Introduction

A literary theory, typically, is a set of concepts and general statements about them that serves as a framework for the study of particular literary works or broader phenomena. It is not at all surprising that a study of such an immensely complex object as a work of literature can't even be attempted without some kind of a framework. The success of a theory, then, is judged by how well it aids in studying literature, i.e. whether it helps to reveal hitherto hidden connections, dependencies and relationships, unify apparently unrelated phenomena as exemplars of a general concept, and so on. It is equally obvious that there can't be a single "right" framework, but rather a large number of them approaching the object from different angles and focusing on different aspects of it (see Howard 2005).

Although there is nothing intrinsically wrong with the notion of theory in humanities, the scientific model holds an allure of objectivity that is inevitably felt in comparison whenever the goal of a researcher is to obtain knowledge (as opposed to, for example, effecting social change). As a quite arbitrary example, consider M. Shapir's (1995: 34) statement that a text "can be poetic or prosaic, it can be neither, but it can't stand somewhere in the middle, between poetry and prose". This is a factual statement by its form, but how can one observe that it is true or false? To be sure, there is extensive argument in support of the statement, with references and examples, but there is no possibility for an experimentum crucis, or any observational procedure for that matter, that a disinterested observer could conduct and obtain some kind of a corroborating evidence.

One could argue that the nature of the proof in humanities is akin to that in mathematics, as opposed to natural sciences, namely, that it is a sequence of logical steps, each of which is undeniable (or at least, sufficiently convincing for the given audience), not by the force of empirical evidence, but by the force of being self-evident. However, mathematical proof is very well subject to falsification in Popper-like sense: if somebody claims to have proved that all numbers are even, just hand him — or her — the number three. Nothing of this kind is possible with such statements as quoted above. Or is it?

I think that the Positivist-style "assault on scholarship in the humanities", as portrayed by B.A. Wilson in the epigraph, is essentially misguided. The right emphasis would not be on "articulating statements capable of being either true or false", because one can produce verifiable, but uninteresting statements in abundance ("there are 12 words in War and Peace" is certainly capable of being false), — but on finding ways to endow interesting statements with the capability of being true or false. That is, finding ways to verify them. Andrey Bely was, probably, the first to do this by introducing "exact methods" in literary scholarship. In his studies of Russian poetic rhythm (Bely 1969), he invented a method to quantify the rhythmic richness of a poem or set of poems, i.e. to make it a matter of mechanical calculation, rather than intuitive judgment. And that allows one to make interesting statements "capable of being true or false" (although one can still contest the significance of the result and whether a statement means what it was intended to mean).

The first scholar to systematically develop the exact methodology was Boris Yarkho (Yarkho 2006; see also an excellent concise review in Gasparov 1994). Unfortunately, his pivotal works of the 1920s and 1930s remained largely unpublished until the 70s, and the first complete edition appeared only in 2006. As an example of Yarkho's approach, consider his study of Corneille's plays. In it Yarkho set up to find formal criteria that would correspond to the distinction between the genres of tragedy and comedy as designated by the author. Yarkho studied over 30 statistical characteristics of texts, such as the average cue length, the proportion of cues ending in the middle of a line, the number of events, the ratio of events on the scene to events behind the scene, and so on. It turned out that about half of the traits clearly differentiate comedies from tragedies, and that many of the distinctions can be interpreted as comedy being "faster" and "lighter". That in itself is not surprising — what is remarkable is the possibility, discovered by Yarkho, to quantify rather nebulous traits that before could only be assessed intuitively. Moreover, when Yarkho applied his indicators to Corneille's plays with non-standard genre designation (e.g. tragi-comedies), he found that they indeed occupied intermediate position between unequivocal comedies and tragedies.

Note that there is a striking structural similarity between the questions addressed by Yarkho and by Shapir (cited above): both concern a binary opposition (poetry/prose, comedy/tragedy) and whether it exhaustively classifies objects of the study. Of course, the scope of Yarkho's study is much more limited, but what is important is the very possibility of a definitive answer that he demonstrated. It is possible now to sketch out a research program that would involve more material from different historical periods and literary genres and generalize Yarkho's indicators. Obviously, this would contribute significantly to our understanding of the nature of genres. It is also worth emphasizing that such results would transcend any particular literary theory. That is, any theory that has anything to say about genres could — and, indeed, would have to — make sure its statements agree with the data. This is essentially the role of experiment in sciences, and so such methods in literary studies can be referred to as experimental (even though they do not involve any techniques of laboratory experiments).

