Next Article in Journal
Hydrocolonialism, Countersurveillance, and “America Independent”: Poetic Framings of Revolutionary Tea Parties
Previous Article in Journal
Not New Poems but Translations: Ezra Pound’s Image-Centered Cathay from Chinese Tang Poetry
Previous Article in Special Issue
The Shapes of Cinderella: Emotional Architecture and the Language of Moral Difference
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

“The Sweetheart in the Forest” and the Synthetic Storytellers

by
Anne Sigrid Refsum
Center for Digital Narrative, Department of Linguistic, Literary and Aesthetic Studies, University of Bergen, 5020 Bergen, Norway
Humanities 2025, 14(12), 230; https://doi.org/10.3390/h14120230 (registering DOI)
Submission received: 8 October 2025 / Revised: 20 November 2025 / Accepted: 22 November 2025 / Published: 25 November 2025
(This article belongs to the Special Issue Depiction of Good and Evil in Fairytales)

Abstract

What happens to a Norwegian traditional folktale when told by a Large Language Model (LLM)? As machine-generated text becomes increasingly omnipresent, the need to understand such texts through analysis using literary scholarship and seeing them through the lens of folkloristics becomes apparent. For the purposes of examining basic structures of LLM narrative, this article uses the folktale “The Sweetheart in the Forest” (ATU 955) to examine how the style and telling of folktales is adapted by LLMs, including how LLMs display a tendency towards “floating” motifs and imagery, and how the LLMs relate to the cultural specificity of the Norwegian variant.

1. Introduction

This is not a love story. The first half of the title “The Sweetheart in the Forest” may imply a tale of sweetness and romance, but as any reader of folktales knows, the second half—“the forest”—tells us that there is mystery, darkness and fear ahead. Indeed, “The Sweetheart in the Forest” is the most well-known Norwegian variant of the folktale type ATU 955, known as The Robber Bridegroom, a tale where a young woman, while knowing she is the next victim, witnesses the murder of another young woman.1 The famous Bluebeard by Charles Perrault (ATU 312) may be seen as belonging to the same family of stories of violence against women as The Robber Bridegroom type, but the version of ATU 955 printed by Jacob and Wilhelm Grimm resembles the Norwegian variants more closely.
The following is a preliminary study in a planned series of articles investigating the folktale structures and motifs found in Large Language Models’ (LLMs’) textual outputs and comparing them to traditional variants. The use of LLMs such as ChatGPT is now so widespread that machine-generated text may be a significant, if not the main, source of written text to many users. This means that the structural and linguistic tendencies of LLM-generated texts ought to be studied from a humanities perspective, using literary scholarship. Folktales are a type of text that is structurally predictable, yet the different variants available means there is a diversity in terms of detail, scenery, style and motifs used. In an international context, there will also be diversity in culturally specific values and practices, as discussed by Jack Zipes (2012). This makes the comparison between generated tales and traditional folktales a useful method for studying the structures and cultural bias of AI-generated text. What happens to these tales and their style and content as they encounter the new media that LLMs are? To what extent do the outputs of LLMs mirror the culturally specific content of the original material?

2. The Language Model and the “Pool of Tradition”

When studying the textual outputs of LLMs I find useful analogies in folkloristics. A notorious problem of oral song and storytelling studies is the ephemerality of the performance and of the variant, as it will change with the setting, the context and the audience. The variant texts provided by different LLMs at different times, using slightly different prompts, and without insight in how the LLMs are programmed, have a similar quality of being hard to grasp and complicated to analyze in a meaningful way.
Very simply put, Large Language Models generate text by running through vast amounts of training data collected from the Internet, regardless of text genre. When presented with a prompt, calculating the most likely string of text that would make a satisfying response to the prompt. I think of the training data of LLMs as analogous to the folklorist Lauri Honko’s (2000) term “pool of tradition”. Honko explains how the various genres of oral performance are made by imagining a “model (…) of narrative elements bubbling freely in a pool of tradition, ready to float in many directions and fuse in novel ways” (Honko 2000, p. 18). Honko deliberately uses the word pool, “meaning both a body of water and a fund to which many contribute and from which many can draw” (ibid.). The rules and models for that fusion vary depending on genre. “Whatever is shared by more than one singer,” Honko writes, “belongs to the pool of tradition. The pool holds a multiplicity of traditions, mostly in a latent state, only parts of it being activated by the individual” (ibid.). The websites, forums and social media communities used as material for LLMs consist, at least for now, largely of human thoughts, expressions, and narrative elements and structures. In the case of LLMs, the activation of the latent multiplicity of traditions is carried out by a machine. The properties of language models and the programming tweaking the results provided means an LLM should be considered a synthetic storyteller.2
In this article I will consider Large Language Models as storytelling actors drawing on a pool of tradition of an enormous scale. Folktales and fairy tales have a relatively consistent narrative structure, which make them easily replicable for LLMs and human authors alike. How Large Language Models relate to literary genre has not been widely studied, but we may assume that genre is mainly constituted in two ways: firstly, as inherent structures in the texts of the datasets, replicated by the LLMs’ modelling of patterns in the data they process, and secondly, through RLHF (Reinforcement Learning Human Feedback), a process where users provide feedback on the suitability of the LLM’s responses (Burchell 2024).

