Deconstructing Nagaru Tanigawa (ruminations of style)

[sarsaparilla]

As a writer of Haruhi fan fiction, I have had persistent difficulties with trying to imitate the style of the original novels in a convincing manner. The shortcoming is particularly conspicuous in the context of this series, as it's exactly the idiosyncratic style of narration that gives it a large part of its charm.

As it would appear to be that writing style reflects -- at least to some extent -- the personality of the author, it is really too bad that my own personality is too far removed from the protagonist of the novels for me to be naturally able to write in the same style. I've taken to heart the observation offered by a random reviewer on FF.net -- that I would indeed appear to be quite like Sasaki, as my works feel more as if they were narrated by her than by Kyon. This realization of insufficiency was one of the primary factors in my main writing project grinding to a halt right after the prologue.

Since it clearly isn't in me to achieve the goal in an effortless, natural manner, I finally decided that a more artificial approach might work instead. To this effect, I did some statistical analysis on a single story of the original series, hoping to find basic guidelines for improving my style. While this approach is not unlike trying to understand the concept of cookies with the aid of a mass spectrograph, after disintegrating them into their constituent molecules, I hope that the results I derived might be helpful for others who also want to study the relation between their own style and the original one. The treatment stays on a rather basic, mechanical level, without even trying to capture any finer points of style -- the reasoning being that those objectives might be within reach once the foundation is solid enough.

As with all translated works, there is always the question of whether a particular aspect of style is integral to the original vision of the author, a coincidental feature of the original language and culture, or an irrelevant artifact of translation. To minimize the effect of this uncertainty, I have tried to concentrate on features that should remain reasonably constant regardless of the language in which they are expressed. The main quantities that I estimated were Word per Sentence (WpS) ratio, as a proxy of sentence complexity, and the relative frequencies of different elements of narration.

I chose the short story Random Numbers -- currently the most recently published piece in the series -- as the target of my somewhat questionable activities. At a bit less than 7,000 words, it is a fairly representative sample of Tanigawa's writing, a light and fluffy story about the SOS Brigade visiting a shrine during New Year festivities. There is some verbal sparring between Kyon and Koizumi and subtle Kyon/Haruhi shipping.

I'll be dividing the results into three main topics: dialogue, narration and 'Kyonisms'.

Dialogue

Counting by words -- 2,114 in total -- dialogue represents 31.6% of the story. This should classify the piece as rather dialogue-heavy, but that is not immediately apparent when looking at the bulk of the text. There are two main factors contributing to this impression: Tanigawa's habit of interspersing narration with dialogue, even in ways that most seasoned writers would find inadvisable, and Kyon's imprecise rendering of his own dialogue, without quotation marks.

The interspersing issue looks like a cultural or translational artifact. It would appear to be quite possible to fix the formatting in adherence to usual typographical conventions, without losing anything essential. Therefore, it wouldn't seem to be worthwhile to even try to imitate the original style on issues like paragraphs containing dialogue from several different characters.

Broken down by character and mode, the dialogue consists of:

Snt Qst Exl  WpS Character
 24   9   0  8.0 Kyon (direct)
 39  10   1 10.7 Kyon (indirect)
  6   0   0  7.7 Kyon (4th wall)
 61   7  23  9.5 Haruhi
 59  10   0 11.4 Koizumi
  7   5   0  1.9 Asahina
  6   0   3  7.4 Tsuruya
  4   1   0  0.5 Nagato
  1   0   1  7.0 Imouto

Judged by the number of sentences (Snt), Kyon, Haruhi and Koizumi share the focus rather evenly between the three of them, with all the other characters having only passing contributions.

Kyon makes about as many questions (Qst) as Haruhi and Koizumi together; the rest don't have enough sentences to provide any meaningful statistics.

Exclamations (Exl) are almost exclusively made by Haruhi, Tsuruya and Imouto, and this should be seen as a stylistic choice to underline their 'genkiness'.

The average WpS ratio for dialogue is 10.1 which makes it uniformly rather plain, even clipped in appearance. Koizumi is the most verbose character, but not by much -- the imprecisely rendered part of Kyon's dialogue has about the same WpS value. Asahina and Nagato strongly deviate in the opposite direction, with Nagato managing the implausible feat of uttering, in average, less than one word per sentence.

