Open Linguistics 2016; 2: 557–571

Research Article Open Access

Eleni Bozia*

Atticism: The Language of 5th-century

Oratory or a Quantifiable Stylistic
Phenomenon?
DOI 10.1515/opli-2016-0029
Received March 7, 2016; accepted October 27, 2016

Abstract: This paper comparatively explores Atticism as it first appeared in fifth-century Greek oratory
and was later revived by Imperial Greek authors. Using Dionysius’ of Halicarnassus and his appreciations
of oratory and orators as a frame of reference and then expanding his inferences on works of Imperial era,
I attempt to parameterize Atticism as a phenomenon. Ultimately this study will apprise us of the usage of
Atticism in Imperial Roman oratory as well, as it then becomes obvious that Atticism has transcended the
boundaries of language and has transformed into a constructional rhetorical system. This paper employs
a unified node-based metric formulation for implementing various syntactical construction metrics,
indicative of the syntactical attributes of the sentences. The developed metrics were applied to annotated
texts of six authors, which were then comparatively examined using Principal Component Analysis.

Keywords: Attic oratory, computational linguistics, treebanks, node-based metrics, Principal Component
Analysis

1 Introduction
5th- and 4th-century BCE orators, Lysias, Isocrates, Demosthenes, are the par excellence representatives of
Attic oratory and paragons of Atticism due to their proper usage of Attic forms and expressions. Asiatic
oratory ensues along with a general decadence in rhetorical productions that has been credited to the
influx of eastern elements until Atticism is revived in Imperial times. Dionysius of Halicarnassus, the
1st-century BCE teacher of rhetoric, attributes this literary upward shift to the Romans - either that means
a re-appreciation of Greek eloquence or implies a political vindictiveness that expresses itself through
language and literature on behalf of the Roman subjects against their rulers. The fact remains, though,
that there is an overwhelming production of grammar books and lexica focusing on Atticism, all the while
determining Greekness as well. Additionally, Cicero and Quintilian express their esteem of Atticism and
instruct Roman orators on how to achieve it. This acknowledgment alone is enough to put forward not
only an issue of political standing of Greek language and literature, but also to contend that Atticism
filtered through centuries and literary genres was modulated into a stylistic construction that transcends
language. Cleanness of expressions, comprehensiveness, cohesiveness and other structural attributes
alongside linguistic appropriateness in the language that is used in each case bear the tag of “Atticism.” In
this paper, my intention is to parameterize Atticism, attempting to determine attributive characteristics of

Article note: This paper belongs to the special issue on Treebanking and Ancient Languages, ed. by Giuseppe G.A. Celano and
Gregory Crane.

*Corresponding author: Eleni Bozia, University of Florida, Gainesville, FL United States; Alexander von Humboldt- Lehrstuhl
für Digital Humanities Institut für Informatik Universität Leipzig, Leipzig, Deutschland, E-mail: [email protected]

© 2016 Eleni Bozia, published by De Gruyter Open.

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivs 3.0 License.
558 E. Bozia

Lysias, Isocrates, and Demosthenes that have also been pinpointed by Dionysius of Halicarnassus in his
treatises on the orators. Additionally, Imperial orators - Lucian, Dio Chrysostom, and Aelius Aristides - will
be examined so as to determine whether their style is traditionally Attic or it is simply the usage of Attic
dialect that has led modern scholars to the conclusion that Imperial orators revived Atticism. Computational
linguistics, stylometry, and network analysis have been employed in several cases to pursue linguistic and
stylistic studies as well as author attribution. Passarotti et al. (2013) perform a lexical-based comparative
examination between Seneca, Cicero and Aquinas, employing Principal Component Analysis (PCA). Eder
et al. (2013) and Burrows (2002, 2006) used stylometry to achieve a multilevel analysis of texts. The critical
difference with the present study, though, is that stylometry does not account for syntactical characteristics
and constructions. Ferrer I Cancho et al. (2004, 2010) and Passarotti (2014) used network theory to study
linguistic constructions. Bamman et al. (2008, 2009), Passarotti (2010) and Mambrini et al. (2012, 2013)
examine issues in Latin and Ancient Greek dependency respectively. For detailed bibliographical references
on similar studies, see also Bozia (2015). In this study I utilize node-based metrics to parameterize Attic
constructions.
Section 1.1 furnishes a theoretical presentation of Attic dialect, its appearance as linguistic choice and
its subsequent appropriation of both linguistic and political preponderance. Section 1.2 then proceeds
to discuss the Imperial revival of Atticism as presented by Dionysius of Halicarnassus in his treatises on
oratory and the orators, determining whether there are syntactical attributes that could qualify an orator
as Attic that can be discerned in both classical Attic orators and their Imperial descendants. Section 2
presents the methods developed for the quantification of rhetorical style. Epideictic speeches of six orators
were syntactically annotated and then node-based metrics were developed in order to analyze the style of
each author. Section 3 discusses the logic behind the developed metrics, the way they work as numerical
descriptors of the sentences and the significance of the acquired results.

1.1 Atticism: one of the Hellenic dialects or a symbol of Greekness?

Close study of the works of lexicographers and grammarians reveal a number of unexpected constituents
regarding Atticism. Atticism was perceived as the Attic way of expression alongside the Aeolic, Doric, and
Ionic dialects. Diomedes, the grammarian, states that there are five languages of Greek - Ionic, Doric, Attic,
Aeolic, and Koine. (G.L. 1.4401)
Hence, contrary to the minute paradigms and extensive discussions of barbarism and solecism
(linguistic, pronunciation, and syntactical infelicities that on occasion are attributed to dialectical
deficiencies), Diomedes perceives the dialects as derivatives of the same language.
In a fragment of Posidippus, there is a blatant contradistinction between Atticism and Hellenism in
favor of the latter. The author strongly criticizes a user of Atticism as pretentious and too ornate. The point
is rather interesting especially if we consider that the same oppositional thesis rose between Atticism and
Asianism.

