This paper introduces a large annotated corpus of public domain German poetry, covering the time period from 1600 to the 1920s with 65k poems. We describe how the corpus was compiled, how it was cleaned (including duplicate detection), and how it looks now in terms of size, format, temporal distribution, and automatic annotation. Besides metadata, the corpus contains reliable annotation of tokens, syllables, part-of-speech, and meter and verse measure. Finally, we give some statistics on the annotation and an overview of other poetry corpora.

A prerequisite for the computational study of literature is the availability of properly digitized texts, ideally with reliable meta-data and ground-truth annotation. Poetry corpora do exist for a number of languages, but larger collections lack consistency and are encoded in various standards, while annotated corpora are typically constrained to a particular genre and/or were designed for the analysis of certain linguistic features (like rhyme). In this work, we provide large poetry corpora for English and German, and annotate prosodic features in smaller corpora to train corpus driven neural models that enable robust large scale analysis. We show that BiLSTM-CRF models with syllable embeddings outperform a CRF baseline and different BERT-based approaches. In a multi-task setup, particular beneficial task relations illustrate the inter-dependence of poetic features. A model learns foot boundaries better when jointly predicting syllable stress, aesthetic emotions and verse measures benefit from each other, and we find that caesuras are quite dependent on syntax and also integral to shaping the overall measure of the line.

In this work, we design an end-to-end model for poetry generation based on conditioned recurrent neural network (RNN) language models whose goal is to learn stylistic features (poem length, sentiment, alliteration, and rhyming) from examples alone. We show this model successfully learns the ‘meaning’ of length and sentiment, as we can control it to generate longer or shorter as well as more positive or more negative poems. However, the model does not grasp sound phenomena like alliteration and rhyming, but instead exploits low-level statistical cues. Possible reasons include the size of the training data, the relatively low frequency and difficulty of these sublexical phenomena as well as model biases. We show that more recent GPT-2 models also have problems learning sublexical phenomena such as rhyming from examples alone.

This paper describes the system Clustering on Manifolds of Contextualized Embeddings (CMCE) submitted to the SemEval-2020 Task 1 on Unsupervised Lexical Semantic Change Detection. Subtask 1 asks to identify whether or not a word gained/lost a sense across two time periods. Subtask 2 is about computing a ranking of words according to the amount of change their senses underwent. Our system uses contextualized word embeddings from MBERT, whose dimensionality we reduce with an autoencoder and the UMAP algorithm, to be able to use a wider array of clustering algorithms that can automatically determine the number of clusters. We use Hierarchical Density Based Clustering (HDBSCAN) and compare it to Gaussian MixtureModels (GMMs) and other clustering algorithms. Remarkably, with only 10 dimensional MBERT embeddings (reduced from the original size of 768), our submitted model performs best on subtask 1 for English and ranks third in subtask 2 for English. In addition to describing our system, we discuss our hyperparameter configurations and examine why our system lags behind for the other languages involved in the shared task (German, Swedish, Latin). Our code is available at https://github.com/DavidRother/semeval2020-task1

Most approaches to emotion analysis of social media, literature, news, and other domains focus exclusively on basic emotion categories as defined by Ekman or Plutchik. However, art (such as literature) enables engagement in a broader range of more complex and subtle emotions. These have been shown to also include mixed emotional responses. We consider emotions in poetry as they are elicited in the reader, rather than what is expressed in the text or intended by the author. Thus, we conceptualize a set of aesthetic emotions that are predictive of aesthetic appreciation in the reader, and allow the annotation of multiple labels per line to capture mixed emotions within their context. We evaluate this novel setting in an annotation experiment both with carefully trained experts and via crowdsourcing. Our annotation with experts leads to an acceptable agreement of k = .70, resulting in a consistent dataset for future large scale analysis. Finally, we conduct first emotion classification experiments based on BERT, showing that identifying aesthetic emotions is challenging in our data, with up to .52 F1-micro on the German subset. Data and resources are available at https://github.com/tnhaider/poetry-emotion.

Due to its semantic succinctness and novelty of expression, poetry is a great test-bed for semantic change analysis. However, so far there is a scarcity of large diachronic corpora. Here, we provide a large corpus of German poetry which consists of about 75k poems with more than 11 million tokens, with poems ranging from the 16th to early 20th century. We then track semantic change in this corpus by investigating the rise of tropes (‘love is magic’) over time and detecting change points of meaning, which we find to occur particularly within the German Romantic period. Additionally, through self-similarity, we reconstruct literary periods and find evidence that the law of linear semantic change also applies to poetry.

We present the first supervised approach to rhyme detection with Siamese Recurrent Networks (SRN) that offer near perfect performance (97% accuracy) with a single model on rhyme pairs for German, English and French, allowing future large scale analyses. SRNs learn a similarity metric on variable length character sequences that can be used as judgement on the distance of imperfect rhyme pairs and for binary classification. For training, we construct a diachronically balanced rhyme goldstandard of New High German (NHG) poetry. For further testing, we sample a second collection of NHG poetry and set of contemporary Hip-Hop lyrics, annotated for rhyme and assonance. We train several high-performing SRN models and evaluate them qualitatively on selected sonnetts.

While there is wide acknowledgement in NLP of the utility of document characterization by genre, it is quite difficult to determine a definitive set of features or even a comprehensive list of genres. This paper addresses both issues. First, with prototype semantics, we develop a hierarchical taxonomy of discourse functions. We implement the taxonomy by developing a new text genre corpus of contemporary German to perform a text based comparative register analysis. Second, we extract a host of style features, both deep and shallow, aiming beyond linguistically motivated features at situational correlates in texts. The feature sets are used for supervised text genre classification, on which our models achieve high accuracy. The combination of the corpus typology and feature sets allows us to characterize types of communicative purpose in a comparative setup, by qualitative interpretation of style feature loadings of a regularized discriminant analysis. Finally, to determine the dependence of genre on topics (which are arguably the distinguishing factor of sub-genre), we compare and combine our style models with Latent Dirichlet Allocation features across different corpus settings with unstable topics.

We describe a systematic and application-oriented approach to training and evaluating named entity recognition and classification (NERC) systems, the purpose of which is to identify an optimal system and to train an optimal model for named entity tagging DeReKo, a very large general-purpose corpus of contemporary German (Kupietz et al., 2010). DeReKo ‘s strong dispersion wrt. genre, register and time forces us to base our decision for a specific NERC system on an evaluation performed on a representative sample of DeReKo instead of performance figures that have been reported for the individual NERC systems when evaluated on more uniform and less diverse data. We create and manually annotate such a representative sample as evaluation data for three different NERC systems, for each of which various models are learnt on multiple training data. The proposed sampling method can be viewed as a generally applicable method for sampling evaluation data from an unbalanced target corpus for any sort of natural language processing.