Poetic rhythm is the object most amenable to exact study (e.g. Kolmogorov, Rychkova 2000; Tarlinskaya 1992), since it is most easily formalized. However works of Yarkho and M.L. Gasparov with co-authors (e.g. Gasparov, Skulacheva 2005) amply demonstrated that other aspects of a literary text, up to the level of thematic motives, can be brought under investigation based on statistical analysis of texts. The present work is, however, based on a different kind of data, characterizing not so much text per se as the reader's perception of it. The data comes from a large-scale experiment I have been conducting since 2003. The experiment is discussed at length in (Manin 2008), but for the sake of completeness we need to give a basic description here.

The Experiment

Not a line is drawn without intention... As Poetry admits not a Letter that is Insignificant so Painting admits not a Grain of Sand or a Blade of Grass Insignificant much less an Insignificant Blur or Mark.
William Blake, A Vision of the Last Judgment

Every word must be mandatory.
Daniil Kharms (2002)

... the reader regards the given text (if it is a perfect work of art) is the only possible one: "No word can be dropped from a song".
Yu. Lotman (1998), Chapter 1

In my talk about poetry I considered the example of Eugene Onegin. The entire poem is an irreducible unity, where EVERYTHING is necessary and interconnected, not a single word can be omitted without affecting integrity, etc.
A.N. Kolmogorov (1997)

The poet's conviction that nothing is left to chance in poetry refutes the puerile reasoning of some literary scholars who think that "a poem may contain structures that are not connected with its literary function and impact".
R. Jakobson (1987)

Some philologists maintain that philology is the science of slow reading. The art of seeking a meaning in works of poetry is certainly an art of ever slower reading. For every word, every punctiation mark arises not accidentally, but by slowly crystallizing into a whole as complex as the very world that we call the lyric poem.
Andrej Bely (1969)

Another assumption about a work of literary art is that it should involve unexpected, striking, surprising words, tropes, turns of phrase and subject, etc. It is typically accepted implicitly, but was made explicit by Viktor Shklovsky (1917) in his theory of defamiliarization (ostranenie). The notion came to the foreground again in a different guise in the 1950s and 60s, when Shannon's information theory became popular. Since mathematical information is, in a sense, a measure of unexpectedness, many believed that literary texts can be rigorously proved to have higher information density than utilitarian texts. In fact, Shannon himself argued in his groundbreaking paper (1948) that Joyce's Finnegans Wake "is alleged to achieve a compression of semantic content" compared to texts in Basic English because of its larger vocabulary and hence, lower predictability. Both Yuri Lotman (1998) (developing ideas of A.N. Kolmogorov, see Uspensky 1997) and Umberto Eco (2006) attempted to apply the ideas of information theory to literature.

The experiment described here is based on the idea that both notions of interconnectedness and unexpectedness of the elements in a literary text can be subjected to a direct verification, at least on the level of individual words. Indeed, unexpectedness can be measured by the probability for a reader to guess an omitted word. As for interconnectedness, if the word in text is (perceived by the reader as) the most appropriate one, the reader should be able to easily tell the original word from any replacement, i.e. presented with the original word and a replacement, select the correct one.

This scheme was implemented in the form of a Web-based literary game in Russian (http://ygrec.msk.ru). It contains a number of text fragments which are presented to the voluntary participants in trials of one of three types:

• type 1: fill in an omitted word;
• type 2: determine whether the highlighted word is original or replaced;
• type 3: select the original word from two alternatives.

The "open" nature of the experimental procedure makes it possible to obtain large amounts of data necessary for statistical validity of the results: thus, by Summer of 2007, over 8000 participants collectively generated over 1 mln data points. Of course, some difficulties are also introduced by this methodology; they are discussed in detail elsewhere (Manin 2008).