Even though there is one Hellas, there are, however, many cities.
You, on the one hand, Atticize whenever you speak such utterances of yours in this language. We, the Greeks, Hellenize.
Why you occupy yourself with syllables and letters and drag liveliness to odiousness? (Posidipp. 28)

In the Scholia ad Aristophanem 1A De Comoedia, Hellenism and Atticism are juxtaposed. The
distinctions listed are ten — analogy, etymology, figures, formation of nouns, allegory, numbers, genders,
breathings, tenses, and accents.
One needs to pinpoint that there is no evaluation of Hellenism and Atticism, or any remark
regarding their respective potency or refinement. However, it should be noted that the differences
being mentioned, such as differences in the gender of words, resemble the characteristics of
barbarisms and solecisms. Would that indicate that initially the evolution from the classical, typical

1 Grammatici Latini
  Atticism: The Language of 5th-century Oratory or a Quantifiable Stylistic Phenomenon? 559

Greek to Attic entailed alterations to the degree of language debasement or in the case of Atticism
over refinement? Therefore, it is interesting to consider that up to a point attikizein was appreciated as
differing from hellenizein. That realization prompts us to ponder on the gradual validation of Atticism
as representative of Greekness.
It seems that progressively Atticism, the most popular of the dialects, assumes an even more
independent role that especially during the Imperial era transcends the boundaries of linguistic usage.
Paramount literary works are written in Attic dialect, and Attica is the geographical area that through its
history embodies Hellenism. Swain (1996) and Whitmarsh (2001) among others discuss Imperial Attic
authors and Atticism in the Second Sophistic. Therefore, it is only reasonable that during the Empire
Atticism becomes the flagship promontory against Romans and Romanness, as Bowie (1970) suggests.
Furthermore, the appearance of Asianism, albeit succinct, needed a counterpart, and Atticism had
literary creations in its record that provided a balanced, and structurally formulated style to which
authors could resort. For discussions on Atticism and Asianism, see: (Rohde 1886), (Schmidt 1887),
(Radermacher 1899), (Wilamowitz 1900), (Norden 1915) and (Dihle 1977). (Goudriaan 1989), (Gelzer 1979)
and (Wisse 1995) also provide us with comprehensive overviews of the debates along with elucidating
parameterization of classicism and Atticism. There have also been comparative stylistic considerations
of Atticism and Asianism, attempting to consider the linguistic aspects of those movements. De Jonge
(2008) succinctly summarizes the debates against the backdrop of Dionysius of Halicarnassus. Another
significant turn in Atticism’s history is its modulation into a writing style beyond language. Cicero in
Brutus and De Oratore (Brut. 27ff; 284-285; De or. 3.28) and Quintilian in Institutio Oratoria (Inst.10.1.76-
80; 12.10.21-24) among others espouse Atticism, theorize upon its construction, characteristics and
merits and wish to teach it for the benefit of Roman oratory. O’Sullivan (1997) discusses Caecilius’,
Cicero’s, Quintilian’s, and Dio’s modulating canons of orators and their models of Attic historians.
Innes appraises the Attic revival in Rome as a “mutual cross-fertilization between Greeks and Romans”
(Innes 1989, 246).
Consequently, Atticism effectively appropriates legitimacy, encompassing linguistic propriety,
paideutic standing, and social status. Setting the paradigm, lexicographers start compiling forms and
examples of usages. The distinction, though, is not between Attic, Doric, Ionic, and Aeolic. Instead
it is between Attic, Hellenic, and Koine Greek. More specifically, Moeris sets the tone from the title
“Μοίριδος ἀττικιστοῦ λέξεις ἀττικῶν καί ἑλλήνων κατά στοιχεῖον.2” Moeris differentiates between
usages of the Hellenes, the Attics, and rarely the ancients. More specifically, he differentiates between
Attic versus Hellenic and common usage. Another contradistinction is Hellenic versus common and
common versus Attic (197.5, 198.2). However, the point, which intensely furnishes the spirit of the era,
is the opposition between primary Attic and secondary Attic (194.29). There is in fact a third occurrence
in which Moeris references a middle Atticism (213.2). (The references follow Boeker’s edition of Moeris)
Therefore, the appreciation of Dionysius of Halicarnassus for Atticism fits well within the socio-
political context of Atticism’s reevaluation. In his introductory statement Dionysius in De Antiquis
Oratoribus furnishes a comprehensive history of oratory, which provides an overview of the significant
historical points in the evolution of oratory always against the backdrop of socio-historical changes.

We ought to acknowledge a great debt of gratitude to the age in which we live, my most accomplished Ammaeus, for an
improvement in certain fields of serious study, and especially for the considerable revival in the practice of civil oratory. In
the epoch preceding our own, the old philosophic Rhetoric was so grossly abused and maltreated that it fell into a decline.
From the death of Alexander of Macedon it began to lose its spirit and gradually wither away, and in our generation had
reached a state of almost total extinction… Our own age has demonstrated this. Whether at the instance of some god,
or by the return of the old order of things in accordance with a natural cycle, or through the human urge that draws
many towards the same activities: for whatever reason, the ancient sober Rhetoric has thereby been restored to her former
rightful place of honor, while the brainless new Rhetoric has been restrained from enjoying a fame which it does not
deserve and from living in luxury on the fruits of another’s labors.3 (D. H. Orat.Vett. 1,2).