Other than indirect dialogue being about twice as common as indirect one, there wouldn't seem to be any clear rule for how Kyon divides his dialogue between those modes. His direct dialogue consists mostly of plain, matter-of-fact statements and questions, though some amount of recognizable sarcasm still remains. Of the indirect dialogue, it is somewhat hard to estimate the percentage that is actually spoken instead of only thought; in some cases the fact can be inferred from the reactions of other characters, and in some others Kyon explicitly indicates that the dialogue is internal, leaving yet other cases ambiguous in this respect. However, it is rather clear that the indirect dialogue is more sarcastic and nuanced than the direct one. This tends to give the impression that Kyon is intentionally playing dumb.

There are a few occasions where Kyon either seemingly speaks directly to the reader, or otherwise draws attention to the fact that events are being narrated by him. I have counted these as a separate mode of dialogue, '4th wall breaking', and they can arguably be seen as a particular form of Kyonism.

Attribution

The joining of narrative and quotations, and dialogue tags, is a topic worthy of specific attention, if for no other reason that Tanigawa's way of doing them is particularly unusual -- to the degree that many of his mannerisms would normally be considered stylistic errors. Again, it is unclear which part of the issue is rooted in Japanese conventions, and what might have been introduced by slightly too literally-minded translators.

There are altogether 19 recognizable dialogue tags, making them somewhat rare. On the other hand, the average WpS value of 7.0 for just the tags suggests that when they are used, they are unusually complex. Tanigawa seems to prefer a construction of the type, 'said <character>, doing <this and that>,' with possibly some character description thrown in for good measure, as well. On an even other hand, there are also constructs that don't look like a valid tag at all, but are still used for that purpose. Examples of those would be:

As for Nagato, "<dialogue>."

or -- even more pathologically -- just:

So, "<dialogue>."

About 30% of the tags use 'said' as the descriptive verb, with the rest either going for a more eclectic alternative, or even dropping the verb altogether as in the examples above.

Differential diagnosis

In my own works, the WpS value for dialogue is significantly higher than in the original novels, and this error happens consistently in the same direction for all characters. In order to get closer to the proper style, I must reduce the complexity of spoken sentences. Specific attention must be given to Nagato's dialogue, which should in most situations be practically nonexistent.

I must pay closer attention to how Kyon divides his dialogue between direct and indirect modes, and how those modes differ stylistically. Compared to my works, a larger percentage of his original dialogue is questions rather than statements. Likewise, Haruhi should be making more exclamations in general.

Until now, my own works haven't had any dialogue tags. While it's arguably not that different from the 9% frequency of the original style, I shouldn't shy away from using them in situations where attribution is necessary but more indirect methods would appear unwieldy.

Narration

The part that isn't dialogue consists of Kyon's narration of events. At 4,572 words, it represents 68.4% of the story. The average WpS value of 17.7 is almost twice as high as for dialogue. This body of text was further divided into categories based on the primary narrative function of each sentence. These functions are listed below, in order of frequency, along with the respective WpS value and the number of Kyonisms (Kyo) found.

   %  WpS Kyo Function
21.9 16.4   5 Statement
17.5 19.8  20 Character
14.1 20.5   3 Action
 7.0 15.5   2 Internal
 4.8 12.8   2 Rambling
 3.2 21.3   1 Scenery

'Statements' are generic declarations of fact pertinent to the story, providing necessary context. Basically, this is the pile of sentences that wouldn't fit in any other, more specific category. A representative example of a sentence in this category would be:

In fact, it was just yesterday that we five returned from the villa that she so graciously invited us to stay in, free of charge.

The 'character' category contains all character descriptions, gesturing, and actions that are primarily used for characterization instead of advancing the plot. This consists of statements like:

I could still see reproach on Koizumi as he looked at me with a thin smile.

I guess that it might be useful to divide this category further according to the object of description, thus giving some indication on how Kyon's attention is divided between characters.

The 'action' category consists of plain descriptions of events happening, narrated in a more or less objective manner, as in:

I tore one of the sides of the envelope, turned it upside down, and two photographs fell out.