It should be noted that, in contrast to statistical methods, our data do not characterize texts as such, but rather the reader's perception of them. The Russian readers of the late 19th century would have different success rates guessing words in Tyutchev's poems (not to mention Vvedensky) than contemporary readers. In contrast to psychological and psychophysiological methods of studying reader response, our data do not really tell us anything about the reader's overall emotional reaction to the text or aesthetic evaluation of it. On the other hand, it provides a "high-resolution", detailed view, potentially on the level of individual words, like text analysis. The hypothesis is that the quantities being measured have something to do with the construction of the text, on the one hand, and with the reader's emotional/intellectual response, on the other. This hypothesis needs to be verified by preliminary data analysis, after which we can use the data to obtain new results.

The results are interpreted in terms of two quantities, unpredictability U and constrainedness B. Unpredictability is defined as the negative binary logarithm of the guess rate in trials of type 1 (guess a word): U = −log₂p₁. It can range from 0 (all words are reliably guessed) to infinity (no word is ever guessed). Constrainedness characterizes the success rate of type 3 trials, B = log₂p₃/(1−p₃) and can range from 0 (success rate no better than chance, i.e. replacements are perceived as being no different than the original word) to infinity (all trials are successful, the original word is easily distinguished from replacements).

The text fragments are presented to the participants in random order, and the word to be used in any particular trial is selected randomly (only words of 5 or more letters are used). The average fragment length roughly corresponds to 4 lines of iambic pentameter. Texts are grouped in categories, usually by author. Some categories are narrower, such as an author's poems from a particular period, and some are broader. There are two sets of texts (stages) in the experiment. Stage 1 is intended to explore the wide range of texts: poetry and prose, classical to contemporary, traditionalist to avant-garde, amateur to professional. It has 3439 fragments in 33 categories. Upon completing all texts in Stage 1, participants proceed to Stage 2, where texts are selected to explore more specific distinctions, in particular the differences between prose, free verse, rhymed and non-rhymed metric verse, and various intermediate forms. Details will follow, but first we need to outline some of the results of (Manin 2008).

Poetry vs. prose in terms of unpredictability and constrainedness

The difference between typical poetry and typical prose is revealed by the constrainedness metric: it is significantly higher for poetry (0.84—1.3) than for prose (0.42—0.61). The "difficult", avant-garde poetry is different in that it has higher unpredictability than both prose and traditional poetry (1.9—2.2), while its constrainedness is in the prosaic range. Thus, both traditional and avant-garde poetry differs from prose by the increased sum of unpredictability and constrainedness. This is consistent with the intuitive notion that poems are more tightly constructed, structurally richer than prosaic texts.

It is also possible to isolate the effect of poetic meter. Indeed, the probability to guess a word is a product of the probability to guess the word's syllabic length by the probability to guess the word, once its length is known. Since logarithm of a product is the sum of logarithms of the multipliers, the total unpredictability can be represented as the sum of metric and non-metric parts. Metric unpredictability is simply the (negative binary) logarithm of the probability to guess the word length, while non-metric unpredictability is the success rate in type 1 trials where the answer (correct or incorrect) has the right syllabic length. Not surprisingly, metric unpredictability in poetry is very low (0.09—0.32), while in prose it is responsible for the larger part of the total unpredictability (0.8—0.98). That is, it is very easy to guess the number of syllables in a word omitted from a poetic line, but that doesn't help much to guess the word itself; while correctly guessing the word length in prose is two-thirds of the task. Correspondingly, non-metric unpredictability is significantly greater in poetry than in prose, which quantifies the notion of "poetic license" with respect to the freedom of word selection and combination.

Poetic meter affects the selection of the original word in type 3 trials as well. Obviously, it is much easier to reject a word replacement if it has a wrong syllabic length (doesn't fit in the meter). This effect is characterized by metric and non-metric constrainedness. Non-metric constrainedness is defined similarly to non-metric unpredictability: it is constrainedness calculated only over the trials where replacement is of correct syllabic length. In other words, it measures how difficult it is to distinguish the original word from replacement, based on semantics, phonetics and syntax, but not on rhythm, and so is an interesting measure for prose to poetry comparison. It should be noted that rhythm depends on the position of the stress in a word, and not just on its length, but for technical reasons we only account for word length in this work.