3. “The Sweetheart in the Forest” as Norwegian Folktale

“The Sweetheart in the Forest” is the title of the ATU 955 variant printed in Norway’s canonical and hugely influential collections of folktales collected, edited and published by Peter Christen Asbjørnsen (1812–1885) and Jørgen Moe (1813–1882).3 While most of the Norwegian folktale corpus consists of tales of magic, this is considered a “romantic tale”, as there are no supernatural elements in the story.4 This tale was not printed in Asbjørnsen and Moe’s earliest collections (1845/1852/1866), rather, it was only printed in 1871 in an edition made by only Asbjørnsen. The reason for the late printing could be the violent content, but could also be Moe’s reluctance to print material he considered to be booklore, i.e., not part of the actual local tradition but too inspired by printed material (Hodne 1979, p. 62). The reason for his consideration is the style of telling particular to the informant, but also the background of the informant, Karen Ancher Huss (1785–1859), herself.
When Moe encountered Karen Ancher Huss in 1847, she was known as Kari Prestedatter (Kari the Priest’s Daughter).5 Her tragic, and, for a folktale informant exceptional life story, made her a person of interest for Moe, and for later folklorists such as Ørnulf Hodne (1979, 1998). Most informants of folktales were among the poorest of any parish. They were contract farmers; “innerster”, the landless lodgers of contract farmers; or “legdslemmer”, beggars living upon the parish in a system of local farms housing and feeding them by rota. Some suffered various impairments, or were famous for their facial deformities. At the time of Asbjørnsen and Moe’s collection travels, folktales were considered old fashioned and mostly suited for children, dismissed as “nursery talk” and “traveller stories” (Hodne 1998).6 Nevertheless, to the informants keeping these tales alive, they may have served as a form of social and cultural capital.
In 1847 Karen was in her sixties, and had been the wife of a poor contract farmer’s lodger for 30 years, but she was born the daughter of a local clergyman, and on both sides descended from clergy going back many generations.7 However, her family fell on hard times around the time she was born. The year of Karen’s birth her father had lost his position to his brother-in-law, Karen’s maternal uncle, and died while in Copenhagen petitioning the king for a new posting. Her mother was remarried to a sexton some distance away and went to live with him without her children. Later Karen’s stepfather fell to drinking and was convicted for extortion together with his father-in-law, Karen’s maternal grandfather (Hodne 1979, p. 133). Hodne has not been able to establish where Karen spent her youth, but finds it likely that she lived with relatives, possibly in a nearby town, and was educated there (ibid.). She turns up in the records again when she gets married at age 30 to the 22 year old son of her clergyman uncle’s estate manager, a landless contract farmer. While they must both have lived at the parsonage for a few years preceding their marriage, the class divide between them was significant. Karen’s husband Søren Tostensen had by then become a corporal in the army, most likely to try to raise himself in society and become good enough for her family. However, the marriage may also have been an act of urgency: Their first child was baptized three weeks after their wedding. They lived partly as poor contract farmers, but Søren probably made most of his income producing hard spirits, and when he died the cause of death was determined to be “excessive consumption of liquor” (Hodne 1998, p. 139).8
However, before their marriage both Karen and Søren would, at Karen’s uncle Simon Ancher Bruun’s parsonage, have had access to a rich cultural environment. Bruun was wealthy and exceptionally literate, even for his class and standing, being in possession of a library of 319 books, including even English books, a foreign language not well known by the Norwegian educated class at the time (Hodne 1998, p. 137). Karen’s repertoire of folktales may have been learnt from an unknown foreign book of fairy tales. Hodne speculates whether the stories themselves had been part of Karen’s literary inheritance, but they may also have been local. Perhaps only the style of telling was influenced by Karen’s upbringing and education. There is a profound sensibility of the dualism between good and evil in Karen’s tales, says Hodne, with strong and explicit details in both directions, but her tales also contain vivid descriptions of love (Hodne 1979, p. 178). Hodne (1979) believes the tragic and romantic tenor of Karen’s folktale repertoire can be connected to her personal experiences.
Hodne argues that while most Norwegian folktales feature a setting that is close to the Norwegian common man’s reality, Karen’s style is distinguished by picturing foreign worlds. There are castles and palaces, and the scenery is that of foreign lands, with sandy beaches by the sea, roses and lilies, and green meadows. The imagery does not resemble Norwegian landscapes (Hodne 1979, p. 179). However, it is the imagery of the local ballads of oral tradition, of the written popular culture accessible to most of the rural population at the time, and of many other folktales not included in Asbjørnsen and Moe’s collections for the same reason. Additionally, regardless of the differences in style, the contents of the variants of ATU 955 from elsewhere in the country are consistent with the contents of Karen’s variant.
“The Sweetheart in the Forest” opens with a young woman, who “was so beautiful that she was renowned across many kingdoms, and the numbers of suitors who came to her were as many as the leaves that fall in an autumn” (Asbjørnsen and Moe 2024, p. 59).9 One man makes himself out to be richer than the others, and as he also is handsome, he is the preferred suitor. He frequently visits her, but one day he asks her to come visit him. He cannot fetch her himself, but he tells her he will leave a trail of peas along the way for her to follow.
