{"id":"3f0aaadc-843b-4c31-99de-2da294f204b4","arxiv_id":"1910.01441","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"Sentiment analysis of To the Lighthouse reveals an emotional arc distributed across characters, a pattern the authors call a distributed heroine model.","lead":"This paper applies off-the-shelf sentiment analysis to Virginia Woolf's To the Lighthouse and finds an emotional arc spread across characters, which it calls a distributed heroine. It argues that even a plotless modernist novel can show latent emotional structure when computational smoothing is combined with close reading.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Distributed-heroine conclusion rests on an untested 10% smoothing window; the 5% window is rejected post hoc, so the central claim is at risk.","rationale":"The reader's weakest_assumption exactly identifies the most load-bearing weakness: the paper's central conclusion depends on a smoothing window chosen because it produces that conclusion, while the alternative 5% window is dismissed on interpretive grounds. The authors provide no independent theoretical or empirical justification for 10%, and the word-salad control only demonstrates non-random sentiment, not the correct scale. The Syuzhet/VADER comparison and the close-reading checks are suggestive but do not settle the parameter question. Because the paper is an exploratory study and the finding is genuinely interesting if robust, the appropriate verdict remains CONDITIONAL rather than outright rejection. A pre-registered multi-window comparison with independent human valence ratings would convert the conditional claim into a testable one. No additional concerns rise to the same level: the paper's acknowledgment of criticism and its comparative use of multiple methods are honest and add credibility, but they do not fix the central circularity in window selection.","tokens_in":17968,"tokens_out":3373,"duration_ms":34220,"concrete_test":"Run a pre-registered robustness analysis: compute rolling-mean sentiment arcs for To the Lighthouse with Syuzhet.R at window sizes 5%, 7.5%, 10%, 12.5%, and 15% (plus a data-driven optimal window selected by cross-validated agreement with independent human valence ratings on a held-out sample of roughly 50 passages). Then extract local extrema for each window and have annotators blind to the hypothesis classify each extremum passage for valence plausibility and thematic coherence using the authors' own criteria. If the 10% window is not among the windows with the best agreement, or if the distributed-heroine reading fails to emerge at 7.5% or 12.5%, then the central claim is best treated as an artifact of the smoothing parameter.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that To the Lighthouse has an emotional structure distributed across characters, a 'distributed heroine.' The load-bearing premise is that the 10% rolling mean window is the right scale at which to capture readerly emotional experience. The paper's own account shows this premise is unsupported. In 'Analyzing the Rolling Mean Using Close Reading,' the authors write that 'we can only surmise' how to choose a filter and explicitly ask whether 5% would be better. They test 5%, find its extrema imply a 'failed love/marriage plot' centered on Lily and Mr. Bankes, and reject it on that basis. Since the 10% window was retained because it supports the distributed-heroine reading and the 5% window was rejected because it supports a less welcome reading, the conclusion is vulnerable to post-hoc parameter selection. The word-salad control in Figure G shows the sentiment signal is non-random, but it does not establish that 10% is the correct scale, nor does it distinguish the distributed-heroine interpretation from other possible arcs. Similarly, the Syuzhet/VADER comparison validates lexicon-level distributions, not the specific 10% window. The close-reading checks are performed by authors who know the intended interpretation, with no independent or pre-registered measure of 'coherence.' The claim could survive if a principled, outcome-independent reason for 10% were given, or if neighboring windows produced the same distributed structure; neither is currently supplied.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper applies lexical sentiment analysis (Syuzhet.R and VADER) to Virginia Woolf's To the Lighthouse, a modernist novel often described as plotless. It compares several smoothing techniques (LOESS, rolling mean, DCT) and reports broad agreement among them. The authors then perform a 'middle reading' that pairs rolling-mean inflection points with close reading, leading them to propose that the novel's emotional structure is distributed across characters, which they term a 'distributed heroine' model. They also benchmark Syuzhet against VADER, run a randomized word-salad control, and address earlier published critiques of Syuzhet in an appendix.","tokens_in":18405,"tokens_out":3583,"duration_ms":33649,"significance":"If the main claim were convincingly established, the paper would make a useful contribution to computational literary studies by showing that sentiment analysis can reveal an interpretable emotional structure in a novel that resists traditional event-based plot analysis, and by articulating a 'distributed heroine' concept that could inform modernist narrative theory. The paper has genuine strengths: it uses publicly available tools and texts, includes a randomized-control experiment (Figure G) that demonstrates the sentiment signal is not random, and transparently documents known limitations of the software. However, the central interpretive claim is currently supported chiefly by the authors' own close reading of a parameter choice that is itself justified by the interpretation