2 “Attic and Greek Diction of Moeris the Atticist according to the parts of speech.”
3 All translations of Dionysius of Halicarnassus are by Usher (1974)
560 E. Bozia

1.2 Defining Atticism

Considering the lexicographers and grammarians cited in the previous section, one concludes that Atticism
pertains both to linguistic and structural choices. However, the polarity between Hellenic and Attic and
primary and secondary Attic among others, render the quantification and parameterization of Atticism
problematic. A pivotal criterion, which is beyond the scope of this paper, lies in the time frame within which
each of the works was authored. The perceptions of what is current or ancient vocabulary and whether
Hellenic and Attic are synonyms alter according to the socio-political status quo.
Foregoing inner dialectical and political conflicts, should we attempt to take for granted and use as a
basis the conclusions of theoreticians, such as Dionysius of Halicarnassus, we might be able to determine
some generic Attic constructions—a process that could ultimately help us reverse engineer syntactical
metrics-markers for Atticism as a whole and then determine the minute dialectical differences as furnished
in the various definitions. To this end, Dionysius’ theoretical commentary is being considered alongside
syntactically annotated sentences of the respective authors, attempting to validate and ultimately quantify
his inferences.
Studying Dionysius’ treatises on the orators punctiliously, one should notice his underlying commentary
of Atticism and the modulations of style that seem to be in tune with the lexicographers’ appreciations of
the dialects.
More specifically, Dionysius considers Lysias the paragon of Atticism.

He is completely pure in his vocabulary, and is the perfect model of the Attic dialect—not the archaic dialect used by Plato
and Thucydides, but that which was in general currency in his day. (D.H. Lys. 2)

Upon closer study of the treatise, one needs to probe into what makes Lysias the model of Atticism. Dionysius
furnishes the readers with examples, such as the following:

Purity of language, correct dialect, the presentation of ideas by means of standard, not figurative expressions; clarity,
brevity, concision, terseness, vivid representation. (D.H. Lys. 13)

As to his composition, it is absolutely simple and straightforward. He sees that characterization is achieved not by periodic
structure and the use of rhythms, but by loosely constructed sentences. (D.H. Lys. 8)

Using Lysias as his framework, Dionysius proceeds to cognitively interpret and describe the other Attic
orators. Dionysius discusses Isocrates and comparatively studies his style with that of Lysias. Such a
comparative analysis gives us a more perspicuous account of Atticism, something that purveys us a
framework with attributes that would be deemed Attic.

His style has the following characteristics: it is as pure as that of Lysias; not a word is used at random; and the language
conforms closely to the most ordinary and familiar usage. (D.H. Isoc. 2)

I pronounced both to be masters of lucidity and vividness, but found Lysias the more successful in the concise expression
of ideas, and Isocrates the superior at rhetorical amplification…I judged Lysias to be the simpler in sentence-structure and
Isocrates the more elaborate; the former more convincing in creating the illusion of truth, the latter the more powerful
master of technique. (D.H. Isoc. 11)

Other interspersed proclamations regarding the two orators seem to be justified when we analyze their
sentence structure. Dionysius asserts that Isocrates’ diction is as pure as that of Lysias (D.H. Isoc. 2). Lysias
also excels in succinctness, while Isocrates in amplification.

On the other hand, it is compact, closely-knit style like the other…it sprawls and overflows with its own exuberance. (D.H. Isoc. 2)

The comparison is visualized in the figure below.

  Atticism: The Language of 5th-century Oratory or a Quantifiable Stylistic Phenomenon? 561

Fig 1. Syntactical comparison of Lysias’ Olympiacus 1.1 (top) and Isocrates’ De pace 3.1 (bottom).

The Treebank annotation furnished showcases that Lysias’ sentence is more tightly composed with just one
secondary clause that is parallel to the prepositional phrase introduced with ἕνεκα. The two sub-branches
depending on the coordination καί give the sentence a structured and balanced form. Isocrates, on the
other hand, after having introduced two major coordinations δέ, also explores the affluence of the Greek in
participles and with one coordination introduces three sub-brances of three supplementary participles, all
dependent on the same main verb.
Even though the Atticism of Isocrates is never questioned, Dionysius insists upon the need for
syntactical brevity. Dionysius proceeds with a profound examination of Isocrates’ style. He dissects the
562 E. Bozia

latter’s sentences, pinpointing parallelisms and rhyming clauses and constructions. He also draws the
reader’s attention to the symmetry of the clauses, which is not always regarded as a merit.

Which is artificial rather than natural. In three clauses of equal length is characteristic of Isocrates’s arrangement. (D.H.
Isoc. 20)

Such a treatise that combines the theoretical framework of oratory alongside technical aspects of actual
speeches further apprises us of the structural essence of Attic oratory. Additionally, in Isocrates 20
Dionysius offers his insight by rewriting some of the orator’s sentences. The first picture in the figure below
shows Isocrates’ sentence and the second Dionysius’ suggestion that opts for clarity and brevity. The major
disparity between the original and the rewrite is that the latter transformed an entire secondary clause,
which clearly elongated the sentence, to a single participle that condenses the same meaning within one
word. Even visually the syntactical tree is simpler and more comprehensive.

Fig 2. Syntactical comparison of Isocrates’ (left) and Dionysius’ (right) version.