“Now, however it did or didn’t happen, he scattered the peas a day early” (p. 59). This is the stroke of destiny in the story, and it becomes his downfall. As she arrives a day earlier than expected, he is not at home to greet her. Inside his beautiful house there is “nothing to see other than a wondrous bird that hung in a cage beneath the roof” (p. 59).10 The girl proceeds further into the house, where there are many riches, but each time she enters a new chamber, the bird cries out: “Beautiful maiden, be bold, but be not too bold!” (p. 60). As she goes, she finds one chamber filled with buckets of blood, and the last is full of dead bodies “and of skeletons of slain women” (p. 60). The bird tells her to hide under the bed. She does so: “[s]he crawled so far in against the wall as she could; yes, she was so scared that she would have liked to have crawled into the wall, had she been able” (p. 61). Presently, her suitor enters the room with another maiden. He rips off “her everything—both clothes and gold—even to a ring she had on her little finger” (p. 61). When he cannot get the ring off, he chops off the maiden’s finger, which falls onto the floor and bounces under the bed. While he proceeds to murder the maiden, the girl under the bed takes the finger with the ring on it and hides it.
The suitor has a little boy with him, who is tasked with finding the finger. When discovering the girl, the boy does not tell on her, but lets the girl know when the robber is out and the coast is clear for her escape. She runs home and tells her father everything. Some time later, her suitor comes to visit, “and he was so fine that it dripped from him”, and asks why she did not come when she promised. Her father makes an excuse for her, but says they have people coming for a feast, and they may as well turn it into an engagement party. When they have finished eating and are seated at the table, a sort of rhetorical game begins where the girl reveals the truth by telling every step of what she has experienced as if it were a dream.
“I dreamt that I walked upon a broad road, and peas had been scattered, there where I walked”.
“Yes, that is like when you walk to my place, my dear,” said the sweetheart (p. 63).
In this way, they go back and forth about the great size of his house and of the many fineries it contains, and he agrees with her. But then she gets to the unpleasant parts:
“When I went into the next chamber, the bird began to scream: ‘Beautiful maiden! Beautiful maiden, be bold but be not too bold!’ And in that chamber there stood barrels and buckets around all the walls, and they were full of blood!”
“Fie, that is terrible! It is nothing like at my place, my dear,” said the sweetheart. Now he was hurting, and wanted to leave.
“It is just a dream that I am telling, of course,” said the daughter of the house (pp. 63–64).
She proceeds to tell everyone of the skeletons and dead bodies and the whole scene with the maiden he undressed and killed. When she comes to the chopped off finger and the ring that went under the bed, she pulls out the finger as evidence. And so “they took him and killed him and burned both him and the house in the forest” (p. 64). Balance is restored, and the girl is safe.
While the details of the violence committed contribute to the fictionality of the tale by being grotesque, the rest of the tale would have seemed realistic and relatable to Karen’s contemporary audience. In early modern times up to the mid-nineteenth century in rural Norway many matches were made by the young people themselves through night-courting.11 This was a custom where young men would visit young women at night, to talk and get to know one another. Several girls would share sleeping quarters, and the young men would arrive as a group, and each would lie down in a girl’s bed, both of them fully clothed, and talk to her. When a relationship had been formed, the chosen suitor would visit on his own, after the group had finished their visit. This is the type of visits implied at the opening of “The Sweetheart in the Forest”. The young woman going to visit his house would not surprise anyone, as she would have to inspect her future home and the material circumstances he had to offer. The question of choosing a trustworthy partner and incidences of violence within marriage would have been topics of real-life experience and interest to a traditional folktale audience. This tale will have served not only as warning, but also as entertainment due to its exciting content and structure; and as comfort and a community-based processing of the existence of such horrific events.
There are nine Norwegian variants in total, all of them from the southwestern (Western Telemark, Åseral, Setesdal, Lista) or western parts (Sogn, Sunnmøre, Surnadal) of Norway. Some have been printed in local history or folklore books, while others only can be found in manuscript collections. The basic structure shared by all of them is: A young lady is betrothed to a man, she goes to visit his splendid house, she realizes he is violent and has to hide. While in hiding she witnesses the murder of another woman and comes into possession of that other woman’s finger with a ring on it. She manages to escape, sometimes aided by another, and when he comes to her house to ask about her, she entraps him and he is punished.
The nature of the antagonist, the suitor, is sometimes rendered more or less dangerous by the mode of telling of the variant. Two of the variants, Faret (1872) and Fylling (1874) present the story not as folktale, but as legend, meaning there is an implied core of historical events, which adds a touch of frightening realism. The Faret variant even gives him a name (“Stefalun”) and a specific location for his house, “at Spind, by the post road”.12 The Fylling variant places the suitor as inhabitant of Jølster, a neighbouring area to the place where the events supposedly took place—but several hundreds of years ago, as the protagonist’s father is reported to be a “chieftain”. However, there is also the possibility of reducing the sense of danger by adding a layer of fictionality, and making the antagonist a “hulderkall”, a man of the mythical underground people (Lona 1880). This means that when he is revealed to be a murderer in front of other humans he disappears entirely, and subsequently ceases to be of danger to anyone.
Most of the traditional variants provide the girl with a helper, a little boy as in “Kjæresten i skogen”, but usually it will be an old woman working as a housekeeper for the suitor telling the girl to hide before her suitor arrives. This helper is consistent with many other tales, and is also found in German versions of The Robber Bridegroom (Ashliman 2023). The bird giving her the warning is only seen in two of the Norwegian variants, as well as in German versions, and turns out to be persistent image in the variants I generated using Large Language Models.