It is somewhat harder to grasp the definition of metric constrainedness,

B_m = −log₂(p₃^T/p₃^F)

where p₃^T,F is the probability of the correct answer in type 3 trials when the replacement length is correct or incorrect (T and F superscripts resp.). It indicates the degree to which word length alone helps to distinguish the original word from a replacement; or, in other words, how strongly the text confirms the metric expectation that it creates in the reader. Higher metric constrainedness means that it is easier for the reader to reject a replacement just because of its different syllabic length. Zero metric constrainedness means that syllabic length doesn't have a role in distinguishing between the original word and its replacement.

It was already mentioned that the first set of texts (Stage 1) of the experiment contains almost exclusively prose and metric rhymed poetry. In contrast, the second set (Stage 2) also contains free verse (vers libre), metric blank (unrhymed) verse, and "poeticized prose", i.e. several generally prosaic texts that, for the lack of a better term, lean towards poetry in different ways (see Appendix). Now that we understand the properties of prose and metric verse with respect to unpredictability and constrainedness, we can turn to the proper topic of the present work: to characterize free verse in this coordinate system.

Existing theoretical perspectives on Free Verse

The main question raised in these discussions is whether free verse is "just" prose broken into lines, or it is something more than that. The broad consensus of its practitioners, as demonstrated by a large number of quotes in (Orlitsky 1995), seems to be that free verse must possess some kind of a "rhythmical breathing" and "poeticity" that is very hard to define. Thus, Yu. Linnik writes: "As it sheds the purely external confines of meter, free verse requires an extremely complicated internal organization of intonation. For this, the poet must have perfect pitch. It is easy to versify within the boundaries of common verse — it is impossible to imitate, to fake the true free verse."

According to Yu. Orlitsky (2008) and M. Gasparov (1997), there are two main positions with respect to the existence of rhythm in free verse. One, exemplified by E. Polivanov, A. Zhovtis and O. Ovcharenko, is that "units of repetition" (mery povtora) exist, but are fluid and constantly changing. The other, represented by Orlitsky, M. Gasparov, and V. Baevsky, denies the existence of any repetitive metric structures and defines free verse as completely irregular accentual verse. Orlitsky notes that the notion of constantly changing units of repetition grants too much freedom to the inquiring researcher to find what he looks for. Gasparov, in a typically insightful passage in (Gasparov 1997), notes that the difference between the two positions can be related to the difference between the first reading of the text and its re-reading. In the first reading, one tends to form a metric expectation that is constantly frustrated. In re-reading, the text is approached without expecting any repetitions. He further remarks that as a reader, he tends to avoid forming any metric expectations, but as a free verse practitioner, he knows that the rhythm of every line being composed only exists in its interaction with the adjacent lines.

It should be noted that even in English-language poetry, where free verse has been dominant for several decades now, these same questions arise as well, with very similar answers being proposed. Thus, G. Burns Cooper writes in his book (1998: 10−11): "There's little agreement about what makes this verse metrical, or rhythmic, or what makes it poetry at all. ... Poets who write free verse ... usually want to insist that free verse is not the same thing as prose chopped up into lines, and that at least part of this difference is rhythmic or musical." He goes on to prove by analyzing texts that "poetry tends to be more compressed, and to have more regular alternation, than other forms of discourse." With respect to Russian free verse, M. Gasparov and T. Skulacheva (2005: ??−??) analyzed Aleksandria Songs by M. Kuzmin and found that the interaction of rhythm and syntax there works to cement the unity of a poetic line and so emphasize the role of line breaks. This line of research is very promising, but to build a wider perspective, many more texts have yet to be studied.