Dionysius, when it comes to his discussion of Demosthenes, pursues what could be perceived as the first
form of computational linguistics and scientific inquiry in the field of literary analysis. He states that there
are three distinct types of stylistic diction—an elaborate that is best used by Thucydides, a simple whose
best representative is Lysias, and a third that is a perfect amalgamation of the previous two. Demosthenes
is the archetypal writer in this category alongside Isocrates and Plato.

This passage illustrates the striking, elaborate style which is remote from normality and is full of every kind of accessory
embellishment. Thucydides is the standard and pattern of this style, and no subsequent writer employed it to greater effect
or imitated him with complete success. (D.H. Dem. 1)

The second kind of style is plain and simple. Its artistry and power seem to consist in its resemblance to the language of
ordinary speech…The man who perfected it and realized its potential as a distinct style was Lysias the son of Cephalus.
(D.H. Dem. 2)

The third kind of style was a mixture formed by combining the other two…It is impossible to find any other writers, except
Demosthenes, who practiced the essential and ancillary virtues of this style to greater effect, or who expressed themselves
in more beautiful language and adorned it more skillfully with additional touches of artistry. (D.H. Dem. 3)
  Atticism: The Language of 5th-century Oratory or a Quantifiable Stylistic Phenomenon? 563

Another characteristic of Demosthenes’ constructive pattern is, according to Dionysius, the complexity
of his sentences, in which he begins with a clause, which is then interrupted in order to introduce a second,
which in turn is also broken unfinished, while a third is introduced.

In the first place, before rounding off the first idea (if it should be so called), a second idea is introduced; then a third
is subjoined before the second is complete, and material belonging to the second is tacked on after the third has been
completed. (D.H. Dem. 9)

Figure 3 demonstrates one of Demosthenes’ sentences syntactically analyzed. The complexity of the
sentence, the secondary clauses, and dependent phrases are on par with Dionysius’ apprehension of the
orator. Furthermore, the developed metrics discussed in this paper can contribute to the quantification of
the sentence’s complexity.

Fig. 3. Demosthenes, Philippica 1.1

Our analysis becomes more intriguing when we examine Imperial Greek oratory, in which the revival of
Atticism is manifest. Lucian of Samosata, the second-century CE orator, imitates classical Attic dialect and
prides himself in his acquired Greekness, or chastises himself whenever he commits a linguistic blander.

To be honest, however, their praise caused me considerable annoyance, and when they had gone and I was left alone, I
reflected as follows: “So this is the only attraction in my writings, that they are unconventional and keep off the beaten
track, while good vocabulary, conformity to the ancient canon, penetration of intellect, power of perception, Attic grace,
good construction, general competence, perhaps have no place in my work. Otherwise they would not have ignored these
qualities and praised only the novel and strange element in my style. I, fool that I was, had thought when they rose in
approbation that perhaps this particular feature too had some attraction for them — I remembered the truth of Homer’s
remark that the new song takes the fancy of an audience; but I did not think to attribute so much — indeed all of it — to
novelty, but supposed novelty to be a kind of additional ornament making some contribution indeed to the approbation
of my work, the audience’s real praise and commendation, however, going to those other qualities. As a result my elation
overstepped its bounds — to think I nearly believed them when they called me unique and in a class apart in Greece and
other flatteries of this kind. 4 (Zeux., 2)

There have also been extensive discussions on Dio’s and Aristides’ Atticism. Schmidt (1887-1897: i.72-191),
Swain (1996, 27-42; 187-241; 254-297), Whitmarsh (2001, passim) elaborate on implicit and explicit Atticism—
both on linguistic level and the assumption of different literary personae all imbued with different forms

4 Translation by Harmon (1913)

564 E. Bozia

and levels of Greekness. Lucian, Dio, and Aelius Aristides embrace Atticism in both their linguistic choices
and sentence structure. However, upon closer reading, one notices significant variations. Of course the
purpose of their delivered orations and the different intended audiences had also predetermined the
stylistic variations. How do we consider these variances, though? Should they be appreciated as forms of
Atticism or do they err significantly from the traditional constructions? Cicero was already one of the first to
elaborate on stylistic differences in Brutus, arguing in favor of the variegated nature of Atticism and against
branding any intricacy of style as Asiatic.

But if he bestows the name of Atticism on a half-starved, a dry, and a niggardly turn of expression, provided it is neat,
correct, and elegant, I cannot say, indeed, that he bestows it improperly; as the Attic orators, however, had many qualities
of a more important nature, I would advise him to be careful that he does not overlook their different kinds and degrees
of merit, and their great extent and variety of character. The Attic speakers, he will tell me, are the models upon which
he wishes to form his eloquence. But which of them does he mean to fix upon? for they are not all of the same cast. Who,
for instance, could be more unlike each other than Demosthenes and Lysias? or than Demosthenes and Hypereides?5 (Cic.
Brut. 285)

In this study I work on the same basis, as I attempt to establish a framework within which Atticism seems
to be working. A computational analysis therefore could provide us with measurable and appreciable
characteristics.

2 Methods: Quantifying Rhetorical Styles

In this section I develop metrics to quantify the syntactical structure of the sentence, parameterizing the
morphology of the tree. For the purpose of this analysis, epideictic speeches of the aforementioned orators
were annotated syntactically in the form of treebank annotation, as shown in figures 1 and 2, using the
Arethusa annotation framework through Perseids (www.perseids.org). I contend that this formational
consideration of rhetorical speeches will shed light on statistical distributions of syntactical phenomena
and consequently the evolution of oratory and the involvement of Atticism. More specifically, a more or
less intricate authorial style translates into the morphology of the tree, namely the width and depth of the
branches and the syntactical tags describing each lemma. In the next section a quantification framework is
presented using node-based sentence metrics.