4. Floating Motifs and Imagery: LLMs Telling Tales

For the purposes of this article, I generated tales meant to resemble “The Sweetheart in the Forest”. This was carried out using the LLMs ChatGPT, DeepSeek, Gemini and Claude every three to four days between 10 September and 22 September 2025, ending in a total of 32 tales.13 The prompts were made as simple as possible on purpose, and the LLMs were accessed on their websites and without logging in. With Claude I also used the incognito version. This, as well as the simplicity of the prompts, was chosen in order to see what the LLMs came up with without too much interference. The prompts used were as follows:
  • “Fortell eventyret ‘Kjæresten i skogen’”, or the relatively similar “Fortell det norske folkeeventyret ‘Kjæresten i skogen’” (Literal translation: “Tell the folktale ’The Sweetheart in the forest’”, “Tell the Norwegian folktale ‘The Sweetheart in the Forest’”).
  • “Fortell en norsk versjon av eventyrtypen ATU 955” (Literal translation: “Tell a Norwegian version of the folktale type ATU 955”).
The results of both prompts were different for each attempt. The first prompt type yielded a diverse corpus of tales, with only very few closely resembling an ATU 955 type tale, but most of them complete fabrications in what resembles folktale style. The second more specialized prompt using the ATU number provided more predictable results with marked similarities to the Grimm Brothers’ “The Robber Bridegroom”. Both types of generated tales displayed many instances of what I call “floating” motifs and images, as well as stylistic features specific to LLMs. This is discussed below.
LLMs generate tales shaped to suit a pedagogical purpose. This comes across in the style of writing, but also in the paratext surrounding the actual tale. At the end of the tale, ChatGPT will offer to make new versions suited for reading aloud to children, or versions written in a more old-fashioned language. The moral of the story is provided thus by Claude: “The moral is that one must trust one’s intuition and not be fooled by beautiful appearances” (Claude 3).14 The end of the tale will be followed by an unprompted analysis, such as a summary of the story in bullet points, a list of motifs compared to other stories, or an explanation of what is “Norwegian” about the version offered up:
**Norwegian features::** The folktale is set in a Norwegian farm with elements like *cured meat*, *barn*, and *sawhorse*. The punishments are not violent, but more psychological and practical, which fits a Norwegian context (DeepSeek 5, original punctuation).15
Some of the analytical comments made by LLMs may be in line with the content of the generated tale, such as the lists of motifs or suggestions to relevant themes, but the further attempts at analysis are of little use. It is likely that this tendency is connected to the use of folk- and fairy tales mainly as children’s reading, and possibly to creators of LLMs’ wish to appeal to an educational market.
The folktale structure is preserved in the entire corpus of generated tales,16 in the sense that there is a formulaic opening: “Det var en gang” (Once upon a time), a female protagonist, a series of events mirroring recognizable folktale motifs, and in half of the texts a resolution ending in “De levde lykkelig alle sine dager” (They lived happily ever after). To achieve “happily ever after”, generated tales added a sentence at the end saying, for example, the girl married “en snill og ærlig mann” (a nice and honest man), and so, despite the carnage and trauma, all is well that ends well. This is in the tales where the male counterpart functions as an actual antagonist, as in “The Sweetheart in the Forest”. In other generated tales, he can be described as more of a co-character, in Greimas’ (1974) terms functioning sometimes as object and sometimes as helper.
Stylistically, LLM tales tend to make the implicit explicit. To a traditional community, and indeed to many readers of folk- and fairy tales today, a tale could leave many “gaps” (Leerstellen, Iser 1978) open to the listener without there being a loss to the experience of the narrative. The gaps may in fact augment the experience. According to John Miles Foley’s theory of traditional referentiality these gaps were filled by the users’ knowledge of other forms, other variants, and other similar stories or songs (Foley 1991). An experienced audience of folk or fairy tales “know” that a promising start must mean complications ahead, that “a girl”, unless designated otherwise, is beautiful and good, or, for that matter, that a forest is a place of mystery, darkness and transformation—but also that riches tend to come to those that deserve them. The LLM text, not coming from a human with a communicative purpose, does not have an implied reader imbued with traditional referentiality in mind.
This tendency is especially evident in the treatment of the protagonist’s feelings towards her suitor. The question of whether it would have been possible to avoid such a man comes up in several of the traditional Norwegian variants, as well as in other European variants. Some variants blame the girl herself, saying that she was too demanding in her requirements for a husband. Torbjørg Gveven’s variant says the girl only wanted to marry a man who had a silver nose. And then the robber turns up, with a silver nose (Gveven 1943). Another Norwegian variant says the girl liked the man, but her father knew he was a murderer, and she runs away to be with him (Fylling 1874). The German “The Robber’s Bride” has the mother persuading the father to allow the suitor to have their daughter (Ashliman 2023). However, by far the main motif seen in this part of the story, as seen above, is that there are no objections from any party—except for some sort of warning as the girl approaches the suitor’s house. This is usually a bird, or a voice, or a sign above the door of the house. We do not need to venture far into psychoanalytic readings to interpret this warning as a manifestation of the girl’s intuition that something is wrong.
There is, however, a sentence in the Grimm Brothers’ version about the girl not liking the man she is to marry, and the LLM tales have used this in every iteration of the tale. The desired sense of dread and danger in the girl is established through explicit descriptions of the girl’s view of her suitor. While the suitor consistently is described in terms such as “staselig”, “rik” og “velkledd” (fancy, rich and well-dressed), the girl responds by not liking him. The situation is explained by Claude: “But the girl felt uneasy when she looked at him—there was something cold in his eyes” (Claude 3).17 Or as by ChatGPT: “[S]he unwillingly became engaged to him, even though something about the man made her uneasy” (ChatGPT 8).18 The uneasiness that shows up in the reader through the knowledge created by traditional referentiality is not provided with room to exist in the text. The LLM, as it does in the paratext, provides the analysis as well as the story. In traditional variants there is more room for the reader/listener to make up their own minds, while literary versions of folktales often will have left less of the interpretation to a sense of traditional referentiality. LLMs seem to amplify this feature of “literarization” of folktales.