At present, the precise nature of rhythm in free verse remains elusive. G. Burns Cooper (1998: 2) concludes that it is created by "more regular alternation" on various levels, from phonetic to thematic. Likewise, Joseph Brodsky (1995) commented on his poem "San Pietro":

The poem is written in vers libre, and when writing vers libre, there has to be some organizing principle. Here we have doublets (dvojchatki) at stanza endings: either literal or psychological. Here: "it's not terracotta and ochre absorbing moisture but moisture absorbing terracotta and ochre". Or: "in a coat and shirtless, in shoes and sockless". Or a doublet in the form of a rhyme: "chugunnaya kobyla Viktora-Emmanuila" (cast iron mare of Victor Emmanuel). I know that, but nobody else knows. There's a lot of such stuff in there.

But this is precisely the kind of question that can be answered by the data from our experiment.

Free Verse in terms of unpredictability and constrainedness

Fig. 1 presents the average unpredictability and constrainedness for each text type (cross-bars denote the standard deviation, a measure of the likely statistical error). As mentioned earlier, the average unpredictability lies in a very narrow range. It is fair to say that the differences between categories are on the verge of being statistically insignificant. It should be noted that the averages hide the variability of texts within types and, generally speaking, will depend on the selection of texts in the studied corpus, specifically on the proportion of "easy", "clear" versus "difficult", "obscure" texts, which can be found in all types. Therefore, one shouldn't overestimate the exact numerical values here; what is significant is the qualitative, approximate equality.

Fig. 1

There are significant differences in constrainedness between text types: according to this indicator, "poeticized prose", free verse and blank verse occupy intermediate position between prose and (metrical rhymed) poetry, in this order. However, as discussed above, both unpredictability and constrainedness depend significantly on the effects of meter and rhyme. While the total unpredictability of poetry is about the same as that of prose, it is mostly non-metric, while unpredictability of prose is predominantly metric. That is, given the length of the omitted word, it is much easier to predict the word itself in prose than in poetry. Or, from the production point of view, while meter imposes significant restrictions on the word choice in metric verse, "poetic license" allows the poet to relax common semantic and syntactic restrictions, thus increasing the total freedom (and unpredictability). Our numerical results only show that these two effects compensate each other almost perfectly, so the resulting freedom of word choice in metric verse tends to be about the same as in prose.

This result was obtained in (Manin 2008), where only prose and metric rhymed verse were studied. How do other text types fit into this picture? Fig. 2 shows the five text types in terms of metric unpredictability and metric constrainedness. Unsurprisingly, the metric unpredictability and constrainedness of blank verse are both very close to those of rhymed poetry, while free verse and poeticized prose, which lack any pronounced rhythmic regularity, are very close to prose. However, both free verse and poeticized prose display a subtle difference from prose: their metric constrainedness is negative.

Fig. 2

To understand what that means, recall the definition of metric constrainedness. It is based on the idea that in metric verse, a replacement word can be rejected just because it is of incorrect syllabic length. In a text with some degree of rhythmic regularity, incorrect length should also help to reject the replacement. Metric constrainedness is a measure of this effect. It is positive, if it is easier for the reader to select the original word in type 3 trials when the two options have different lengths (and hence, the one with the wrong length is the replacement). If the word length has no effect, i.e. there is no rhythmic regularity in the text whatsoever, metric constrainedness should be equal to zero. We can say that it is a measure of how well the text confirms the rhythmic expectation that it creates in the reader. Interestingly, prose is characterized by small, but positive metric constrainedness. This is observed in both the Stage 1 (not shown here) and Stage 2 results. Apparently, it possesses a small, but perceivable, degree of rhythmic regularity.

But what does negative metric constrainedness mean? One way to conceptualize this is to say that such a text tends to violate, rather than confirm, the rhythmic expectation it creates in the reader. Indeed, if the length difference between the original word and replacement, when it is present, hinders the correct choice, rather than facilitating it, then the additional hint provided by the word length tends to be misleading. The reader chooses (mostly, unconsciously) the word length according to the expectations created by the context, and more often than not, this choice is incorrect. This looks very much like Tynyanov's idea of free verse where each line has a "dynamically-successive metrical impulse", but no "dynamically-simultaneous metrical resolution" (Tynyanov 1965), i.e. in Gasparov's (1997) words, where after each line there's an expectation that the next line will be rhythmically similar, and every time this expectation is frustrated. "The question is, how long can this hopeless expectation last?" — Gasparov asks. We can't answer the question, but we have an evidence that this "hopeless expectation" does exist.