2.1 Node-based sentence metrics

The structure of a syntactically annotated sentence is defined as a linearly ordered set of elements S={n1, n2,
… nk}, where each element ni is a tree node (ni ϵ T) and T denotes the space of tree nodes. It should be noted
that every tree node is a tree on its own, while also being the root of that tree. Therefore, a syntactically
annotated sentence is a linearly ordered set of as many trees as the words in the sentence. Furthermore, in
this space the operator children of is defined, which maps each node to a set of children nodes that are also
elements of the same sentence (children of: T→ {Tx}). A node without children is mapped to the empty set
through this operator (i.e. when x=0). The word order in a sentence defines the linear order of the tree nodes
in the set S. Additional operators can also be defined to implement other characteristics of the nodes, such
as syntactical tags, for example the mapping isATR: T→{0,1} could indicate if a given node is an attributive.
In order to quantify the use of Attic in this case, one needs to extract numerical descriptors for each
annotated sentence in a given corpus. Therefore, a set of metrics could be defined within the space S of
sentences that will then allow us to perform further comparative analyses. The syntactical morphology of
the sentence is depicted in the connectivity of the nodes. This paper explores the possibility to establish
a node-based metric so as to quantify the local morphology of each individual node and ultimately assess

5 Translation by Sutton and Rackham (1942)

  Atticism: The Language of 5th-century Oratory or a Quantifiable Stylistic Phenomenon? 565

the complexity of the sentence. A sentence metric is a function that maps each sentence to the space of real
numbers f: S→ℝ, where S denotes the previously defined space of syntactically annotated sentences.
A generalized sentence metric can be expressed as the weighted sum of node-based metrics:

(1)

where μ(ni) is a node metric that operates on node ni and computes a numerical value (μ: T → ℝ). The
weights wi determine each node’s degree of syntactical and positional contribution in the sentence. Figure
4 illustrates the evaluation of Eq. 1 on an abstract sentence with 7 words.
The form of node-based sentence metrics as defined in Eq. 1 is generic enough so that it can implement
a wide variety of sentence metrics that can quantify syntactically annotated sentences. The following
sections demonstrate in six detailed examples the construction of sentence metrics using Eq. 1.

Weights w5 Metric values μ5

w2 w4 μ2 μ4

w3 w6 w7 μ3 μ6 μ7

w1 μ1

FigWeights
4. Illustration of node-based
w5 sentence Metric values
metrics. Each node hasμan
5
assigned weight and a metric value. The total value of the
sentence metric is w1μ1+ w2μ2+...+ w7μ7.

Weights w2 1
w4 μ2
Metric values 1
μ4

2.2wSimple
3 w6 node-based
w7 sentence
μ3 metrics
μ6 μ7
1 1 1 1

A basic
1 example
1 of w
a11simple node-based1sentence μ1 is the one that calculates the number of words in a
1 metric
1
sentence. Such metric can be expressed in the form of Eq. 1 by setting wi=1 and μ(ni)=1 for all nodes in the
sentence. In this case,
1 the value of Eq. 1 will correspond
1 to the number of words in a given sentence as it is
shown in Figure 5.

Weights 1
Metric values 1

Weights 1 1
1 1
Metric values 0
1

1 1 1 1 1 1
1 1 0 0
1 ATR 1
1 1 1 0 1 0
ATR
Fig 5. Calculation of the1“number of words” as a node-based sentence
1 metric. In this example the result is
1x1+1x1+1x1+1x1+1x1+1x1+1x1=7.

Weights
Another simple example
1
of a node-based
Metricsentence
valuesmetric0
is the one that calculates the number of attributives
in a sentence. Such a metric can be expressed in the form of Eq. 1 by setting wi=1 for all nodes in the sentence,
μ(ni)=1 for the
1 attributive nodes,
1 and μ(ni)=0 for
0 the rest of the
0 nodes. In this case, the node metric μ(ni)
implements the is attributive metric and the value of Eq. 1 will correspond to the number of attributives in a
ATR
given1sentence
1 as it1is shown in Figure 6.0 1 0
ATR
1 1
 1 1

566 E. Bozia

Weights 1
Metric values 0

1 1 0 0
ATR
1 1 1 0 1 0
ATR
1 1

Fig 6. Calculation of the “number of attributives” as a node-based sentence metric. In this example the result is
1x1+1x0+1x0+1x0+1x0+1x1+1x0=2.

2.3 Relative sentence metrics

In order to quantitatively compare annotated sentences from one or more corpora, it is essential to be able
to express metrics as percentage of the number of the words in a sentence, as such numerical descriptors
can give a better indication of the complexity of a sentence. For instance, a big number of attributives
in a relatively small sentence could signify a more intricate writing style. Such relative sentence metrics
can be defined in the same way as the previous examples. An example of a relative node-based sentence
metric is the one that calculates the percentage of leaves in a sentence. Such metric can be expressed in
the form of Eq. 1 by setting wi=1/k, where k is the number of the words in a sentence, μ(ni)=1 for the leaves,
and μ(ni)=0 for the rest of the nodes in the sentence. In this case, the node metric μ(ni) implements the
is leaf metric and the value of Eq. 1 will correspond to the percentage of leaves in a given sentence as it is
shown in Figure 7.