Another striking feature of the LLM-generated tales is the unambiguousness of the scenic elements. In every Norwegian variant, the house belonging to the suitor is beautiful and grand. There are chambers filled with gold and silver, there are ample amounts of food and drink on the table, or the farm covers a large area with many different houses on it. After the warning, the reader/listener follows the girl as she experiences the house. This serves the triple purpose of shedding a positive light on the suitor and on the girl’s possible future, increasing the dread caused by the warning—because something must be very wrong for the girl to be warned away from such a lovely house—and through this providing the story with the build up towards the turning point when she discovers the cupboard with a man’s head in it, or the chamber with the barrels filled with blood. With only two exceptions the LLMs describe the house as unpleasant. Regardless of model, the house is some slight variation of dark, derelict and dirty: “The forest was darker and darker the further she got. The birds quieted, and the trees seemed to whisper warnings. Everything was quiet. She went inside” (ChatGPT 6).19 Or: “Finally, they arrived at a large, gloomy house in the middle of the forest” (Claude 4).20 This is also a feature of the Grimm brothers’ version. However, in the LLM corpus the house is described once as beautiful: “After a while she came to a big, beautiful house” (Gemini 9) and once as large and grand: “Deep inside the dark forest she came to a large, grand house” (Claude 6).21 While some of the German version elements are persistent, there are sometimes glimpses of other influences on the tales generated.
Both the ATU 955-resembling and the other tales generated by LLMs displayed a tendency towards “floating” motifs, by which I mean motifs that have “detached” from the tale where they normally would belong. This could be due to certain tales often being published together, or discussed in the same context, so that the LLM would find them in close proximity to one another. Several of the LLM variants of ATU 955 integrate motifs seen in Bluebeard. The two types were often intertwined even for folklorists, such as Rikard Berge (1913) who has named one of the Norwegian ATU 955 variants “Blåskjæg” (Bluebeard).
In the tales that did not resemble an ATU 955 variant, the LLMs displayed having made connections between the title “The Sweetheart in the Forest”/”Kjæresten i skogen” and other tales of magic; especially to ATU 425 “The Animal as Bridegroom”, another story of an active female protagonist. In a Norwegian context this is not surprising, as Asbjørnsen and Moe’s printed variants of ATU 425 “East of the Sun and West of the Moon” and “White Bear King Valemon”, are among the most well-loved and well-known folktales in Norway. In both these variants the heroine goes to live with a polar bear that becomes a man during the night. A curse has made him so, and when she yields to the temptation to take a candle and look at him while he sleeps, he disappears. She suffers hardship and must pass several tests to rescue him. The cleverness and resourcefulness shown by the ATU 425 heroine and the protagonist of «Kjæresten i skogen» provides a clear intertextual connection, as it probably did to the users of the traditional variants. Several of the generated tales use motifs such as falling in love with a bear, transformation into animal form; or the sudden disappearance of the protagonist’s lover, whereupon the protagonist goes to look for him. The LLMs seem to prefer this story of an actual happy ending, to the gory details of “The Sweetheart in the Forest”.
There is an implicit cannibalistic motif present in some of the Norwegian variants, but it is explicit in the Grimm Brothers’ “The Robber Bridegroom”. This motif is favored by the LLMs over the more sexually coded violence in the tale.22 This could be due to the implicitness of the sexual violence in the traditional variants. We as readers or listeners must infer whether there was sexual violence committed or just murder. I would argue towards the former. In addition to the obvious situation in the folktale, the bird or voice of warning calling out to the girl to be “bold, but not too bold!” suggests the balancing act performed by young women at the time between getting to know a young man and achieving the sort of intimacy required for making a match, without crossing the line sexually. The custom of night-courting must have been conducive to making each other’s acquaintance, as well as an arena for temptation (and many couples did have children out of, or barely inside, wedlock, including the informant Karen herself). To the nineteenth century listeners of Karen telling, there could be no doubt as to the specific danger for a girl being alone in the presence of a bad man, and there must be a reason why versions of this tale without exception is about a young woman in dangerous proximity to a man she is about to marry, and not, for instance, about a boy contracted to work as a servant finding himself with a sinister master. There is also the instance of the robber (and, if applicable, his men) tearing off the clothes of the woman about to be murdered. Of course, the clothes are valuable, but the image created is that of a naked and vulnerable woman in the presence of a violent man.
In the corpus of generated tales there is only one instance of an LLM referring to a woman without any clothes, when the girl enters the room full of dead bodies: “There inside hung dead people—women, young like her—some with their clothes torn off, some with open and empty eyes” (ChatGPT 3).23 Most of the LLM-generated variants retain the cannibalistic motif, expressed in that the old woman warns the girl the robbers will eat her, and when the other young woman appears, she is only described as being killed, and her finger with her ring cut off. Sometimes the robbers carve up her body to prepare her for eating: “[C]hopped her into pieces and sprinkled salt over the meat” (Claude 10). In one instance of telling of the finger with the ring on it, it becomes augmented to the whole hand of the murdered young woman, so the protagonist of the story comes in possession of a whole severed hand (Claude 4).24 There is no lack of blood or violence, but neither of the young women of the tale seem to have a body for any other purpose than for the flesh on it to be eaten.
The cannibalistic motif may be the explanation for an issue I call “floating imagery”. As with motifs detaching from stories, imagery will detach itself from the motif it belongs to and find uncanny outlets. For instance, one generated variant describes how the protagonist sees a large red handbag containing a gold ring with a large red diamond on it.25 She steals the diamond, as she recognizes it as her own engagement ring. However, when the crime committed by the antagonist is revealed, the ring turns into a ring made out of flesh (Gemini 2).26 Another tale has the girl entering a forest where “the wind carried a smell of rotten meat” (ChatGPT 3).27 None of these variants have any reference to the eating of bodies, and this meat out of place seems to be remnants of the cannibalistic motif present in the Grimm Brothers’ version.
The bird giving the girl the warning is also present as a floating image. Two generated variants make use of this motif, both from Gemini (6 and 9), but there are also instances of birds turning up where they do not belong in the tale. A DeepSeek tale with little other resemblance to the requested ATU 955 type, presents a number of birds as helpers in all things, in a manner similar to the Disney version of Cinderella (DeepSeek 6). There are also echoes of the bird imagery known from horror films, in variants where they emphasize the frightening appearance of the suitor. ChatGPT has him being followed by ravens and wolves (ChatGPT 1) and Claude fills his house with “dead birds hanging from the ceiling” (Claude 4).28 These instances of floating imagery show how LLMs in generating a structurally fairly simple genre, still will include elements from a number of other genres. I will not go into the specifics of the linguistic style here, but suffice to say, there are breaks from the folktale tenor, especially in the dialogues, evident of such genre mixing.