Another way to look at the negative metric constrainedness is to note that when the selection between two words involves a difference in length, this adds another dimension to the problem: one word may be more appropriate (in the reader's mind) by length, but less appropriate by some other criteria (meaning, phonetics, etc.) than the other. When rhythm is a formal constraint, as in metric verse, the length criterion always wins, and the choice becomes easier. But if the rhythm is an aesthetic dimension in its own right, the choice becomes harder, because there are more contradictory criteria to reconcile.

We'll emphasize once again that the negative metric constrainedness of free verse and poeticized prose is a very weak effect, on the same order as the positive metric constrainedness of prose. As more data becomes available, it may turn out to be just a statistical fluctuation. But if the effect is confirmed, it will prove the intuitive notion that rhythm in free verse indeed plays a very special role, different from that in either metric verse or prose.

Next, we turn to the non-metric unpredictability and constrainedness, Fig. 3. Non-metric unpredictability measures how easy it is to guess a word if its syllabic length is known. It factors out the effect of poetic meter and measures the unexpectedness of semantic and syntactic construction of the text. By this measure, both free verse and poeticized prose lie close to regular prose and apart from metric rhymed and unrhymed verse. On the other hand, free verse and poeticized prose are grouped with metric verse and apart from prose by non-metric constrainedness. This quantity measures the "fitness" of a word to its place in text (by the reader's judgment) with the effect of rhythm, again, factored out, i.e. calculated only from trials where the replacement is of the same length as the original word.

Fig. 3

It turns out then, that free verse (and poeticized prose) does occupy an intermediate position between metric verse and prose by these criteria, but it is not an arithmetic mean. Free verse is similar to prose in one respect and to metric verse in the other respect. Paradoxically, free verse is less free than metric verse to violate common semantic and syntactic constraints on word combination, and is more like prose by this criterion. But free verse is like poetry in the way the words are tightly fit (or woven) together and form a cohesive structure.

Of course, a note of caution is due here: one can easily find examples of free verse with, for example, very unusual, even downright cryptic, word usage. Our results were obtained with a relatively limited, though reasonably diverse, sample and characterize that sample only. Its representativity can be disputed. However this only means that the results should be generalized with caution. It is still the first time that some of the basic theoretical tenets related to the nature of free verse can be tested against objective data.

Conclusions

The most pressing question for the further study seems to be: how does free verse and traditional verse achieve the heightened non-metric constrainedness relative to prose? Is it due to the use of poetic tropes? Does introduction of line breaks alone increase constrainedness? Are there any other identifiable mechanisms? This remains to be seen.

References

Верлибры недостойны почестей, — 
Там, как невеста без венчанья, 
Томится слово в одиночестве,
Не слыша парного звучанья.

(For free verse let no praise be said, —
In it, in solitude confined,
A word is like a bride unwed
And cannot hear a matching sound.)

Такие, как эти, стихи можно писать бесконечно,
Но бумага кончается — и это, увы, реальность.
Жестокая, если угодно.
Если бы я писал это в рифму — ушло бы дней пять,
А так — меньше часа,
Даже жаль, что так быстро,
А еще далеко до обеда...

(Poems such as this, one could write endlessly,
But the paper runs out — alas, this is reality.
You may call it cruel. 
If I wrote this with rhymes — it could take five days,
This way it took under an hour. 
It's a bit of a pity: I'm done, 
And it's still a long time before lunch...)

Appendix: texts used in the experiment

Душа молчит. В холодном небе
Все те же звезды ей горят. 
Кругом о злате иль о хлебе
Народы шумные кричат...

Иные мысли мне приходят ныне:
передо мною кровель строй багряный,
и медь листвы тончайшего чекана,
и ясность зорь, и добрый жар в камине.

По всякой земле
Балаганчик везу,
А что я видал
На своем веку:
Кусочек плоти
Бредет внизу,
Кусочек металла
Летит наверху.

Замерзая, я вижу, как за моря
солнце садится, и никого кругом.
То ли каблук скользит, то ли сама земля
закругляется под каблуком.

