similar to the corpus we present in this article in terms of annotations and size respectively. The Corpus of Czech Verse is impressively large with close to 80 000 poems, and is annotated extensively on the token and verse level. 1[4] Most of the annotations are closely linked to linguistic particularities of the Czech language, and the wide selection of annotation types allows for many diferent statistical analyses.
DISCO is a smaller corpus, containing 4,085 Spanish-written sonnets from the fiteenth
through the nineteenth centuries. 1[
These two corpora provide empirical data for poetry analysis in their respective languages and for their respective literary histories, and the lack of such a resource for Norwegian led us to create NORN Poems. Like Czech Verse, our corpus include phonemic transcriptions for each verse line, as well as a rhyme tag, and a collated rhyme scheme for each stanza. Like DISCO, we encode our corpus in TEI and share it in public repositories on widely used platfo2rTmhse. corpora are similar in size and enable many of the same analyses and discussions. However, DISCO is made exclusively out of sonnets which streamlines the computational analyses in some ways while NORN Poems is composed of a wider variation of poems.
This lyric type variation in NORN Poems enables more comprehensive literary analyses of the genre as a whole in the given time period: What was the gender distribution among published poets? What language did they write poetry in, after almost half a century of discussions about what written Norwegian should look like? What does the distribution of works by canonical authors look like, compared to non-canonical? Can we find non-canonical authors that were popular in their own time, only to be forgotten by the literary histories?
There are 22,591 books published between 1800 and 1899 that have been digitized by the Norwegian National Library. Researchers from the Norwegian literary research prIomjeacgtiNation has manually annotated 489 books of poetry from the nineteenth century, published between 1814 and 1905.3 [11] Out of these, the corpus we present here is made up of the 81 first edition books of poetry published in Norway and Denmark from 1890 through 1899, in their original language.4 The total number of digitized publications of poetry from this decade is 135, however these books include re-publications of older poetry. We have left these out in order to establish the qualities the new poetry from the 1890s shares; this allows us to discuss the labels and features of the decade as identified in literary histories.
There are several reasons for limiting the corpus to these 81 books of poetry. Firstly,
according to many literary historians, this decade allegedly marks significant changes in the lyric
genre. Something ”new” happens when poetry supposedly reappears after a downturn in the
1870s and 1880s. [
Finally, for practical reasons the corpus had to be small enough to easily toggle analysis methods. At the same time, it should be large enough to ensure that statistical patterns and analyses are generalizable to the lyric genre of the period, and not skewed too much by outliers and discrepancies.
All the poems in our corpus were originally printed in books, and digitized and made available by the National Library of Norwa5y.One of the main challenges in this project was structuring the dataset with the poem as the main object of study, instead of the book. The process involved several manual steps, including listing the title of each poem, identifying the pages each poem was located on, and assessing each split.
We received a collection of Alto XML files from the National Library’s DH-lab, containing the optical character recognised (OCR) text from the 81 handpicked, scanned books of poe6trWy.e extracted the individual poems and indexed them. A thorough manual cleaning process was conducted on the resulting text files. We corrected line breaks and indentations, fixed OCR errors, and removed unwanted artifacts such as page numbers, stamp text, and page breaks. For data analysis and processing, as well as preservation and sharing, we structured the poem 3ImagiNation: https://www.ntnu.edu/isl/imagination 4Although the majority of the literary production in Norway was published in Norwegian and in Norwegian cities, there were still a large influence by the Danish literary scene. Most major Norwegian writers were published in Denmark until the turn of the century. 5The National Library of Norway’s online librarhytt:ps://www.nb.no/search 6The DH-lab: dh.nb.no texts and metadata in the TEI format7. The hierarchy of textual elements in each poem consists of stanzas, verses, word tokens, and syllables. Additionally, rhyme schemes are encoded with the TEI attribute @rhyme on the stanza text elements.
