Tristan Miller

Austrian Research Institute for Artificial Intelligence · Freyung 6/6 · 1010 Vienna · Austria
+43 1 5324621 3 tristan@logological.org ()

I'm a computational linguist with research interests in lexical semantics, historical online corpora, and computational detection and interpretation of humour. I currently head the Computational Pun-derstanding: Computer-Assisted Translation of Humorous Wordplay project at the Austrian Research Institute for Artificial Intelligence (OFAI).



Publications

Waltraud Kolb and Tristan Miller.
Human–computer interaction in pun translation.
In James Hadley, Kristiina Taivalkoski-Shilov, Carlos S. C. Teixeira, and Antonio Toral, editors, Using Technologies for Creative-Text Translation. Routledge, 2022. To appear.
We present and evaluate PunCAT, an interactive electronic tool for the translation of puns. Following the strategies known to be applied in pun translation, PunCAT automatically translates each sense of the pun separately; it then allows the user to explore the semantic fields of these translations in order to help construct a plausible target-language solution that maximizes the semantic correspondence to the original. Our evaluation is based on an empirical pilot study in which the participants translated puns from a variety of published sources from English into German, with and without PunCAT. We aimed to answer the following questions: Does the tool support, improve, or constrain the translation process, and if so, in what ways? And what are the tool's main benefits and drawbacks as perceived and described by the participants? Our analysis of the translators' cognitive processes gives us insight into their decision-making strategies and how they interacted with the tool. We find clear evidence that PunCAT effectively supports the translation process in terms of stimulating brainstorming and broadening the translator's pool of solution candidates. We have also identified a number of directions in which the tool could be adapted to better suit translators' work processes.
@incollection{kolb2022human,
author       = {Waltraud Kolb and Tristan Miller},
editor       = {James Hadley and Kristiina Taivalkoski-Shilov and Carlos S. C. Teixeira and Antonio Toral},
title        = {Human--Computer Interaction in Pun Translation},
booktitle    = {Using Technologies for Creative-Text Translation},
year         = {2022},
publisher    = {Routledge},
note         = {To appear},
}
Jörg Wöckener, Thomas Haider, Tristan Miller, The-Khang Nguyen, Thanh Tung Linh Nguyen, Minh Vu Pham, Jonas Belouadi, and Steffen Eger.
End-to-end style-conditioned poetry generation: What does it take to learn from examples alone?
In Proceedings of the 5th Joint SIGHUM Workshop on Computational Linguistics for Cultural Heritage, Social Sciences, Humanities and Literature (LaTeCH-CLfL 2021), November 2021. To appear.
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.
@inproceedings{woeckener2021end,
author       = {J{\"{o}}rg W{\"{o}}ckener and Thomas Haider and Tristan Miller and The-Khang Nguyen and Thanh Tung Linh Nguyen and Minh Vu Pham and Jonas Belouadi and Steffen Eger},
title        = {End-to-end Style-Conditioned Poetry Generation: {What} Does It Take to Learn from Examples Alone?},
booktitle    = {Proceedings of the 5th Joint SIGHUM Workshop on Computational Linguistics for Cultural Heritage, Social Sciences, Humanities and Literature (LaTeCH-CLfL 2021)},
month        = nov,
year         = {2021},
note         = {To appear},
}
Alexandra Uma, Tommaso Fornaciari, Anca Dumitrache, Tristan Miller, Jon Chamberlain, Barbara Plank, Edwin Simpson, and Massimo Poesio.
SemEval-2021 Task 12: Learning with disagreements.
In Proceedings of the 15th International Workshop on Semantic Evaluation (SemEval-2021), pages 338–347, August 2021. ISBN 978-1-954085-70-1. DOI: 10.18653/v1/2021.semeval-1.41.
Disagreement between coders is ubiquitous in virtually all datasets annotated with human judgements in both natural language processing and computer vision. However, most supervised machine learning methods assume that a single preferred interpretation exists for each item, which is at best an idealization. The aim of the SemEval-2021 shared task on Learning with Disagreements (Le-wi-Di) was to provide a unified testing framework for methods for learning from data containing multiple and possibly contradictory annotations covering the best-known datasets containing information about disagreements for interpreting language and classifying images. In this paper we describe the shared task and its results.
@inproceedings{uma2021semeval,
author       = {Alexandra Uma and Tommaso Fornaciari and Anca Dumitrache and Tristan Miller and Jon Chamberlain and Barbara Plank and Edwin Simpson and Massimo Poesio},
title        = {{SemEval}-2021 {Task}~12: Learning with Disagreements},
booktitle    = {Proceedings of the 15th International Workshop on Semantic Evaluation (SemEval-2021)},
pages        = {338--347},
month        = aug,
year         = {2021},
isbn         = {978-1-954085-70-1},
doi          = {10.18653/v1/2021.semeval-1.41},
}
Tristan Miller.
Dmitri Borgmann's rotas square articles.
Notes and Queries, 67(3):431–432, September 2020. ISSN 0029-3970. DOI: 10.1093/notesj/gjaa113.