В горах Сицилии, в монастыре забытом, 
По храму темному, по выщербленным плитам,
В разрушенный алтарь пастух меня привел,
И увидал я там: стоит нагой престол,

Как хорошо лицо свое иметь...
Тот смотрит робко, этот смотрит гордо.
Тот любит в лодке с удочкой сидеть.
Другой в восторг приходит от кроссворда.

Мой суфий сердца трезв как тамплиер
Уста души невинней баядеры
Смирения двугорбый дромадер
Берет подъем не претыкаясь в еры

и не было ни одного среди всех
живущих, кто мог бы всерьез исповедовать мне свой грех,
говорит железная печка: не трехногому коту 
исповедовать мне свою хромоту,

О ты, пространством бесконечный,
Живый в движеньи вещества,
Теченьем времени превечный,
Без лиц, в трех лицах божества!

Тревожьтесь обо мне
Пристрастно и глубоко,
Не стойте в стороне,
Когда мне одиноко.

Достойно, главное, достойно
любые встретить времена,
когда эпоха то застойна,
то взбаламучена до дна.

Закинув на спину трубу,
Как бремя золотое,
Он шел, в обиде на судьбу.
За ним бежали двое.

Движется нахмуренная туча,
Обложив полнеба вдалеке,
Движется, огромна и тягуча,
С фонарем в приподнятой руке.

Любимая, когда впервые мне
ты улыбнулась ртом своим беззубым,
точней, нелепо растянула губы,
прожженный и потасканный вполне,

я вдруг поплыл — как льдина по весне,
осклабившись в ответ светло и тупо.
И зазвучали ангельские трубы
и арфы серафимов в вышине!

Чего же нам бояться, друг Боэций,
в виду возможной смены декораций —
нам, умыкавшим Девушек с Трапеций,
и побеждавшим персов, как Гораций,

Кросс — это самый трудный бег,
Такая проба сил,
Где сразу виден человек:
Могуч он или хил.

В том краю Килиманджаро, 
Где о берег бьет прибой,
Жил меж грядок с кабачками
Мистер Йонги-Бонги-Бой.

Где слог найду, чтоб описать прогулку,
Шабли во льду, поджаренную булку
И вишен спелых сладостный агат?
Далек закат, и в море слышен гулко
Плеск тел, чей жар прохладе влаги рад,

Печальный Демон, дух изгнанья,
Летал над грешною землей,
И лучших дней воспоминанья
Пред ним теснилися толпой;

К н я з ь
Ах, никогда мне это не забыть...
Вы жизнь мою спасли...
А р б е н и н
                      И деньги ваши тоже
(г о р ь к о)
А право, трудно разрешить,
Которое из этих двух дороже.

Мечом снопа опять разбуженный паук
Закапал по стеклу корявыми ногами.
Мизерикордией! Не надо лишних мук. 
Но ты в дверях жуешь лениво сапогами.

Уже прекрасное светило 
Простерло блеск свой по земли
И божие дела открыло:
Мой дух, с веселием внемли;
Чудяся ясным толь лучам,
Представь, каков зиждитель сам!

Где римский судия судил чужой народ,
Стоит базилика, — и, радостный и первый,
Как некогда Адам, распластывая нервы, 
Играет мышцами крестовый легкий свод.

Как люб мне натугой живущий, 
Столетьем считающий год,
Рожающий, спящий, орущий,
К земле пригвожденный народ.

Нетрудно порошок принять,
Но надобно его понять.
Вот так и вас хочу понять я —
И вас, и наши обоюдные объятья.

Весь день я спал, и рушась от загона,
На всем ходу гася в колбасных свет,
Совсем еще по-зимнему вагоны
К пяти заставам заметали след.

Мелко исписанный инеем двор!
Ты — точно приговор к ссылке
На недоед, недосып, недобор,
На недопой и на боль в затылке.

Во всем мне хочется дойти
До самой сути.
В работе, в поисках пути,
В сердечной смуте;

Я всю жизнь свою провел в мытье посуды
И в сложении возвышенных стихов
Мудрость жизненная вся моя отсюда
Оттого и нрав мой тверд и несуров

Соперники в искусстве брани,
Не знайте мира меж собой;
Несите мрачной славе дани
И упивайтеся враждой!