We have encoded certain key features in the metadata of the corpus which will enable further investigations. Easily quantified measures include gender and language distributions. A more interpretive feature is whether or not the poets are canonized in literary historical accounts of the period. We have encoded canonization of an author in three wa8ys:
The NORN Poems are written in three diferent languages: Danish, Landsmål and Riksmål. While Danish had been the standard written language in Norway for many centuries, the nation-building project of the 1800s included debates about the development of a standardized Norwegian language. These debates resulted in the establishment of two new written languages: Landsmål, based on Norwegian dialects, and Riksmål, based on orthographic modification of Danish, today known as Nynorsk and Bokmål.
Riksmål is overrepresented in the corpus with 60 of the 81 publications, while 16 books are written solely in Landsmål. Two books alternate between Landsmål and Riksmål for each poem, and only three of the publications are written in Danish. The distribution is visualized in Figure1. The three languages pose a linguistic challenge for the project, both in terms of inconsistencies in the optical character recognition that generated the digital text, as well as the linguistic features and vocabularies.
From the ImagiNation corpus as a whole, we have manually identified fiteen female poets of lyric poetry in the nineteenth century1.[1] Four of these female poets are represented in our corpus. 56 authors are male, bringing the total to 60 poets in our poetry selection from the 7TEI P5: https://www.tei-c.org/release/doc/tei-p5-doc/en/html/VE.html 8A full overview can be found in appendiAx. 9https://snl.no 10https://www.idunn.no/journal/edda 1890s.11 One of the female poets stands out in quantitative terms: Karen Nilsen is represented with four books in this corpus. She wrote another during the same time period, which has not been digitized and is therefore not included here. These five books make her one of the most productive poets of this decade, only matched by Vilhelm Krag.
Sigbjørn Obstfelder is by far the most frequently mentioned poet in the six canonical literary histories, as can be seen by the overview in tableA. At the same time, there are 11 poets that are not mentioned anywhere in the literary reference works. The 48 other poets in the corpus are either mentioned a handful of times, or with a frequency of double digits. There is not necessarily a connection between the number of produced works and their position in literary history. For instance, it is interesting to note that Vilhelm Krag is one of the most prolific authors, but is mostly remembered by a single poem, ”Fandango”. Menawhile, Sigbjørn Obstfelder published only one book of poems in his lifetime, which includes the poem ”Jeg ser” which has become emblematic for the entire decade.
Karen Nilsen is not mentioned in any of the canonical literary histories before 2012, and was
ifrst described explicitly in 1988 inNorsk kvinnelitteraturhistorie [
The field of lyric theory has generated a number of diferent definitions of the lyric genre. Many of the most prominent features of modern genre theory, can be difÏcult to operationalize. However, some genre features from the nineteenth century still hold a place in Western lyric traditions. The features include that the poems are relatively short, often separated into stanzas with a certain pattern, and often contain the presence of a lyric speaker. 5[, p. 89] In these preliminary explorations of NORN Poems, we want to examine how these three traditionally important lyric features appear in the corpus.
By means of counting verse lines in NORN Poems, we can confirm that the majority of the poems are short. We find only a couple of outliers with closer to 1000 lines as shown in Figure 2. More than 75% of all poems are shorter than 42 lines. Additionally, 50% of the corpus have 26 lines or shorter, which confirm the claim that poem length is a distinctive genre feature in NORN Poems.
Poem length is also a distinctive feature within the longest poems in the corpus, although these include greater variety in form. The 25% longest poems span between 42 and 993 lines, and only 55 of the poems are longer than 200 lines, as illustrated by Fig2u.rHeowever, the 55 longest poems are written by 22 diferent poets, which can indicate an interest in playing with the poem length tradition. These long poems have a large variation in line length, stanza length, and whether they are even stanzaic, which confirms the claim that the long poem as a concept is ”a generic hybrid”. 1[0, p. 2] (a) The majority of the poems are short.