it produces, so the significance of the finding is conditional on resolving that circularity.","major_comments":[{"comment":"The choice of the 10% rolling mean window is not justified by any independent theoretical, empirical, or reader-based criterion. The paper itself states that 'we can only surmise' how to choose a filter (p. 10), and when the 5% window is tested it is rejected because its inflection points suggest a 'failed love/marriage plot' rather than the distributed-heroine reading (p. 24). Because the 10% window is retained because it supports the distributed-heroine interpretation and the 5% window is dismissed because it does not, the central claim is vulnerable to post-hoc parameter selection. The authors need an outcome-independent justification for the window size, or a demonstration that neighboring windows (e.g., 7.5%, 12.5%) yield substantially the same distributed structure rather than only the single selected window.","section":"Analyzing the Rolling Mean Using Close Reading (pp. 12–25)"},{"comment":"The 'readerly' validation of the emotional arc rests entirely on the authors' own close reading of the selected inflection points P1–P22. There is no independent human annotation, no pre-registered definition of 'coherence,' and no quantitative measure of agreement between the computational arc and reader judgments. The same interpretive judgment that selects the 10% window is also used to confirm that the resulting arc is meaningful, which is circular. The word-salad control (Figure G) shows that the smoothed sentiment signal is distinguishable from random word order, but it does not show that the specific 10% window captures readerly experience or that the 'distributed heroine' interpretation is the correct one among the many arcs extractable at different scales.","section":"Validation throughout 'Analyzing the Rolling Mean Using Close Reading'"},{"comment":"The Syuzhet/VADER comparison is reported only qualitatively. The paper claims 'very similar distributions with very similar means, variances and slight negative skews' and, in Appendix A, 'almost identical frequency distribution,' but no numerical values, plots, or statistical tests are provided. Since the purpose of this comparison is to establish that Syuzhet's simple lexical approach is not statistically distorted by negation and intensifier errors, the reader cannot assess whether the claimed similarity is meaningful. Reporting summary statistics (means, variances, skewness) or a distance test (e.g., Kolmogorov-Smirnov) would make the methodological defense concrete.","section":"Comparing Models section (pp. 7–11) and Appendix A"}],"minor_comments":[{"comment":"The abstract's claim to be 'the first to undertake a hybrid model that fully leverages the strengths of both computational analysis and close reading' is unsupported by the cited literature and overlooks prior work that combines distant and close reading, including the paper's own footnote 4 reference to Laurie Taylor. Recommend softening the novelty claim.","section":"General"},{"comment":"In 'of it’s naive lexical approach,' 'it’s' should be 'its'. Also, on p. 3, 'more thorough statistical modelling' should likely be 'more thorough statistical modeling' (spelling aside, the phrase is awkward).","section":"p. 4"},{"comment":"The abbreviation 'LPS' is introduced as 'low pass filter (LPS)' but the standard term is LPF, and the abbreviation is used inconsistently (LPF appears elsewhere). Please standardize.","section":"p. 25"},{"comment":"The description of rolling-mean clipping is confusing: 'the first sentiment value is calculated at the 5% point as the mean value for the midpoint of the 10% sliding window.' A small diagram or a clearer formula would help readers understand the edge handling.","section":"p. 12"},{"comment":"The upper-bound sentence-splitting error analysis is a nice robustness check, but the sample sizes (56 sentences for Woolf, 72 for Dickens) are small; reporting a confidence interval or exact binomial bounds would strengthen the claim that the error rate is negligible.","section":"Appendix A"}],"recommendation":"major_revision","confidential_remarks":"The paper is an interesting and readable contribution to digital humanities, but its central 'distributed heroine' claim currently rests on a circular parameter choice. I would be willing to reconsider after the authors provide a stability analysis across window sizes, an independent justification for the 10% window, or reader-based validation. The novelty claim in the abstract is also stronger than the evidence warrants. The paper fits the journal's scope as a methodological and interpretive piece, but it needs the load-bearing revision described in the major comments."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper is a sincere, exploratory digital-humanities study. Its contribution is not the tools—Syuzhet, VADER, DCT, rolling means are all prior art—but the text-specific claim that To the Lighthouse has an emotional arc distributed across characters, a 'distributed heroine.' The close reading of inflection points is genuine literary work: the authors actually check whether the computational peaks and valleys make sense in context, and they find thematic patterns (separation vs. coherence, distance vs. connection) that dovetail with existing Woolf scholarship. The VADER/Syuzhet comparison is a useful sanity check for simple lexical sentiment analysis on this text, and the word-salad control in Figure G is a legitimate test against randomness.