К душе твоей, как к пламени свечи,
Тянусь во тьме,
И сумрак с ней не страшен,
Когда бредет безрадостно в ночи
Озябший ветер меж старинных башен.

Что тебя я не люблю —
День и ночь себе твержу.
Что не любишь ты меня —
С тихой грустью вижу я.

О сладкая Мечта! о неба дар благой!
Средь дебрей каменных, средь ужасов природы,
Где плещут о скалы Ботнические воды,
В краях изгнанников... я счастлив был тобой.

Я ношусь во мраке, в ледяной пустыне,
Где-то месяц светит? Где-то светит солнце?
Вон вдали блеснула ясная зарница,
Вспыхнула — погасла, не видать во мраке,

Разбросаны в разбросанных столицах,
мы, одиноки и неисчислимы.
играли в первозданного Адама, 
дарующего миру имена.

Мой Телемак,
Троянская война
окончена. Кто победил — не помню.
Должно быть, греки: столько мертвецов
вне дома бросить могут только греки...
И все-таки ведущая домой
дорога оказалась слишком длинной.
как будто Посейдон, пока мы там
теряли время, растянул пространство.

Леночка, будем мещанами! Я понимаю, что трудно,
что невозможно практически это. Но надо стараться.
Не поддаваться давай... Канарейкам свернувши головки,
Здесь развитой романтизм воцарился, быть может, навеки.

Чудесное со мной не происходит. 
Как будто зареклось происходить.
Как будто крест поставило на мне
Чудесное. И вот что я придумал.

На берегу сидел слепой ребенок,
И моряки вокруг него толпились;
И улыбаясь он сказал: "Никто не знает,
Откуда я, куда иду и кто я,
И смертный избежать меня не может,
Но и купить ничем меня нельзя.[..."]

Всегда, к тому ж, терпимость равнодушна.
Ручаюсь я, что прежде двух годов
Весь мой народ, вся северная церковь
Признают власть наместника Петра.

Ты Петр был царем.
О слава дней минувших!
взлети как пламя трепанное ввысь.
А я уйду. Уйду с болот жестоких,
Прощай, Россия! навсегда прощай!

Горацио считает это все
Игрой воображенья и не верит
В наш призрак, дважды виденный подряд.

Улица, улица...
Тени беззвучно спешащих
Тело продать,
И забвенье купить,
И опять погрузиться 
В сонное озеро города — зимнего холода...

Спите. Забудьте слова лучезарных.

Новая Англия. Утро.
Сворачиваю в сторону Крэги.
И в тысячный раз вспоминаю,
что "Крэги" — шотландское слово,
а сам корень "крэг" — по истокам кельтский.

Третью неделю туман не слезает с белой
колокольни коричневого, захолустного городка,
затерявшегося в глухонемом углу
Северной Адриатики. Электричество
продолжает в полдень гореть в таверне.

Когда мне вдруг захочется
Увидеть сразу весь мир —
Я закрываю глаза.

Не правда ли — руки, когда они высоки
И пять пальцев на каждой
Взгляни на это дважды
Ради ветра и моря

Где ты ищешь счастья, моя богиня?
     Грозные скалы, шумные бури,
     Задумчивые закаты,
     Благоуханные рощи над воспетыми берегами.

До войны я носил фамилию Гангнус.
На станции Зима
               учительница физкультуры
с младенчески ясными спортивными глазами,
с белыми бровями
и белой щетиной на розовых гладких щеках,
похожая на переодетого женщиной хряка,

Однажды король остготов
посмотрел в зеркало
и увидел там
короля вестготов.
Устрашенный, он созвал диван мудрецов,
но ни один не смог истолковать сего сна.

Моряки старинных фамилий,
влюбленные в далекие горизонты,
пьющие вино в темных портах,
обнимая веселых иностранок;

Глядим на лес и говорим:
— Вот лес корабельный, мачтовый,
Розовые сосны,
До самой верхушки свободные от мохнатой ноши,
Им бы поскрипывать в бурю,
Одинокими пиниями,
В разъяренном безлесном воздухе.