The 'internal' category consists of statements describing Kyon's internal state, his feelings, desires and attitudes, thus revealing something about his own character, as in:

I was relieved at the fact that she didn't seem to be angry at me.

'Ramblings' are short flights of fancy, showing Kyon's tendency to go off tangent into unusual and sometimes sarcastic directions, as when pondering the purpose of shrine apparel:

What's the purpose of such a big bell anyway? A substitute for an intercom?

As with the 4th wall breaking dialogue, these are essentially a form of Kyonism.

'Sceneries' are descriptions of the places where events happen, or in other words, the background element, such as:

Below it were the big vermilion doors, with many words written on there by various visitors.

When looking at the relative frequencies of different categories, one can immediately see some salient features. Representing 17.5% of all the words in the entire story, characterization is the single most important specific function. If this is combined with the 'internal' category, which is essentially Kyon's characterization of himself, the total goes up to almost 25%. When counting the dialogue as well the sum goes beyond 50% of all text. Based on this, it appears evident that the original series is strongly character focused.

It has been claimed that we know little about Kyon's own character because he is deliberately terse in that regard; in light of these numbers that wouldn't appear to be true. While he doesn't (naturally) describe himself the same way he does the other characters, there is still enough material to get a good understanding of his personality.

Another quite surprising issue is scenery, or the almost complete lack of it. The events happen at a place that has not been established elsewhere in the series, but Tanigawa is still content to leave the description to a couple of sketchy remarks, or the bare minimum required to tie the events to any place at all. If this story was rendered in manga format, 95% of the panels would have empty backgrounds!

The floating feeling caused by the vagueness of scenery is further reinforced by Tanigawa's habit of advancing different scenes side by side, without clearly separating them from each other. The first part of the story is a jumble of three or possibly four temporally distinct plot threads, without any scene transitions at all, making it rather hard to determine the temporal order of pieces of text. This might be an intentional effect.

WpS values are relatively consistent across different categories. Possibly the only outlier is 'ramblings'. This makes sense, as the boundary between ramblings and Kyon's indirect dialogue is somewhat arbitrary; thus, it is to be expected that the WpS values are closer to each other.

Differential diagnosis

The difference in WpS values is less for non-dialogue narration than it was for dialogue, but I still manage to write longer than original sentences in narration, as well. I should consciously avoid excessively complex sentence structures, even at the price of reduced precision of expression. There can be long and complex sentences -- even excessively so -- in the original material, but they are an exception rather than a rule.

The relative frequencies of narrative functions represent possibly the largest difference between my natural style and the original one. Compared to the original style, mine is somewhat lacking in characterization, especially when it comes to gestures. On the other hand, I tend to overshoot the amount of scenery description, but the worst outlier is the 'ramblings' category -- it is used in the originals to give a bit of local flavor, but in my stories it has a tendency to become the bulk of the work.

Kyonisms

There are some personal quirks in Kyon's narration; here, all such instances are called 'Kyonisms'. Because they are rather conspicuous, any writer who tries to imitate the original style should know what they are -- and just as importantly, how exactly they are used.

A couple of eccentricities -- 4th wall breaking and ramblings -- were already mentioned earlier. In addition to these, Kyon uses some unusual expressions, mostly similes and metaphors, to 'flourish' certain passages of narration. It doesn't take long to notice that the use of these flourishes is far from random. It is rather obvious that -- like 'love bubbles' in shoujo manga -- they are directly related to the enthusiasm with which Kyon narrates the passage in question.

Altogether, there were 33 recognizable Kyonisms in the story, such as:

While the three girls could be described as reasonably fair, dressed to their utmost in the clothes of the Japanese soul, they stood out like the three stars of Orion's belt in the winter sky.

which, containing both a metaphor and a simile, is arguably a double Kyonism.

Looking at the distribution of these expressions among different narrative elements, there aren't any Kyonisms in the bulk of dialogue, even in Kyon's own part of it. For the non-dialogue narration, the occurrences are evenly divided between different narrative categories, with one very significant exception: about 60% of the occurrences happen in relation to characterization. Thus, it is not just that the series is character focused, but Kyon himself is most engrossed when describing characters, by a very wide margin.

Thus, a good first order approximation would be to use Kyonisms only in relation to characters, since all other uses seem to be relatively incidental.