Weights 1/7 Metric values 0

1/7 1/7 0 1

1/7 1/7 1/7 1 1 0

1/7 1

Fig 7. Calculation of the “percentage of leaves” as a node-based sentence metric. In this example the result is 1/7x1+1/7x0+1/
7x1+1/7x1+1/7x0+1/7x1+1/7x0=4/7.
Weights 1 Metric values 4
Similarly, the example in Figure 6 can be expressed as a relative sentence metric by setting wi=1/k for all
the nodes in
0 the sentence.
0 In this case the value
3 of Eq. 1 will1correspond to the percentage of attributives in
a sentence.
0 0 0 1 1 2

2.4 Recursive0sentence metrics 1

Although the previous examples demonstrated the construction of basic sentence metrics using Eq. 1, more
complex metrics can be defined by setting the weights wi and the node metric μ accordingly. It should
be noted that, despite the linear form of Eq. 1, non-linear metrics can also be established, using wroot=1,
wi≠root=0, and setting μ(nroot) to be a non-linear function that can operate on the entire sentence tree by
Weights
recursively traversing Metric values
1/7 it from the root. 3

0 0 3 0

0 0 0 0 0 1

0 0
Weights 1/7 Metric values 0

 1/7 Atticism:
1/7 The Language of 5th-century
0 Oratory
1 or a Quantifiable Stylistic Phenomenon? 567

1/7 1/7 1/7 1 1 0

An example of a recursive sentence metric is the one that calculates the height of the syntactical tree of
a sentence. Such1/7 metric can be expressed in the form1 of Eq. 1 by setting wroot=1 and wi≠root=0 for the rest of
the nodes in the sentence. The node metric μ(ni) will recursively implement the height of the tree metric for
Weights
the sub-tree ni. In1/7 Metric
this case, the value values
of Eq. 1 will correspond
0 to the height of the syntactical tree of a given
sentence as it is shown in Figure 8.
Weights 1/7 1/7 0
Metric values 1
1 4
1/7 1/7 1/7 1 1 0
0 0 3 1
1/7 1
0 0 0 1 1 2

0 1

Weights 1 Metric
Fig 8. Calculation of the “height of the tree” values sentence
as a node-based
4
metric. In this example the result is
0x1+0x3+0x1+0x1+1x4+0x1+0x2=4.

0 0 3 1
Another example of a recursive sentence metric is the one that calculates the relative size of the largest
Weights 1/7 Metric values 3
family
0 in a0syntactical
0 tree. Similarly to
1 the previous
1 2 example, such metric can be expressed in the form
of Eq. 1 by setting wroot=1/k, where k is the number of words in the sentence and wi≠root=0 for the rest of the
0 0 0 3 1 0
nodes in the sentence. The node metric μ(ni) will recursively implement the largest family of the tree metric
for the sub-tree ni. In this case, the value of Eq. 1 will correspond to the size of the largest family in the
0 0 0 0 0 1
syntactical tree of a given sentence calculated as percentage of the total number of words in the sentence
as it is shown in Figure
0 9. 0

Weights 1/7 Metric values 3

0 0 3 0

0 0 0 0 0 1

0 0

Fig 9. Calculation of the “size of the largest family as percentage of the words”. In this example the result is 0x1+0x3+0x1+0x1
+1/7x3+0x1+0x1=3/7.

2.5 Computer implementation

This analytical framework was implemented in JavaScript/HTML5 as a cross-platform programming

environment. The developed interface includes API for defining custom node-based sentence metrics and
is compatible with the Ancient Language Dependency Treebank (ALDT) format version 1.5 currently used
by Perseids. In this interface a custom metric is defined as:
1. var example=new NodeMetric(“Num of attrib.”);
2. example.weight=function(node){return 1;};
3. example.metric=function(node){return
4. node.getRelation()==“ATR”;};
5. var value=example.apply(sentence);

The above example shows a case in which wi=1 and μ(ni)=1 if ni is ATR (attributive), or 0 if otherwise. This
implements the node-based sentence metric demonstrated in Figure 6. The last line in the example shows
568 E. Bozia

how a metric can be applied to a Treebank sentence, which is given as a TreebankSentence object variable
(here named sentence).

3 Results
The purpose of this framework is to set certain metrics and apply them on a collection of treebanks.
This process will produce several numerical descriptors of each sentence that can quantify syntactical
construction. These numbers can then be used as an input in any classification or pattern analysis algorithm,
such as Principal Component Analysis, to examine similarities between works, authors, and writing styles.
For this pilot study, approximately one hundred sentences were annotated, using the Perseids Treebank
Annotator. Five node-based metrics were implemented, using the presented framework, which are the ratios
of the: a) number of leaf nodes, b) number of ATR nodes, c) tree height, d) tree width, and e) maximum
number of branches in a single node, to the total number of nodes in the sentence. Considering that this
pilot study was conducted on a relatively small corpus, in an attempt to produce trustworthy results, the
metrics that were used were selected as robust descriptors of the morphology of the sentence and were not
prone to noise deriving from local syntactic variations. Using these metrics, each sentence became a point
in a 5D space. This point cloud was processed using PCA in Matlab to map the data onto the plane of the
largest spread as shown in Figure 10. The plane of the largest spread was computed using the principal
and secondary Eigen axes calculated by the PCA. Table 1 presents the contribution of each metric to the
principal and secondary Eigen Axes.

