Technodiscursive Analysis of Twitterbot Poetry

The paper situates Twitterbot poetry within the broader ecology of digital-era textuality characterized by hypertext, algorithms, generativity, and multimodality. On platforms like Twitter, literary micro-texts (tweet-poems/twittexts) combine text with emojis and other media. The study targets Leonardo Flores’ protest Twitterbot @Protestitas (Spanish version of @TinyProtests), active 2018–2021, which engaged with issues such as Hurricane Maria, Black Lives Matter, and COVID-19. The central purpose is to understand how a Twitterbot produces its poetic output as executed code and how the tweet-poems should be read. The authors frame this as a technodiscursive problem and propose a Four-Dimensional Analysis (4DAs) integrating technology, semiotics, art, and language to interpret the relationship between bot-generated text, emojis, and code, and to assess the bot’s socio-political communication and authorial intention.

The paper traces Twitterbot literature to algorithmic and combinatorial traditions in 20th-century movements such as OuLiPo and A.L.A.M.O., and to practices like Dadaist cut-ups, Surrealist techniques, and ALIRE’s codified poetics. It also references contemporary generative literature and uncreative writing. Scholars (Parrish, Veale et al., Jhave) argue that bots are capable of political innovation and creativity similar to human artists. Within the tiny-bot ecosystem hosted by Cheap Bots, Done Quick (CBDQ), most bots generate immersive, apolitical scenes; @Protestitas is distinctive for explicit political engagement. The review notes the shutdown of CBDQ’s large-scale API access in 2023 and situates emoji use within historical pictographic systems and contemporary semiotics (e.g., emoji as pictograms with systematic structure). It also invokes debates on technodiscourse and the author’s intention encoded in devices and code (Paveau, Bootz), and positions Twitterbot authorship in light of Couffignal’s robot poet test and meta-author theories (Lem, Balpe).

The study employs a technodiscursive analysis (4DAs), integrating: (1) technology of discourse (Twitter platform constraints, CBDQ infrastructure, Tracery/JSON grammar), (2) semiotics (emojis, spaces, punctuation), (3) art (visual composition/layout of emojis, ASCII), and (4) language (twittexts/slogans). It focuses primarily on composite tweets (text plus emojis/spacing) from @Protestitas and analyzes ten representative excerpts. Critical code studies are applied to the bot’s source code (via CBDQ/Tracery) to connect variables, values, and replacement grammar to on-screen outputs. The Tracery replacement grammar and JSON structure are examined: variables (e.g., origin → #protesttype#) expand into emoji-configurations (#infront#, #blockedgates#, etc.) and slogans (#featuredslogans#), with control characters (e.g., \n) shaping spatial layout. The methodology compares source code structures, emoji groupings, and displayed micro-texts to decipher meaning, reading tweet-poems as rebuses whose semantics emerge from the interplay of code, emojis, and text over time-bound, stochastic generations.

• The on-screen tweet-poems and the bot’s source code share the same “literary spark,” encoding coherent political, cultural, and socio-economic messages across both layers.
• Emoji functions are multifold: (1) independent language (visual grammar needing translation), (2) referents to twittexts (illustrating or anchoring slogans), (3) complements (amplifying or specifying textual claims), and (4) interlingual translations (transposing text content into icon sequences).
• Spatial layout and blank spaces are semantically active (“icono-/space-to-be-seen”), contributing to meaning and highlighting gaps (e.g., masses vs. government/police protection of elite spaces).
• Case analyses (e.g., “¡QUE LA DEUDA LA PAGUEN LOS RICOS!”, “¡NO A LA DEFORMA EDUCATIVA!”, “¡CERO SUBCONTRATOS!”, “¡NO MAS CORRUPCION!”) show consistent mappings of urban institutions (🏛️, 🏢, 🏟️) and security/emergency services (👮, 🕵, 🚓, 🚑) to critiques of neoliberalism, subcontracting, educational deformation, and corruption.
• Emoji assemblages are polysemous and culturally situated; their arrangement (rows/columns, proximity/distance) indexes power relations (e.g., leaders/weather symbols above, masses below; racialized distributions among police/detective emojis).
• Temporal variation (“spatiotemporal dynamics of display”) changes expression and content: repeated slogans with different emoji arrays modulate the poem’s semantics across generations.
• Source code inventory and composition: approximately 70 variables; 1,005 emojis and 225 English–Spanish slogans (1,230 values). The broader database includes 964 Anglo-Hispanic lexicons (512 Spanish, 452 English), totaling 1,969 icono-micro-text elements (1,005 emojis + 964 lexicons), numerically linked by the authors to 1969 Puerto Rican student protests.
• Language distribution in slogans: 125 Spanish (55.6%) vs. 100 English (44.4%), suggesting Spanish dominance in content while algorithmic logic is in English, reflecting PR–US sociolinguistic relations.
• Face-emoji demographics: 625 activist faces, with 170 representing women (≈27.2%), implying women’s underrepresentation in public engagement relative to men in the dataset.
• Output cadence: designed to post at 1–3 hour intervals (often cited as ~every 3 hours), implying about eight poems per day and up to 8,760 per year (as reported by the authors), contingent on server/device/network conditions.
• Urban focus: Code object categories (buildings, government facilities, university, stadiums, etc.) emphasize urban protest geographies; rural signifiers are largely absent in the codebase.
• Overall, @Protestitas functions as a protest bot aligned with anti-racist, anti-fascist, anti-neoliberal discourses, with Google Translate facilitating multilingual display (e.g., Spanish to French), and emoji-text composites performing political critique.

The findings address the research questions by demonstrating how @Protestitas produces its outputs through Tracery-based replacement grammar and JSON-structured variables that deterministically/stochastically assemble emoji layouts and slogans. Reading the tweet-poems as technodiscourse reveals that meaning arises from the coordinated action of platform constraints, code structures, emojis, and micro-texts—captured in the 4DAs framework. The equivalence of “literary spark” between code and display confirms that authorial intention (meta-poet/programmer-artist) is embedded in both the data structures and the generative process. Emoji configurations govern semantic framing (independent language, referent, complement, translation), making visible power asymmetries (state/security, elite institutions, racialized bodies) and highlighting urban protest geographies. Temporal regeneration alters expression/content, enabling synecdochic address from Puerto Rico and the US to global audiences. The language mix (Spanish content, English algorithmic scaffolding) mirrors PR–US sociopolitical and cultural entanglements. Overall, the bot evidences that Twitterbot poetry is a viable medium for socio-political communication and digital activism, with code studies providing a precise lens to recover intention and interpretive nuance.

The study introduces and applies a Four-Dimensional Analysis (technology, semiotics, art, language) to Twitterbot poetry, showing that @Protestitas’ source code and on-screen outputs share coherent political aesthetics. By coupling technodiscursive analysis with critical code studies, the authors demonstrate how emojis, blank spaces, and micro-texts co-produce meaning, and how replacement grammar encodes authorial intention and protest configurations. They conclude that @Protestitas effectively contributed to sustained digital activism (2018–2021), aligning against right-wing politics and “trumpism,” and recommend Twitterbots as tools for social good in socio-political discourse.

The authors explicitly limit the analysis to composite tweets (tweets complemented with emojis) and to the technology of discourse (platform, code, display), not to the full range of interactive affordances. The bot’s non-interactive, logocentric discourse contrasts with typical interactive digital media. Findings are also contingent on CBDQ/Twitter infrastructure and device/browser/network variability. Additionally, the codebase emphasizes urban geographies; rural contexts are not represented, which may constrain the generalizability of spatial interpretations.