Next, I divided the existing Kyonisms according to theme, with the following results:

6 Generic fanciful simile
5 Generic mundane simile
5 Unusual metaphor
5 Animal motif
3 Japanese culture
2 Plant motif
2 Sports
2 Military
2 Generic culture
1 Sciences

Kyonisms with a science theme are probably the most memorable, if not for other reason than how fanciful they are, but as the numbers above suggest, they are an outlier rather than the main course. In the course of the short story, Kyon manages to compare Haruhi to -- I'm not making this up -- a duck, a rabbit, a puppy, a mole and a beetle. As suggested by this list, the relation between themes and objects is far from random as well. Besides fauna, military similes seem to be Haruhi's exclusive property, as well.

Then, there is one final insight that can be derived by arranging Kyonisms according to their object of comparison. In order of frequency, the things that inspire Kyon's poetic side are:

16 Haruhi
 3 Asahina
 3 SOS girls as a group
 2 Nagato
 2 Kyon himself
 7 Other or indefinite

I believe that it is right here where Kyon, even unwittingly, reveals something that he usually does his best to hide. Sure, he may gush endlessly about how beautiful Asahina-san looks in her borrowed kimono, but when it comes to pursuing that extra metaphor like a baseball player going for a home run, the person in Kyon's mind is none other than Haruhi.

Well, I think that's pretty cute. ^_^

Final notes

When examined in detail, an unavoidable conclusion seems to be that on a purely technical level Tanigawa's writing is surprisingly poor. The irregularities are too profound to be explained away as artifacts of translation. Probably the best word to describe how dialogue is handled would be 'messy'; scene establishment is consistently weak. Combined with the apparent shortcomings in crafting longer stories -- as demonstrated by, e.g., novels nine to eleven -- one must ask what is it then that makes Haruhi novels worth reading, and thus, which elements of the original material should be carefully conserved to stay true to the series.

Tanigawa's most compelling works are light, relatively sketchy character studies -- vignettes, even -- with little if any plot to steal the focus from characterization. There are certain fundamental facts of the basic setting that are -- quite intentionally -- never revealed; furthermore and for whatever reason, the author is clearly reluctant to explore certain obvious avenues of development. This in turn sets hard limits on what is possible to accomplish within the given format. Tanigawa's own attempts at introducing more complex plots have been less than successful; it is unclear whether this tells something about the skill of the author, or perhaps it would indeed be infeasible to write a strongly plot-driven story that stays within the implicit constraints.

Keeping the above in mind, the conclusion would appear to be that the most important element to get right is characterization -- not just the idiosyncracies of each character but the overall focus on character interactions and the specific methods used in portraying them. With these in order, it might be even inadvisable to try to single-mindedly follow other quirks of the author. WAFFy one-shots would be easiest to get 'right', as the setting naturally leans in that direction. Stories with a plot of any significant substance most likely transgress canon boundaries one way or another, anyway.

[Jason_Miao]

Questions: Did you analyze the Japanese light novels, or the translation?  And are you interested in writing a fanfic which narrates like Kyon, or writing a fanfic which English readers will be comfortable with as being narrated by Kyon?

I tend to write posts in ways that occasionally imply meanings that I didn't intend.  So let me say this clearly here: however you cut it, this is a remarkable analysis and I don't want to come across as denigrating it in any way.  The only comparison in anime fandom I've seen was to count the number of times each character used a mallet to disprove the popular conception that its use was particular to a the romantic lead.  This goes far above and beyond that and can be generally useful for any Haruhi writer.

That said, since you prefaced your analysis by a comment that you don't write in the same style, I wanted to mention that while the writing style reflect the writer, a translation also reflects the translator.   How one approaches differences in grammar and idiom to reflect the meaning of the original work are also a large part of the skill of translation.  Since I am unfamiliar with what reference sources you use, if you have only used the English translations, it is possible that what you are taking as Kyon's narration habits may be an effect of the translation rather than the actual character*.

So, depending on your goal, you may or may not want to look further.  If you want to write "like Kyon" then you may also want to compare Tanigawa Nageru's original story with the translation to see what differences may have been introduced - perhaps .  If you're more interested in writing in a way that readers will be familiar, what you have is probably best.