Table 1. The contribution of each metric to the computed Eigen axes

Node-based Sentence Metrics Principal Eigen Axis Secondary Eigen Axis

Relative Number of Leaves 18.96% 24.97%

Relative Height 7.57% 50.78%

Relative Width 34.06% 0.14%

Relative Largest Family Size 35.71% 4.61%

Relative Number of Attributives 3.70% 19.50%

Fig 10. Plots of the sentence dataset on the plane of the two dominant eigenvectors. Each sentence is represented by one
point (circle) in this plot.
  Atticism: The Language of 5th-century Oratory or a Quantifiable Stylistic Phenomenon? 569

The plots in Figures 10 and 11 clearly furnish the affinities or lack thereof between the discussed orators.
There are three different viewpoints from which to examine and interpret these plots. Starting with individual
orators, the plots indicate affinities between Lysias and Lucian. The latter proclaims in several places in his
work that he opts and strives for Attic writing style. His dialect is distinguished for its Attic purity, even though
the orator dates in the second century CE. Therefore, the sentence plot in Figure 11 is an attestation of the
close proximity between Lysias’ and Lucian's ellipses. Isocrates is the other ellipse that coincides with the
aforementioned authors. Based on the analysis presented in Section 2, Isocrates, albeit less succinct, is still
regarded for the cleanness of his expression. On the other hand, Dionysius’ of Halicarnassus explication of
the style of Demosthenes as a perfect mixture of simple and florid expects the locality of the orator in the
figure. Additionally, Dio converges significantly with Demosthenes, Lysias and Aelius Aristides, which can be
explained when one considers that Dio constantly alters his style according to his intended audience, but also
the fact that Dio as well as Demosthenes and Aristides had clearly politicized their rhetorical practice. Finally,
Aelius employs Attic constructions, having, however, resorted to a more convoluted form of expression.
Hence, he does not relate closely with the majority of the other orators.

Fig 11. Visualization of the point set of Fig.9 as Gaussian distributions. The distribution of the sentences of each author is
depicted as an ellipse. The center of each distribution is marked by a circle, and the median is marked by an asterisk.

The second viewpoint from which to consider the orators is in groups—the classical Attic and the Imperial
orators. It is interesting to note that Demosthenes and Isocrates do not coincide, but Lysias is the connecting
ellipse between the two. This showcases that, albeit classical Attic orators, they have distinct writing styles.
They are, however, undeniably Attic, and the affinities with Lysias, the quintessential Atticisist, prove that
they share common traditional elements. Close examination of the Imperial orators proves that they also
have distinct similar elements. All three ellipses coincide considerably, which would mean that the revived
Atticism opted for standardized schemas for self-validation. In the future, isolating certain constructions
and determining whether they are particular to Atticism in general, classical, or Imperial Atticism will
further this analysis.
The third approach to the plots is the apparent unity of Atticism. It should be considered that regardless
of stylistic differences and the ultimate attempt to establish a syntactical territory for each one of these
authors, hence parameterizing Atticism, a major point that is obvious through the plot is that there is no
author who is entirely disconnected from the group, thus reinforcing the idea that there was indeed an Attic
framework which was first established by the 5th- and 4th-century Attic orators and was then recanonized
and revived in Imperial times. Additionally, we should not discount the possibility that the metrics set for
570 E. Bozia

this study may have not be adequately discriminating. On the one hand, we can establish that Atticism is
a formalized phenomenon, but we also need to define different metrics, pertaining to certain syntactical
constructions, that could perhaps be more descriptive of each author. Consequently, the structural
analysis of Atticism—classical and Imperial—will be further dissected, and our understanding will be more
percipient.
To conclude, in this paper a framework was presented for the analysis of stylistic similarities in certain
5th- and 4th-century and Imperial Attic orators. A limited pilot study was performed as a proof of concept.
In the future, I intend to annotate a larger amount of those authors’ epideictic speeches and create more
complex metrics that will serve as more discrete descriptors of stylistic patterns. Therefore, the comparisons
will apprise us of the stylistic attributes of certain authors that are considered seminal in the formation of
Atticism and of others who have revived Atticism, using it as a medium for the entire literary movement
of the Second Sophistic in the Imperial Era. Subsequently, I will determine whether 5th- and 4th-century
Attic and then respectively Imperial writers utilize certain preset motifs that would validate them as Attic
as well as whether there are features that distinguish Classical from Imperial Atticism. Ultimately, I expect
to identify constructional characteristics of Atticism as a phenomenon, something that will enhance our
understanding of Atticism as a linguistic and rhetorical current.