\n\nThe soft spot is the one the stress-test flags, and the paper deserves to be taken down a notch for it. The choice of a 10% rolling window is not independently justified; it is retained because it yields the distributed-heroine reading, while the 5% window is rejected because it yields a less welcome 'failed love/marriage plot.' The authors even say they can 'only surmise' how to choose a filter. That is post hoc parameter selection. The word-salad control shows the signal is not noise, but it does not establish that 10% is the correct scale, nor does it distinguish the distributed-heroine interpretation from other possible arcs. The close-reading checks are performed by people who know the intended interpretation, with no independent annotation or pre-registered criterion.\n\nThat said, the paper is more honest than many: it explicitly raises the possibility that 'we might be able to justify any underlying pattern,' and it acknowledges that 'more exploration surrounding this question is needed.' The flaw is real but proportionate—this is an exploratory case study, not a definitive proof. The authors also overstate novelty with the claim they are 'the first' to hybridize computation and close reading; that is not credible, though the distributed-heroine framing is new.\n\nWho is this for? Digital humanists working on sentiment analysis and narrative, and literary scholars curious about what computational methods can offer modernist studies. It deserves a serious referee, but the referee should ask for a more principled treatment of the smoothing window, ideally with a stability analysis across windows and, if possible, another novel. As it stands, the distributed-heroine is a suggestive hypothesis, not a well-supported discovery.","headline":"An honest exploratory DH paper with a suggestive reading of To the Lighthouse, but its central 'distributed heroine' claim rests on a smoothing window chosen after the fact.","tokens_in":18758,"tokens_out":1706,"would_cite":false,"duration_ms":18244,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Sentiment analysis reveals that Woolf's seemingly plotless novel To the Lighthouse carries an emotional arc distributed across characters rather than a single hero.","keywords":["sentiment analysis","Virginia Woolf","To the Lighthouse","modernist novel","emotional arc","distributed heroine","Syuzhet.R","close reading"],"falsifier":"Sweep the same pipeline over window sizes from 1 to 20 percent: if the distributed-heroine arc appears only at 10 percent and disappears at nearby windows, or if randomized shuffled versions of the novel sometimes produce arcs as structured as the original, the claim that the structure is inherent to To the Lighthouse is falsified. Alternatively, collect passage-level emotional ratings from human readers and test whether the 10 percent window's peaks and valleys predict those ratings better than the 5 percent window or random chance.","tokens_in":17780,"feed_emoji":"📈","tokens_out":10246,"duration_ms":83254,"temperature":0.7,"pith_summary":"This paper tries to establish that sentiment analysis can uncover emotional structure in a modernist novel that resists conventional plot analysis. Using Virginia Woolf's To the Lighthouse as the test case, the authors argue that the novel's emotional arc is not centered on one hero but is distributed across many characters, a pattern they call the distributed heroine model. They first defend the method by comparing the simple lexical tool Syuzhet.R with the more sophisticated VADER system, finding nearly identical sentiment distributions, and by showing that ten randomized reorderings of the novel produce noise from which the true sentiment plot stands apart. They then read the emotional highs and lows of a 10 percent rolling mean against the surrounding text and against recognized thematic patterns of connection and separation, arguing that the arc matches readerly experience. If this is right, sentiment analysis gives literary critics a new way to see structure in modernist fiction.","feed_headline":"Sentiment analysis finds a hidden arc in Woolf's 'plotless' novel","feed_subtitle":"The arc is carried by many characters together, not one hero — a 'distributed heroine' structure, the authors argue.","key_machinery":"The load-bearing mechanism is the smoothed emotional-valence curve produced by lexical sentiment analysis, specifically the Syuzhet.R package with its sliding rolling mean and DCT smoothing, cross-checked against VADER. A numeric sentiment score is computed for each sentence by summing the polarity values of words found in a sentiment lexicon, and a sliding window then averages these scores to expose an underlying arc buried in sentence-level noise; the paper settles on a 10 percent window, roughly 350 sentences, because it balances granularity against noise. The named identity that carries the interpretation is the distributed heroine model: the emotional peaks and valleys are not assigned to a single protagonist but emerge from multiple characters whose sentiments reinforce one another at moments of human connection and scatter at moments of separation. The 'middle reading' procedure, which compares each computational inflection point with its surrounding text and with known thematic patterns, is what turns the statistical curve into a literary claim.","core_discovery":"The paper's central finding is that To the Lighthouse, far from being emotionally unstructured, reveals an underlying emotional structure distributed between characters, which the authors name the distributed heroine model. The emotional valence of the narrative, measured sentence by sentence and smoothed with a 10 percent rolling mean, rises at moments of connection and coherence among characters, such as the dinner party, Paul and Minta's shared 'we,' and Lily's completed painting, and falls at moments of separation, distance, and dissolution, such as the extinguished lamps, Mrs. Ramsay's death reported in brackets, and Cam's loss of the summer