* And I had a fairly long example in mind comparing Viz and Ranma 1/2 with AnimEigo and Urusei Yatsura, before I realized that none of it would likely mean anything to most people these days anyway.

[Muphrid]

Fascinating analysis.  Tanigawa's emphasis on character (and comparative lack of attention to scenery description) is really interesting to me.  I can appreciate how painstaking this process must've been.  I do agree with Jason's point that some of this could be introduced by the translator--that was ultimatemegax, right?  remember that he loves Koizumi, and that he's also a scientist himself.

Now, something that could be interesting might be to look at most frequent adjectives and adverbs.  It might require a larger sample, but this article comparing Collins, Meyer, and Rowling's styles with a frequency analysis of different words caught my eye a while back.  The kinds of adjectives and adverbs one uses betrays one's emphasis in telling the story:  it's pretty telling how much Meyer's adjectives and adverbs focus on describing people and their actions, compared to the other two.  I experimented with trying to do this myself, but the big roadblock I ran into was identifying the parts of speech.

Perhaps something from this list of textual analysis tools would be useful in that respect.  Having played with such tools now and then, I know that they like to look at average word length and sentence length and such; I think you took it to the next level in breaking down sentence length not only by function of the sentence (narration vs dialogue) but also by speaker (in the case of dialogue).  That's something an automated tool probably can't do at this time, but on the other hand, I can imagine it would require considerably less personal effort.

Another thing I've used in lieu of the above is a word cloud, but these tend to end up being names and nouns and are thus less useful.

----

Let me say some more things about the analysis you've performed, though.  You noted that Kyon has as many questions as Haruhi and Koizumi put together.  By ratio, Kyon seems to have quite a bit more.

The dialogue tag stuff is really interesting to me because I only remembered dialogue being alone on its own lines.  This is something I'll have to keep in mind as I try to wrap up c&c for Ergoemos.

I think you've drawn solid conclusions about Tanigawa's shortcomings; character is clearly most important to him, and he seems quite a bit weaker in building a strong plot or exploring the details of the setting.  I mean, what's the strongest single plot in the series?  One might say it's Disappearance, and in turn one could say it's strong because it's really focused on Nagato and Kyon.  Some of the plot details of that story aren't so good:  why should Kyon have to gather all the Brigade in the club room to activate the failsafe?  It gives him a reason to get Haruhi and Koizumi, but that could've easily been dropped.  It only rewards the audience to see that.

[sarsaparilla]

Questions: Did you analyze the Japanese light novels, or the translation?  And are you interested in writing a fanfic which narrates like Kyon, or writing a fanfic which English readers will be comfortable with as being narrated by Kyon?

I believe that the linked story is a fanslation; I don't have the necessary tools to process original, Japanese material in the same way. On the other hand, a fanslation is quite likely closer to the original in a literal sense than a professional translation, and on an even other hand, the results that I derived -- when interpreted in the proper way -- should be relatively constant regardless of the underlying language.

For the second question -- since I'm writing in English, my primary goal is to write in a style that English readers find familiar, and thus it's a reasonable choice to use English text in the analysis instead of the original Japanese.

However, and perhaps most importantly, I did this excercise to force myself into taking a closer look at how exactly my style differs from the originals, and this seems to have been a success. After the analysis I re-read one of my old fics and could immediately see what was wrong with particular passages, and how they should be modified. Thus, the next step on my quest to becoming a better writer will probably be to rewrite that fic, concentrating solely on getting the style right.

[Muphrid]

So after some fiddling, I discovered AntWordProfiler.

It has a couple word lists that are based on the most common 2000 words in the English language, including words that are built upon those base words.  While the words most frequently used in the top 1000 are too common to draw great insight, I thought the top 20 words in the second 1000 might be interesting.  I looked at my text files of the Haruhi novels and found this:

Code Select Expand

Group uf_1
club 818
probable 614
sudden 544
tea 365
bit 364
thank 364
slow 345
guess 334
cat 333
quick 325
worry 325
inside 310
during 299
sorry 275
hullo 271
mystery 269
pick 257
lot 252
exact 239
origin 227


Words like "club" and "tea" don't surprise me.  Things like "sudden," "guess," "probable," "quick," and "worry" strike me as potentially distinctive, though.