In 1979 and 1980, Word Ways: The Journal of Recreational Linguistics printed a series of articles on the early history, religious symbolism, and cultural significance of the rotas square, an ancient Latin-language palindromic word square. The articles were attributed to Dmitri A. Borgmann, the noted American writer on wordplay and former editor of Word Ways. While they attracted little attention at the time, some 35 years after their publication (and 29 years after Borgmann's death), questions began to be raised about their authorship. There is much internal and external evidence that, taken together, compellingly supports the notion that Borgmann did not write the articles himself. This paper surveys this evidence and solicits help in identifying the articles' original source.
@article{miller2020dmitri,
author       = {Tristan Miller},
title        = {{Dmitri Borgmann's} Rotas Square Articles},
journal      = {Notes and Queries},
volume       = {67},
number       = {3},
pages        = {431--432},
month        = sep,
year         = {2020},
issn         = {0029-3970},
doi          = {10.1093/notesj/gjaa113},
}
Tristan Miller and Denis Auroux.
GPP, the generic preprocessor.
Journal of Open Source Software, 5(51), July 2020. ISSN 2475-9066. DOI: 10.21105/joss.02400.
In computer science, a preprocessor (or macro processor) is a tool that programatically alters its input, typically on the basis of inline annotations, to produce data that serves as input for another program. Preprocessors are used in software development and document processing workflows to translate or extend programming or markup languages, as well as for conditional or pattern-based generation of source code and text. Early preprocessors were relatively simple string replacement tools that were tied to specific programming languages and application domains, and while these have since given rise to more powerful, general-purpose tools, these often require the user to learn and use complex macro languages with their own syntactic conventions. In this paper, we present GPP, an extensible, general-purpose preprocessor whose principal advantage is that its syntax and behaviour can be customized to suit any given preprocessing task. This makes GPP of particular benefit to research applications, where it can be easily adapted for use with novel markup, programming, and control languages.
@article{miller2020gpp,
author       = {Tristan Miller and Denis Auroux},
title        = {{GPP}, the Generic Preprocessor},
journal      = {Journal of Open Source Software},
volume       = {5},
number       = {51},
month        = jul,
year         = {2020},
issn         = {2475-9066},
doi          = {10.21105/joss.02400},
}
Tristan Miller.
Reinhold Aman, 1936–2019.
Humor: International Journal of Humor Research, 32(1):1–5, February 2020. ISSN 0933-1719. DOI: 10.1515/humor-2019-0085.
@article{miller2020reinhold,
author       = {Tristan Miller},
title        = {Reinhold {Aman}, 1936--2019},
journal      = {Humor: International Journal of Humor Research},
volume       = {32},
number       = {1},
pages        = {1--5},
month        = feb,
year         = {2020},
issn         = {0933-1719},
doi          = {10.1515/humor-2019-0085},
}
Edwin Simpson, Erik-Lân Do Dinh, Tristan Miller, and Iryna Gurevych.
Predicting humorousness and metaphor novelty with Gaussian process preference learning.
In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics (ACL 2019), pages 5716–5728, July 2019. ISBN 978-1-950737-48-2. DOI: 10.18653/v1/P19-1572.
The inability to quantify key aspects of creative language is a frequent obstacle to natural language understanding. To address this, we introduce novel tasks for evaluating the creativeness of language---namely, scoring and ranking text by humorousness and metaphor novelty. To sidestep the difficulty of assigning discrete labels or numeric scores, we learn from pairwise comparisons between texts. We introduce a Bayesian approach for predicting humorousness and metaphor novelty using Gaussian process preference learning~(GPPL), which achieves a Spearman's~$\rho$ of 0.56 against gold using word embeddings and linguistic features. Our experiments show that given sparse, crowdsourced annotation data, ranking using GPPL outperforms best--worst scaling. We release a new dataset for evaluating humor containing 28,210 pairwise comparisons of 4,030 texts, and make our software freely available.
@inproceedings{simpson2019predicting,
author       = {Edwin Simpson and Do Dinh, Erik-L{\^{a}}n and Tristan Miller and Iryna Gurevych},
title        = {Predicting Humorousness and Metaphor Novelty with {Gaussian} Process Preference Learning},
booktitle    = {Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics (ACL 2019)},
pages        = {5716--5728},
month        = jul,
year         = {2019},
isbn         = {978-1-950737-48-2},
doi          = {10.18653/v1/P19-1572},
}

Projects

Funded research projects

Events & organizations

Software

Publishing & documentation


Miscellany

My interests in language, math, and computers were sparked and strengthened by exposure to the works of Willard R. Espy, Louis Phillips, Mike Keith, Dmitri Borgmann, Jim Butterfield, and others. These writers share a great talent for making technical or linguistic topics fun and accessible to a general audience. You can check out my own contributions to popular and recreational mathematics and linguistics, plus a few other odds and ends.

I also maintain an index of miscellaneous documents and websites I've produced which don't really fit into any other section.