pith. sign in

arxiv: 1906.08733 · v1 · pith:XIBP4CUFnew · submitted 2019-06-20 · 💻 cs.CL · cs.AI

Autonomous Haiku Generation

Rui Aguiar , Kevin Liao This is my paper

Pith reviewed 2026-05-25 19:29 UTC · model grok-4.3

classification 💻 cs.CL cs.AI
keywords haiku generationdeep learningpoetry generationartificial intelligencenatural language generationcreative computingtext synthesis
0
0 comments X

The pith

Deep learning models trained on haiku data can generate new haikus that humans judge to be of high quality.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper sets out to apply artificial intelligence and deep learning to the task of generating haikus autonomously. It starts from the observation that poetry is language at its most distilled and powerful, then treats haiku creation as a learnable creative process rather than one requiring explicit rules. The central effort is to train models on existing haiku examples so they can produce new ones without further human guidance. A sympathetic reader would care because the work tests whether quantitative methods can reach into domains traditionally seen as intuitive and rule-free. If the models succeed, creative writing joins the list of tasks where data-driven systems can operate independently.

Core claim

The paper claims that deep learning can be used to generate high quality haikus by training models on haiku data, thereby extending AI from quantitative tasks to creative ones that are difficult to define precisely.

What carries the argument

A deep learning model for text generation trained on a dataset of haikus.

If this is right

  • Creative tasks like poetry become feasible targets for autonomous AI systems.
  • Training on example poems allows models to internalize syllable and structural constraints without manual encoding.
  • Human judgment serves as the final validation metric for machine-generated creative output.
  • The same training approach could transfer to other short poetic forms.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Poetic form may be learnable as statistical patterns rather than requiring explicit symbolic rules.
  • Generated haikus could serve as starting points for human poets or as teaching examples.
  • Success here would prompt similar experiments on longer or less structured creative writing tasks.

Load-bearing premise

That a deep learning model trained on haiku data will produce outputs that humans judge as high quality poetry.

What would settle it

A side-by-side human evaluation study in which experts rate the generated haikus as consistently lower in quality than human-written haikus on standard poetry assessment scales.

Figures

Figures reproduced from arXiv: 1906.08733 by Kevin Liao, Rui Aguiar.

Figure 1
Figure 1. Figure 1: With each iteration, the model becomes better at predicting the similarity between words, and the difference in similarity goes down. After training this cost function, the costs in our Beam search were the absolute value of the difference between the similarity of the candidate word and what our model predicted the similarity of the next word to be. Our Beam search can be described mathematically as Each … view at source ↗
Figure 2
Figure 2. Figure 2: [4] [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: Figure 5 [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
read the original abstract

Artificial Intelligence is an excellent tool to improve efficiency and lower cost in many quantitative real world applications, but what if the task is not easily defined? What if the task is generating creativity? Poetry is a creative endeavor that is highly difficult to both grasp and achieve with any level of competence. As Rita Dove, a famous American poet and author states, "Poetry is language at its most distilled and most powerful." Taking Doves quote as an inspiration, our task was to generate high quality haikus using artificial intelligence and deep learning.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 1 minor

Summary. The manuscript states that its task is to generate high-quality haikus using artificial intelligence and deep learning, taking inspiration from Rita Dove's description of poetry as 'language at its most distilled and most powerful.' No further details are supplied.

Significance. If the central claim were substantiated with working methods and evidence, the work would address an interesting application of deep learning to computational creativity. In its current form, however, the absence of any technical content means no significance can be assessed.

major comments (1)
  1. [Abstract] Abstract: the claim that high-quality haikus were generated via deep learning is unsupported because the manuscript supplies neither model architecture, training data, loss function, training procedure, sample outputs, nor any quality metric (human or automatic).
minor comments (1)
  1. [Abstract] The possessive in 'Rita Dove, a famous American poet and author states, ... Taking Doves quote' should read 'Dove's quote'.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the report. We agree that the submitted manuscript provides no technical details, model specifications, data, training procedure, outputs, or evaluation, rendering the central claim unevaluable.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that high-quality haikus were generated via deep learning is unsupported because the manuscript supplies neither model architecture, training data, loss function, training procedure, sample outputs, nor any quality metric (human or automatic).

    Authors: The referee is correct. The manuscript contains only the high-level motivation and contains none of the listed elements. No architecture, corpus, loss, procedure, samples, or metrics are present. We will either withdraw the manuscript or prepare a substantially revised version that supplies these components if the underlying experiments exist; otherwise the work cannot be substantiated. revision: yes

Circularity Check

0 steps flagged

No derivation chain present; claim is unsupported but exhibits no circularity

full rationale

The manuscript provides only an abstract stating the intent to generate haikus via deep learning, with no model architecture, equations, training procedure, outputs, metrics, or any quantitative derivation. No load-bearing steps exist that could reduce to inputs by construction, self-citation, or fitted parameters. The work therefore contains no derivation chain to analyze, making circularity score 0 by definition.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No technical content is supplied, so the ledger cannot enumerate specific free parameters or axioms; the central claim rests on the unstated assumption that standard deep learning will suffice for high-quality constrained poetry.

pith-pipeline@v0.9.0 · 5596 in / 838 out tokens · 18340 ms · 2026-05-25T19:29:43.400203+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

9 extracted references · 9 canonical work pages

  1. [1]

    Rita Dove

    Lannan Foundation. Rita Dove. Lannan Foundation, lannan.org/bios/rita-dove

    work page

  2. [2]

    Ballas, Sam. PoetRNN. Github, 1 Aug. 2015, github.com/sballas8/PoetRNN/tree/master/data

    work page 2015

  3. [3]

    haiku generation

    Aguiar, Rui. haiku generation. GitHub, 7 Dec. 2017, github.com/raguiar2/haiku generation

    work page 2017

  4. [4]

    How Does LSTM Cell Map to Layers? Stack Overflow, Aug

    Stack Overflow. How Does LSTM Cell Map to Layers? Stack Overflow, Aug. 2017, stackoverflow.com/questions/45223467/how- does-lstm-cell-map-to-layers

    work page arXiv 2017

  5. [5]

    Getting Started with the Keras Sequential Model

    Keras. Getting Started with the Keras Sequential Model. Guide to the Sequential Model - Keras Docu- mentation, keras.io/getting-started/sequential-model-guide/

    work page

  6. [6]

    Using AI to Generate Lyrics Ivan Liljeqvist Medium

    Liljeqvist, Ivan. Using AI to Generate Lyrics Ivan Liljeqvist Medium. Medium, Medium, 5 Dec. 2016, medium.com/@ivanliljeqvist/using-ai-to-generate-lyrics-5aba7950903

    work page 2016

  7. [7]

    An evolutionary algorithm approach to poetry generation

    Manurung, Hisar Maruli. An evolutionary algorithm approach to poetry generation. University of Edin- burgh, Institute for Communicating and Collaborative Systems, 2013, www.inf.ed.ac.uk/publications/thesis/online/IP040022.pdf

    work page 2013

  8. [8]

    Welcome to Magenta

    Magenta. Welcome to Magenta. Magenta, 1 June 2016, magenta.tensorflow.org/welcome-to-magenta

    work page 2016

  9. [9]

    What’s a Haiku

    North Carolina Haiku Society. What’s a Haiku. North Carolina Haiku Society, nc-haiku.org/whats-a- haiku/. 10

    work page