5. Conclusions

The Norwegian folktale “The Sweetheart in the Forest” and its variant siblings are part of a large international tradition of tales thematizing men’s violence against their female intimate partners, a story that will never get old. However, when the atrocities committed in the tale focuses on cannibalism, the story becomes one of grotesque body horror, not one of stark reality—as the Norwegian folktale is. With the generated variants of “The Sweetheart in the Forest” using LLMs, this tendency was clear. The cannibalistic motif is present in the Grimm Brothers’ version, and it is likely that this version is more present than any other in the training data of the LLMs because the Grimms’ fairy tales have been a staple in the education of the anglosphere for generations. When looking for any type of “Norwegianness” in the LLM-generated variants, this is found mostly in props present in the setting of some tales and in the suggestions made by the LLMs in their analyses following the generated tales. These analyses are complete fabrications.
The LLM-generated folktales have instances of “floating” motifs and imagery where motifs belonging to a specific tale find their way into a variant of another, or where an image is placed in a new context, creating, in the context of “The Sweetheart in the Forest”, absurd and uncanny results. This must be the result of the synthetic part of the LLM as storytelling actor, where genres and texts are sifted and blended in an artificial way. There is a need for further study of such synthetic stories.

Funding

This research was funded by the European Research Council grant AI STORIES, doi: 10.3030/101142306.

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

Dataset available on request from the author.

Acknowledgments

The author would like to thank Jill Walker Rettberg and Zahra Rizvi for valuable feedback, and Ida Tolgensbakk for helpful discussions on English terminology for Norwegian cultural history terms. While Large Language Models were used for generating the corpus of texts analyzed in this article, the entirety of the written text is the author’s own.

Conflicts of Interest

The author declares no conflicts of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or in the decision to publish the results.