References
Bamman, David, Passarotti Marco, Busa Roberto, Crane Gregory. 2008. The Annotation guidelines of the Latin Dependency
Treebank and Index Thomisticus Treebank. In: Proceedings of the 5th SaLTMiL Workshop. 26 May-1 June 2008,
Marrakech, Morocco. pp. 71-76.
Bamman, David, Mambrini Francesco, Crane Gregory. 2009. An Ownership Model of Annotation: The Ancient Greek
Dependency Treebank. In: M. Passarotti, A. Przepiórkowki, S. Raynand, F. van Eynde (eds.) Proceedings of the 8th
International Workshop on Treebanks and Linguistic Theories. 3-5 December 2009, Milan, Italy. pp. 5-15.
Bozia, Eleni. 2015. Measuring Tradition, Imitation, and Simplicity: the case of Attic Oratory. In: F. Mambrini, M. Passarotti,
C. Sporleder, (eds.), Proceedings of the Workshop on Corpus-Based Research in the Humanities. 10 December 2015,
Warsaw, Polland. pp.23-29.
Bowie, Ewen. 1970. Greeks and their Past in the Second Sophistic. Past and Present 46, pp. 3-41.
Burrows, John Frederick. 2002. ‘DELTA’: a measure of stylistic difference and a guide to likely authorship. Literary and
Linguistic Computing 17(3), pp. 267-287.
Burrows, John Frederick. 2006. All the Way Through: Testing for Authorship in Different Frequency Strata. Literary and
Linguistic Computing 22(1), pp. 27-48.
De Jonge, Casper. 2008. Between Grammar and Rhetoric. Brill: Leiden and Boston.
Dihle, Albrecht. 1977. Der Beginn des Attizismus. A&A 23, pp.162-177.
Eder, Maciej, Kestemont Mike, Rybicki Jan. 2013. Stylometry with R: a suite of tools. In: Digital Humanities 2013 Conference
Abstracts. Lincoln 2013, pp. 487-489.
Ferrer I Cancho, Ramon, Solé Ricard, Köhler Reinhardt. 2004. Patterns in Syntactic Dependency Networks. In: Physical Review
E69 (5), pp. 051915(1)-051915(8).
Ferrer I Cancho, Ramon. 2010. Network Theory. In: P.C. Hogan (ed.). The Cambridge Encyclopedia of Language Sciences.
pp.555-557.
Gelzer, Thomas. 1979. Klassizismus, Attizismus und Asianismus. In: O. Reverdin, B. Grange (eds.) Le Classicisme à Rome aux
1ers siècle avant et après J.-C.. Entetriens sur L’Antiquité Classique. Tome XXV. Vandeoeuvres-Genève. pp. 1-41.
Goudriaan, Koen. 1989. Over classicisme. Dionysius van Halicarnassus en zijn program van welsprekendheid, cultuur en
politiek. Unpublished diss. Vrije Universiteit Amsterdam.
Innes, David. 1989. Augustan Critics. In: G.A. Kennedy (ed.). The Cambridge History of Literary Criticism. Vol.I: Classical
Criticism. Cambridge. pp. 245-273.
Hajič, Jan. 1998. Building a syntactically annotated corpus: The Prague Dependency Treebank. In: E. Hajičová (ed.), Issues of
Valency and Meaning. Studies in Honor of Jarmila Panevová. Prague Karolinum, Charles University Press. pp.12-19.
Harmon, Austin Morris. 1913. Lucian. Vols.1-8. Harvard University Press: Cambridge, MA; London.
Kiel, Heinrich. 1822-1894. Grammatici Latini. Vols.7. Teubner: Leipzig.
Mambrini, Francesco, Passarotti Marco. 2012. Will a Parser Overtake Achilles? First Experiments on Parsing the Ancient Greek
Dependency Treebank. In: I. Hendrickx, S. Kübler, K. Simov (eds.) Proceedings of the 11th Workshop on TLT. 30 November-1
December 2012, Lisbon, Portugal. pp.133-144.
  Atticism: The Language of 5th-century Oratory or a Quantifiable Stylistic Phenomenon? 571

Mambrini, Francesco, Passarotti Marco. 2013. Non-projectivity in the Ancient Greek Dependency Treebank. In: E. Hajičová,
K. Gerdes, L. Wanner (eds.) Proceedings of the 2nd International Conference on Dependency Linguistics. 27-30 August
2013, Prague, Czech Republic. pp.177-186.
Norden, Eduard. 1898. Die antike Kunstprosa vom VI. Jh. V. Chr. Bis in die Zeit der Renaissance. Leipzig.
O’Sullivan, Neil. 1992. Alcidamas, Aristophanes and the Beginnings of Greek Stylistic Theory. Stuttgart.
Passarotti, Marco. 2010. Leaving Behind the Less-resourced Status. In: K. Sarasola, F. M. Tyers, M. L. Forcada (eds.)
Proceedings of the 7th SaLTMiL Workshop, 23 May 2010, Malta. pp.27-32.
Passarotti, Marco. 2013. A Statistical Investigation Into the Corpus of Seneca, In: P. Poccetti (ed.) Proceedings of The 17th
International Colloquium on Latin Linguistics, 20-25 May 2013, Rome, Italy. pp. 1-16.
Passarotti, Marco. 2014. The Importance of Being sum. Network Analysis of a Latin Dependency Treebank. In: R. Basili, A.
Lenci, B. Magnini (eds.) Proceedings of La Prima Conferenza Italiana di Linguistica Computazionale. 9-11 December 2014,
Pisa, Italy. pp. 291-295.
Radermacher, Lambertus. 1899. Studien zur Geschichte der antiken Rhetorik IV: Ueber die Anfänge des Atticismus. RhM 54.
pp. 351-374.
Schmidt, Wilhelm. 1887-1897. Der Atticismus in seinem Hauptvertreten (5 vols.). Stuttgart.
Swain, Simon. 1996. Hellenism and Empire. Oxford: Clarendon Press
Sutton, Edward William, Rackham, Harris. 1942. Cicero, Marcus Tullius. Cambridge, Mass.: Harvard University Press.
Usher, Stephen. 1974. Dionysius of Halicarnassus. Cambridge, MA: Harvard University Press.
Wilamowitz-Moellendorff, Ulrich. 1900. Asianismus und Attizismus. Hermes 35, pp. 1-52.
Wisse, Jacob. 1995. Greeks, Romans, and the Rise of Atticism. In: J.G.J, Slings, S.R., Sluiter, I. (eds.) Greek Literary Theory After
Aristotle. Abbenes. VU University Press: Amsterdam. pp. 65-82.
Whitmarsh, Tim. 2001. Greek Literature and the Roman Empire. Oxford: Clarendon Press.