house. The authors argue that this arc is not a statistical artifact: ten randomized word salads of the novel produce a noisy band from which the original sentiment plot clearly stands apart, and the VADER comparison yields nearly identical distributions. They further claim that a 5 percent window would reduce the novel to a failed love-and-marriage plot centered on Lily and Mr. Bankes, while the 10 percent window captures the larger cast of characters and thereby matches a reader's experience of many perspectives reinforcing a shared emotional arc. As the novel progresses the arc becomes more coherent, mirroring at the level of emotion the movement from chaos to pattern that literary scholars such as Kern have traced in Woolf's themes.","pith_inferences":["If the model generalizes, other multi-perspective modernist novels such as Mrs Dalloway or Ulysses should show similarly distributed emotional arcs rather than single-hero arcs; running the same pipeline on those texts would test that prediction.","The 10 percent window is validated in the paper mainly by the coherence of the reading it produces; an independent anchor, such as continuous reader-response ratings aligned with the arc, would be needed to separate the choice of scale from the interpretation.","A sharper test of the method would use the 10 percent arc to predict which passages readers rate as emotional peaks and valleys, rather than checking inflection points against close reading after the fact.","The striking difference between the 5 percent and 10 percent readings suggests that smoothing scale is itself an interpretive choice in computational literary studies, not merely a technical detail."],"forward_implications":["If the claim is right, sentiment analysis can expose emotional structure in modernist novels that appear plotless by the usual action-event-causality standards.","The distributed heroine model implies that fragmented narrative perspective need not produce fragmented emotion; multiple characters can jointly carry a unified emotional arc.","The VADER comparison supports the practical conclusion that simple lexical sentiment tools are adequate for novel-length literary texts, making the method accessible to critics without deep NLP resources.","The window-size dependence shows that smoothing parameters change the interpretive outcome, so comparative model analysis must accompany any such claim.","The alignment of the arc with connection-versus-separation themes gives computational support to close-reading accounts of To the Lighthouse's pattern emerging from chaos."],"supporting_citations":[{"why":"The open-source R package that provides the lexical sentiment scoring and the DCT and rolling-mean smoothing curves central to the analysis.","marker":"Syuzhet.R"},{"why":"Supplies the emotional-arc framework and the randomized-text control technique the paper adapts to show Woolf's arc is not noise.","marker":"Reagan et al."},{"why":"The more sophisticated rule-based sentiment model used as a benchmark; its near-identical sentiment distributions validate the simpler Syuzhet approach.","marker":"VADER"},{"why":"The close-reading account of fragmentation and unification in Woolf that the paper uses to corroborate the thematic pattern of the emotional arc.","marker":"Kern"},{"why":"Provides the 'transparent minds' concept that lets the authors treat characters' reported emotions as legitimate readerly experience.","marker":"Cohn"},{"why":"Her published critiques of Syuzhet.R are the objections the paper tests and addresses in an appendix, confirming the tool's reliability for literary texts.","marker":"Swafford"}],"fun_headline_variants":["Woolf's plotless novel hides a distributed emotional arc","Sentiment analysis reveals shared emotional structure in Woolf's novel","Distributed heroine: the hidden emotional arc in Woolf's novel","To the Lighthouse's emotional arc is shared, not singular","No hero, but a distributed emotional arc in Woolf's novel"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The central claim rests on the choice of a 10 percent rolling mean as the smoothing scale that reveals the arc: a 5 percent window produces a different, less coherent reading, and the paper justifies 10 percent mainly by the fact that it yields the distributed-heroine interpretation rather than by an independent measure of reader experience.","fun_headline_variants_meta":{"raw":{"variants":["Woolf's plotless novel hides a distributed emotional arc","Sentiment analysis reveals shared emotional structure in Woolf's novel","Distributed heroine: the hidden emotional arc in Woolf's novel","To the Lighthouse's emotional arc is shared, not singular","No hero, but a distributed emotional arc in Woolf's novel"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000666,"raw_usage":{"total_tokens":3074,"prompt_tokens":1012,"completion_tokens":2062,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":628,"completion_tokens_details":{"reasoning_tokens":1978}},"tokens_in":628,"tokens_out":2062,"duration_ms":44447,"temperature":1.0,"reasoning_tokens":1978,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T05:57:20.575836+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Sweep the same pipeline over window sizes from 1 to 20 percent: if the distributed-heroine arc appears only at 10 percent and disappears at nearby windows, or if randomized shuffled versions of the novel sometimes produce arcs as structured as the original, the claim that the structure is inherent to To the Lighthouse is falsified. Alternatively, collect passage-level emotional ratings from human readers and test whether the 10 percent window's peaks and valleys predict those ratings better than the 5 percent window or random chance.","supporting_citations":[],"review_version":1}