Notes

1
ATU numbers refers to the Aarne-Thompson-Uther international folktale classification, see Uther (2004).
2
On the usage of «synthetic» in a Large Language Model context, see Pilipets and Geboers (2025); Rettberg and Wigers (2025); de Seta et al. (2024).
3
The Norwegian title is “Kjæresten i skogen”. While in modern Norwegian this is the general term for romantic partner, “kjæreste” was at the time of telling the word used for sweetheart or fiance/fiancee. For quotes from “The Sweetheart in the Forest” I use the translation carried out by Simon Roy Hughes in 2024, the first ever complete and annotated English translation of all of Asbjørnsen and Moe’s tales. All other translations are my own.
4
A “romantic tale” in terms of the Aarne Thompson Uther folktale classification system. However, the presence of the talking bird and intertextual link between the protagonist of “The Sweetheart in the Forest” and in various tales of magic make me think of the distinction between romantic and magic tales as more of a continuum and depending on the variant at hand.
5
“Kari” being the vernacular version of the more refined “Karen”.
6
“Ammestuesnakk” or “Fantehistorier”. “Fant” was a derogatory term for Norwegian Romani travellers.
7
In most rural Norwegian communities in the 16th to 19th Century, the local clergyman was the sole representative of the upper classes; educated in Denmark, learned in foreign languages, speaking a different spoken language to the local vernacular, and altogether part of a culture distinctly different from his parishioners.
8
“[O]verdreven Nydelse af Brændvin” (“[E]xcessive enjoyment of spirits”).
9
All references to the text of “The Sweetheart in the Forest” from here on out are from Asbjørnsen and Moe (2024).
10
Here, Jørgen Moe’s manuscript has “Papegøie” (parrot) instead of bird, a detail revealing the exotic and un-Norwegian tendencies of Kari Prestedatter’s style of telling, and showing how Jørgen Moe Norwegianized her style.
11
See Hans Henrik Bull (2005) “Deciding Whom to Marry in a Rural Two-Class Society: Social Homogamy and Constraints in the Marriage Market in Rendalen, Norway, 1750–1900”.
12
Spind is part of modern day Farsund municipality, at the very south coast of Norway.
13
The full corpus of generated folktales is available as a dataset, see above. The model names and numbers used correspond with the entries in the dataset.
14
“Moralen er at man må stole på sin intuisjon og ikke la seg lure av vakre fasader”.
15
“**Norske trekk:** Eventyret er satt i en norsk bondegard med elementer som *spekekjøtt*, *låve* og *vedbukk*. Straffene er ikke voldelige, men mer psykologiske og praktiske, noe som passer inn i en norsk kontekst”.
16
LLMs tend to favour a folktale structure in stories, as seen in Rettberg and Wigers (2025).
17
“Men jenta følte seg urolig når hun så på ham—det var noe kaldt ved øynene hans”.
18
“[H]un ble motvillig forlovet med ham, selv om noe ved mannen gjorde henne urolig”.
19
“Skogen ble mørkere og mørkere jo lenger inn hun kom. Fuglene stilnet, og trærne syntes å hviske advarsler. Til slutt kom hun til et gammelt, forfallent hus midt i skogen. Alt var stille. Hun gikk inn.”
20
“Endelig kom de fram til et stort, dystert hus midt i skogen”.
21
“Etter en stund kom hun til et stort, vakkert hus”; “Langt inne i den mørke skogen kom hun til et stort, flott hus”.
22
Two of the Norwegian variants has a reference to cannibalism: The young woman who is to be murdered, is told she can save her own life by eating from a man’s severed head and drinking from a bowl of blood; she cannot do it, and so must die (Gveven 1943; Gravem 1915).
23
“Der inne hang døde mennesker—kvinner, unge som henne—noen med klærne revet av, noen med øynene åpne og tomme”. There was also one instance, on 22 September 2025, of ChatGPT stopping midway in generating a text looking very similar to the Norwegian “The Sweetheart in the Forest”, citing violation of their usage policy. I am inclined to believe this was connected to the content of the text.
24
“hakket henne i stykker og strødde salt over kjøttet”.
25
“en stor rød veske som inneholder en ring med en stor, glitrende rød diamant”.
26
Svanhild, skjelvende, svarer: “Jeg vil fortelle deg hva jeg så. Jeg var forlovet med en mann som bor dypt inne i skogen. Han ba meg komme og besøke ham, men da jeg kom fram, så jeg at hjemmet hans var et røverrede. Jeg så deg og dine menn drepe en stakkars kvinne. Og jeg så at du bar på en stor rød veske som inneholdt min ring.” Svanhild løfter fram hånden sin. Men da ser hun at ringen hennes ikke er der lenger. I stedet ligger en blodig ring av kjøtt på hånden hennes (Svanhild, trembling, replies: “ I want to tell you what I saw. I was engaged to a man who lives deep inside the forest. He told me to come visit him, but when I got there, I saw that his home was a robbers’ den. I saw you and you men murder a poor woman. And I saw you carrying a large red purse containing my ring.” Svanhild puts out her hand. But then she sees that her ring isn’t there anymore. Instead a bloody ring of meat is lying in her hand.).
27
“vinden bar med seg en lukt av råttent kjøtt”.
28
“Hun gikk fra rom til rom, og overalt hang det døde fugler i taket” (“She walked from room to room, and everywhere there were dead birds hanging from the ceiling”).