(b) Exponentially increasing poem length by lines. a a b b a a
Rhyme and rhythm (as sound patterning) play a crucial role in the lyric genre’s aural dimension. Jonathan Culler claims that ”[s]uch patterning is highly seductive, which is in part to say that a given sequence of sounds does not have some fixed, necessary efect, but invites readers to an experience.” [5, p. 134] One of our objectives is to annotate the NORN Poems with end rhyme schemes, because we are interested in recurring patterns in stanzaic poems, and we want to examine the distribution of end rhyme schemes. The annotations will later serve as training data for automatic poetry analysis tools as well.
Reddy and Knight (2011) trained a generative model on data with pre-annotated rhyme
schemes, without any phonetic information1.[
Both Reddy and Knight (2011) and Plecháč (2018) disregard phonetic information, either due to lack of resources or the possible discrepancies in pronunciations over time. They both make use of large datasets that have already been annotated with rhyming patterns, and no such dataset previously exists for Norwegian poetry.
However, there are two recently developed resources for Norwegian pronunciation: one multidialectal pronunciation dictionary and grapheme-to-phoneme models for the written standard Bokmål.1[7, 16] With this phonemic information we segment verse lines into syllable sequences, identify the last stressed syllable, and compare end rhymes across verses within a stanza, exemplified in Table 1. Our approach is rule based and naively annotates any list of lists with letter sequences1.2 In future work, we will explore whether using modern pronunciation information works better than not using it at all for the initial rhyme scheme annotation. 12Link to https://github.com/norn-uio/poetry-analysis
Lyric poetry is often recognized by the presence of a speaker in the poem, and this subject can either point inwards through a subject, or outwards through a lyric address. Throughout the nineteenth century, the subject is often understood as an expression of the poet, which establishes the lyric genre as a specifically private genre. With the development of New Criticism in the mid-twentieth century, the lyric subject is understood not as an act of the poet, but as the ”speech of a persona”.[5, p. 84]
The lyric subject may be present without an explicit mark by personal pronoun, which emphasizes the experience of a speaker in the poem. In order to explore the presence of the lyric subject in NORN Poems, and annotate unambiguously, we defined four groups of words that all indicate that the speaker is present in the poem’s context, and assigned labels for the type of lyric speaker: Explicit subject, explicit object, implicit subject, and deixis. Ta2blleists all the search terms that we used to probe the poems for diferent ways that the lyric subject appears. The ”Indicating words” also illustrate the subtle diferences in spelling between the three languages in NORN Poems.
The search results displayed in Figur3eshow that every poem in NORN Poems contains a reference to the lyric subject by use of one of the indicating words. The distribution plot confirms that the lyric subject is a prominent genre feature in this corpus. On its own, it interesting to note how omnipresent the lyric subject is during this decade. However, further investigations on how the speaker is conveyed and how it acts in the poems will be best explored through close reading.
This article presents the corpus NORN Poems that enables further research into literary and linguistic features of poetry published for Norwegian readers in the 1890s. We have highlighted three contextual features among the encoded metadata in Sectio5n, which enable relevant data partitioning, selection, and exploration. We explored more poem specific qualities in Section 6, and ways to operationalize features of the lyric genre.
Going forward, we aim to investigate other lyric features such as alliteration, anaphores, enjambment, themes and topics, and connections between poets and their poetry. A poetry analysis tool for Norwegian, both for annotating and visualizing the poems is also in development.
This work is funded by the European Union. The views and opinions expressed are, however, those of the authors only and do not necessarily reflect those of the European Union or European Commission. Neither the European Union nor the granting authority can be held responsible for them. [1] G. ”Fram”.Første greb på lyren : Digte udgivne af gymnasialsamfundet ”Fram” (1883-1891).
Hamar, 1892. url: https://urn.nb.no/URN:NBN:no-nb%5C%5Fdigibok%5C%5F201404140 6092.