References

  1. Primary Sources

    Anon. 1872. [informant]. Collected by T. Faret. NFS Moe M. 114. SAMLA. Available online: https://samla.no/viewer/piresolver?id=a2624ef7-0296-426a-8cc4-df54321bf34f (accessed on 13 May 2025).
    Anon. 1879. [informant]. Collected by Ivar Mortensson. NFS Mortensson 8. SAMLA. Available online: https://samla.no/viewer/piresolver?id=136c3f36-c599-48cd-8651-59076d720b42 (accessed on 13 May 2025).
    Berge, Hæge. 1913. [informant]. Collected by J. Berge. Rikard Berge, TEMU/TGM-A-1003/F/L0007/0001: 251–99/251 Fyrisdal, Laardal. Eventyr og viser, 1913, pp. 22–23. Available online: https://www.digitalarkivet.no/ed60149526000014 (accessed on 25 May 2025).
    Fylling, Peder. 1874. “Paa Osen Gaard”. Folkesagn 1. Aalesund: F. M. Olsens Forlag og Bogtrykkeri.
    Gravem, Ann Tronsdotter. 1915. [informant]. “Iva Pænnskjægg”. Collected by E. Langset. Norsk Eventyrbibliotek 11. Guten i gadden. Eventyr frå Sogn og Fjordane, Møre og Romsdal. Edited by Brynjulf Alver. Oslo: Det norske Samlaget, 1980.
    Gveven, Torbjørg. 1943. [informant]. Collected by Tone Gudve Gjellstad. NFS Gjellstad 5. SAMLA. Available online: https://samla.no/viewer/piresolver?id=b4078501-e964-4162-b96f-33929e679194 (accessed on 5 September 2025).
    Lona, Ingeborg Nilsdatter. 1880. [informant]. Collected by Moltke Moe. NFS Moe M. 12. SAMLA. Available online: https://samla.no/viewer/piresolver?id=906ab1ba-c545-4617-8558-248583483ffb (accessed on 13 May 2025).
    Sande, Olav. 1892. “Røvaren og gjenta”. Segner fraa Sogn 2. Bergen: Mons Litlére.
  2. Secondary Sources

  3. Asbjørnsen, Peter Christen, and Jørgen Engebretsen Moe. 2024. New Collection and Additional Folktales. The Complete Norwegian Folktales and Legends of Asbjørnsen & Moe. Volume 2. Translated by Simon Roy Hughes. [Google Scholar]
  4. Ashliman, D. L. 2023. The Robber Bridegroom and Other Folktales of Aarne-Thompson-Uther Type 955. Available online: https://sites.pitt.edu/~dash/type0955.html (accessed on 25 September 2025).
  5. Bull, Hans Henrik. 2005. Deciding Whom to Marry in a Rural Two-Class Society: Social Homogamy and Constraints in the Marriage Market in Rendalen, Norweay, 1750–1900. International Review of Social History 50: 43–63. [Google Scholar] [CrossRef]
  6. Burchell, Julie. 2024. Separating Fact from Fiction in a World of AI Fairytales. NDC Conference London; Available online: https://www.youtube.com/watch?v=lLNJld729bc&pp=ygUNam9keSBidXJjaGVsbA%3D%3D (accessed on 20 July 2024).
  7. de Seta, Gabriele, Matti Pohjonen, and Aleksi Knuutila. 2024. Synthetic ethnography: Field devices for the qualitative study of generative models. Big Data & Society 11: 20539517241303126. [Google Scholar] [CrossRef]
  8. Foley, John Miles. 1991. Immanent Art. From Structure to Meaning in Traditional Oral Epic. Bloomington: Indiana University Press. [Google Scholar]
  9. Greimas, Algirdas. 1974. Strukturel Semantik. Copenhagen: Borgen. [Google Scholar]
  10. Hodne, Ørnulf. 1979. Jørgen Moe og Folkeeventyrene. En Studie i Nasjonalromantisk Folkloristikk. Oslo: Universitetsforlaget. [Google Scholar]
  11. Hodne, Ørnulf. 1998. Det Norske Folkeeventyret. Fra Folkediktning til Nasjonalkultur. Oslo: Cappelen Damm. [Google Scholar]
  12. Honko, Lauri. 2000. Text as process and practice: The textualization of oral epics. In Textualization of Oral Epics. Trends in Linguistics: Studies and Monographs, No. 128. Berlin and New York: Mouton de Gruyter. [Google Scholar]
  13. Iser, Wolfgang. 1978. The Act of Reading: A Theory of Aesthetic Response. Baltimore: Johns Hopkins University Press. [Google Scholar]
  14. Pilipets, Elena, and Marloes Geboers. 2025. Synthetic imaginaries of ‘sensitive AI’: On ambient amplification and jail(breaking) as method. Platforms & Society 2: 29768624251378110. [Google Scholar] [CrossRef]
  15. Rettberg, Jill Walker, and Herman Wigers. 2025. AI-generated stories favour stability over change: Homogeneity and cultural stereotyping in narratives generated by gpt-4o-mini. Open Research Europe 5: 202. [Google Scholar] [CrossRef]
  16. Uther, Hans-Jörg. 2004. The Types of International Folktales: A Classification and Bibliography, Based on the System of Antti Aarne and Satith Thompson. Helsinki: Suomalainen Tiedeakatemia. [Google Scholar]
  17. Zipes, Jack. 2012. The Irresistible Fairy Tale. The Cultural and Social History of a Genre. Princeton: Princeton University Press. [Google Scholar]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Refsum, A.S. “The Sweetheart in the Forest” and the Synthetic Storytellers. Humanities 2025, 14, 230. https://doi.org/10.3390/h14120230

AMA Style

Refsum AS. “The Sweetheart in the Forest” and the Synthetic Storytellers. Humanities. 2025; 14(12):230. https://doi.org/10.3390/h14120230

Chicago/Turabian Style

Refsum, Anne Sigrid. 2025. "“The Sweetheart in the Forest” and the Synthetic Storytellers" Humanities 14, no. 12: 230. https://doi.org/10.3390/h14120230

APA Style

Refsum, A. S. (2025). “The Sweetheart in the Forest” and the Synthetic Storytellers. Humanities, 14(12), 230. https://doi.org/10.3390/h14120230

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop