Matthew Stephenson

Walter Crist
College of Science and Engineering, Flinders University,
Centre for the Arts in Society, Leiden University, Leiden, the Adelaide, Australia [email protected]
Netherlands
[email protected]
Cameron Browne
Éric Piette Department of Advanced Computing Sciences, Maastricht
University, Maastricht, the Netherlands
ICTEAM, UCLouvain, Louvain-la-Neuve, Belgium
[email protected] [email protected]

Dennis J. N. J. Soemers
Department of Advanced Computing Sciences, Maastricht
University, Maastricht, the Netherlands
[email protected]

Computational Approaches for

Recognising and Reconstructing
Ancient Games: The Case of
Ludus Latrunculorum
1. INTRODUCTION ancient counterparts or chosen for other reasons.
THE DIGITAL LUDEME PROJECT This paper uses the example of the ancient Roman
1.1 Approaches to Game Reconstruction game Ludus Latrunculorum to demonstrate the ways
that a computational approach can contribute to the

T
he existence of board games in the ancient processes of identifying games in the archaeological
world has been documented from the early record and reconstructing playable rulesets.3
days of archaeology, and along with the
discovery of boards and their identification with 1.2 The Digital Ludeme Project
games known from written texts, archaeologists The Digital Ludeme Project is a five-year research
and others interested in games have sought to project that seeks to improve our understanding of the
produce playable rulesets to bring these ancient development of (primarily) board games throughout
practices back to life. Some authors who have recorded human history, through computational
attempted to reconstruct the rules for ancient analyses of the available evidence. This task is made
games have paid very close attention to the ancient challenging by the paucity and incomplete nature of
sources,1 while others have been more creative in evidence for ancient games and the unreliable nature
their interpretations.2 Usually, there is a degree of of much of the information that is available about
playtesting involved, to produce a game that seems them, which is often based on interpretations of the
to play well. Nevertheless, these reconstructions are source material filtered through a modern (Western)
subjective to varying degrees, and it is not always lens. Another factor is that the rules for games are
clear whether the rules that have been included in typically carefully crafted and fragile to change,
a particular reconstructed game are inspired from and there is no guarantee that even well researched
games that may be reasonably connected to their reconstructions of ancient rules that are entirely
plausible within their historical and cultural contexts
1 e.g., Schädler (2001); Schädler (1994).
2 e.g, Becq de Fouquières (1869): 454–456; Bell (1979): 86–87. 3 Browne et al. (2019).

GAMES IN THE ANCIENT WORLD: PLACES, SPACES, ACCESSORIES, ALESSANDRO PACE, TIM PENN, ULRICH SCHÄDLER (EDS) | 63
will actually play well as games. They can be related to the equipment (board or
pieces), or rules (game setup, ending condition(s),
The Digital Ludeme Project promotes a new field of
or movement of the pieces), but also to more
research called Digital Archaeoludology4 in which
general aspects such as properties (e.g., time model,
games are digitally modelled as accurately as possible
information type, symmetries). The current list of all
from the available evidence, then missing information
identified concepts used through Ludii is available
about them is interpolated probabilistically based on
on our website.9
how well the candidate rulesets function as games
in addition to their authenticity within the given Thanks to the ludeme representation of games in
historical and cultural contexts. Ludii, during the compilation process, the existence
of specific ludemes or combinations of ludemes
This chapter demonstrates these principles in action.
trigger each of these concepts. Binary concepts are
Using the historical and cultural context to identify
activated by the existence of ludemes while numerical
plausible rule combinations for the reconstruction of
concepts instead have their values set when the
Ludus Latrunculorum, we then apply computational
game is compiled. Consequently, all concepts are
self-play analyses of the resulting candidate rulesets
computed in a few milliseconds. For this reason, they
to identify those that work more successfully as
are a powerful tool to identify similar games.
games. While this approach may not necessarily
reveal how the game was actually played, it can
indicate with some confidence how it was not played, 2. LUDUS LATRUNCULORUM: THE EVIDENCE
which can be just as useful to know. 2.1 Documentary Evidence for Rules
1.3 Ludii Ludus Latrunculorum, or the “Game of Little
Soldiers”, is one of the games played by the Romans
Ludii is a computer program that was developed as about which we have the most documentary
part of the Digital Ludeme Project that constitutes evidence, since several authors seem to have been
a complete general game system for modelling, familiar enough with the game to mention portions
playing, evaluating and reconstructing the full of the rules. A complete recounting of the textual
range of games required for the project. Games are references to the game is beyond the scope of this
modelled digitally as structured sets of ludemes5 to paper, and has been covered extensively elsewhere.10
provide a playable database of both complete games Nevertheless, there are four aspects of the game that
and partial descriptions requiring reconstruction, can be gleaned from the sources.
suitable for computer self-play by artificial intelligence
(AI) agents which can playtest candidate rulesets Placement phase: Two ancient sources confirm that
much more quickly than human players. the board is empty at the beginning of the game. In
Laus Pisonis, by an anonymous author, there is a
The Digital Ludeme Project games database6 passage which praises Piso in the context of a game
compiles what is known about traditional games that holds the same military symbolism as other
from the past five thousand years, including descriptions of the game.11 It is explicitly stated that
geographic, chronological, social, and rules data for the pieces are cunningly placed on an open board.12
1,006 games at the time of writing.7 Compiling all Isidore, writing in the late 6th or early 7th century CE,
of this data facilitates comparison between games, indicates three kinds of pieces or, more precisely, the
particularly in connecting games which are close to states of movement of three kinds of pieces: ordine,
each other geographically, chronologically, and with which move regularly, vagi, which move anywhere,
respect to game concepts. In the future, the process of and inciti, which cannot move.13
connecting games according to these measurements
will be automated, but for this study with a limited Orthogonal movement: Ovid, in his Tristia, indicated
scope an ad hoc approach must be taken and is that the pieces on the board are moved in a straight
sufficient to make the necessary connections. line.14

1.4 Game Concepts Custodial capture: The next phrase in Ovid’s Tristia
states that a piece between two enemies is lost.15 Ovid
Game concepts8 are features designating high-level
also makes this clear in Ars Amatoria, where he says
aspects of games which can be shared between
different games. Each concept is expressed in game-
terms commonly used by players and designers.
9 Accessible at https://ludii.games/searchConcepts.php.
4 Browne et al. (2019). 10 Schädler (1994).
5 A ludeme is a game-related concept that can be expressed as a func- 11 Anonymous, Panegyric on Piso, 192–193.
tional unit. See Parlett (2016). 12 Austin (1934), 30; Schädler (1994), 55–54.
6 Accessible at https://ludii.games/library. 13 Isidore, Origins, 18.67; Schädler (1995), 82.
7 Stephenson, Crist, Browne (2020). 14 Ovid, Sorrows, 2.477; Schädler (1994): 52.
8 Piette et al. (2021). 15 Ovid, Sorrows, 2.477; Schädler (1994): 52.

64 |COMPUTATIONAL APPROACHES FOR RECOGNISING AND RECONSTRUCTING ANCIENT GAMES: THE CASE OF LUDUS LATRUNCULORUM
that one counter perishes between twin foes.16 This
form of capture is commonly known as custodial
capture in games literature.17
Quadrangular board: Varro, writing about the
Latin language in the 1st century BCE, describes a
declination table for the word albus, with horizontal
and vertical lines, as similar to the game board used
to play Ludus Latrunculorum.18 This indicates that
the game was played on a rectangular or square
board (i.e., Types 29 and 30—see Appendix to this
volume), but does not specify its size. Schädler points
out that a declination table would have six lines for
the six cases of Latin nouns, and six columns for the
singular and plural for masculine, feminine, and
neuter genders.19
2.2 Archaeological Evidence for Boards
Despite Varro’s implication, only one six-by-six
board has been found at a Roman site. Therefore, we
must assume that Varro’s description refers generally
to a square or rectangular board, rather than one of
a specific size. Archaeological evidence, then, must
be consulted to provide information about the size of
the quadrangular grid that likely formed the board of
Ludus Latrunculorum. Fig. 1. Four Roman Ludus Latrunculorum boards held at Dover Castle.
Clockwise from top left: Intact 8 x 8 chalk board (14.5 x 26 x 10 cm), from
Quadrangular boards within the limes of the Dover, 2nd c. CE?, Dover Castle DV2000; fragmentary marble board with
8 or 9 x 7+ squares (26 x 28 x 4.5 cm), from Richborough Fort, Roman
empire have been found in many different kinds of in date, Dover Castle 78305715; fragmentary marble board with 10+ x
archaeological contexts (Figure 1). A considerably 5+ squares (19.2 x 23 x 4.2 cm), from Richborough Fort, Roman in date,
Dover Castle 78305708/478; fragmentary sandstone board with 6+ x 5+
large number have been found at fortresses and squares (14.5 x 10.3 x 2.6–3.3 cm), from Richborough Fort, after 280 CE,
military installations, particularly in the provinces Dover Castle 78303093. Photos by Summer Courts.
of Britannia,20 Hispania Terraconensis,21 and
Aegyptus,22 as well as in urban contexts, such as in Zurich,28 appears to have been manufactured as
within Rome itself, Athens,23 and the cities of Asia a board for Ludus Latrunculorum. It is questionable
Minor,24 in spaces devoted to leisure such as the whether all of the grids on ceramic tiles were meant
Antonine Baths at Carthage,25 as well as rural sites to have functioned as game boards; it has also been
such as Gebel el-Silsila in Egypt26 and Thornham, noted that ceramic tiles may have been scored in
England.27 the same manner to aid in fixing them to vertical
surfaces.29 Wooden boards may also have been used
The boards that have been preserved are made of to play Ludus Latrunculorum, but they have not
stone or ceramic. Ludus Latrunculorum tends to survived in the archaeological record.
appear as secondary use on stone surfaces, whether
as graffiti or as a reappropriated object on which the In addition to artifacts used for playing, there are
board is carved. Only one example, formerly held two terracotta models of quadrangular boards which
appear to represent Ludus Latrunculorum. One
comes from Athens, with a group of players sitting
16 Ovid, The Art of Love, 358. around a six-by-seven board.30 Another, found in
17 Bell (1979), 78; Parlett (1999), 199; also called interception capture, Egypt’s Fayyum Oasis, shows a six-by-seven board
Murray (1951), 10. with pieces arrayed on it.31
18 Varro, On the Latin Language, 10.22.
Judging from the boards with an intact gaming
19 Schädler (2007), 361.
pattern, it is apparent that there is a wide range of
20 Austin (1934), 26–27; Schädler (1994), 50; Pace 2015; Pace 2020;
Pace 2022; Courts, Penn (2019). sizes for quadrangular boards from Roman sites. The
21 Carretero Vaquero (1998); Fernández Pintos (2017), 234–240. twenty-five known intact boards (see Figure 2) show
22 Mulvin, Sidebotham (2004), 611–612. that they range in size from six-by-six to eleven-by-
23 Schädler (1994), 49–50.
24 Bell (2007), 98; Ersoy, Erdin (2015), 151. 28 May (1991), 174–175.
25 De Voogt (2019), 91. 29 Courts, Penn (2019): 6–7.
26 De Voogt, Nilsson, Ward (2020), 127–128. 30 Michaelis (1863); Schädler (1994), 53.
27 Gregory, Gurney (1986), 13. 31 Petrie (1927), 55.

GAMES IN THE ANCIENT WORLD: PLACES, SPACES, ACCESSORIES, ALESSANDRO PACE, TIM PENN, ULRICH SCHÄDLER (EDS) | 65
Location Size Reference
Aventicum 6x6 Roman Museum of Avenches 13/15890-01
Athens (model) 6x7 SCHÄDLER (1994): 53
Monte de Santa Tegra 6x7 FERNÁNDEZ PINTOS (2017): 235-236
Aventicum, 6x8 Roman Museum of Avenches X/02924
Qasr Ibrim 7x8 ROSE (1996): 160
Chesters Fort 7x8 AUSTIN (1934): 27
Villa Adriana 7x8 MANDERSCHEID et al. (2011): 514-518
Corbridge 7x8 MURRAY (1913): 30
Gebel el-Silsila 7x8 DE VOOGT et al. (2020): 127
Parthenon, Athens 8x8 KARAKITSOU (2009): 24
Chedworth 8x8 BADDESLEY (1925)
Exeter 8x8 HOLBROOK and BIDWELL (1991): 278
Miletus 8x8 BELL (2007): 98
Smyrna 8x8 ERSOY and ERDIN (2015)
Smyrna 8x8 ERSOY and ERDIN (2015)
Brescia 8x8 MOSCA and PUPPO (2012): 274–275, 278
Basilica Iulia, Rome 8x8 SCHÄDLER (1994): 49-50
Xanthus 8x8 BELL (2007): 98
Kom Ombo 8x9 CRIST et al. (2016): 140-141
Gebel el-Silsila 8x9 DE VOOGT et al. (2020): 127
Carthage 9x10 DE VOOGT (2019): 91
Dover 10x10 PHILIP (1981): 167
Samos 11x12 SCHÄDLER (2007): 361
Puig Castellar de Bosca 11x16 RODRIGO REQUENA and ROMANÍ SALA (2021)
Fig. 2. Intact quadrangular boards found within the borders of the Roman Empire.
Location Preserved Grid Reference
Singidunum 2x2 JANKOVIC (2009): Fig. 11
Cemenelum 2x5 MOSCA and PUPPO (2012): 274–275
Saldum 3x3 JEREMIC (2009): 50–51
Didymoi 3x3 BRUN (2011): 126, 153
Cumidava 3x4 GUDEA and POP (1971): Pl. LII.3
Singidunum 3x4 JANKOVIC (2009): Fig. 12
Didymoi 3x4 BRUN (2011): 121, 143
Petavonium 4x1 CARRETERO VAQUERO (1998): Fig. 2.6
Petavonium 4x2 CARRETERO VAQUERO (1998): Fig. 2.4
Conimbriga 4x4 DA PONTE (1986): 138-139
Halpići 4x4 BUSULADŽIĆ (2017): Pl. 3
Petavonium 4x4 CARRETERO VAQUERO (1998): Fig. 2.3
Abu Sha’ar 4x5 MULVIN and SIDEBOTHAM (2004): Fig. 9.3
Abu Sha’ar 4x7 MULVIN and SIDEBOTHAM (2004): Fig 9.2
Didymoi 4x7 BRUN (2011): 126, 153
Petavonium 5x3 CARRETERO VAQUERO (1998): Fig. 2.5
Vindolanda 5x5 PENN and COURTS (n.d.)
Krokodilo 5x7 MATELLY (2003): 594, 605
Abu Sha’ar 5x9 MULVIN and SIDEBOTHAM (2004): Fig. 9.1
Exeter 6-7x8-9 HOLBROOK and BIDWELL (1991): Fig. 135
Abu Sha’ar 6-8x8-9 MULVIN and SIDEBOTHAM (2004): Fig. 9.5
Bearsden 6x4 BREEZE (2016): 93–94
Aventicum 6x6 DANIAUX (2019): Fig 3
Richborough 6x7 BUSHE-FOX (1928): Fig. 1.3
Aquis Querquennis 6x7 AVELAIRA (2015): 66
Noville 6x7 PÉREZ LOSADA (1993): 1058, Fig. 3
Buciumi 7x3 GUDEA (1971): Pl. LVI.4
Vindolanda 7x9 PENN and COURTS (n.d.)
Carthage 8x10-11 DE VOOGT 2019: 91
Richborough 8x5 BUSHE-FOX (1928): Fig. 1.2
San Chuis 8x5 VILLA VALDÉS (2010–2012): 109–110
Abu Sha’ar 8x6 MULVIN and SIDEBOTHAM (2004): Fig. 9.4
Petavonium 8x7 CARRETERO VAQUERO (1998): Fig. 2.1
Abu Sha’ar 9–10x10–11 MULVIN and SIDEBOTHAM (2004): Fig . 9.1
Birrens, England 9x15 ROBERTSON (1975): 33, 58, 100
Castiellu de Llagú 9x5 BERROCAL et al (2002): 128–129
Richborough 9x7 BUSHE-FOX (1928): Fig. 1.1
Petavonium 9x7 CARRETERO VAQUERO (1998): Fig. 2.2
Abu Sha’ar 9x9/9x10 MULVIN and SIDEBOTHAM (2004): Fig. 9.1
Viladonga 11x12 FERNÁNDEZ PINTOS (2017): 236–237
Chao de Samartín 12x17 VILLA VALDÉS (2000): Fig. 8.1
Inverarvon 13x16 DUNWELL and RALSTON (1995): 562, 564
Fig. 3. Fragmentary quadrangular boards found within the borders of the Roman Empire.

66 |COMPUTATIONAL APPROACHES FOR RECOGNISING AND RECONSTRUCTING ANCIENT GAMES: THE CASE OF LUDUS LATRUNCULORUM
Fig. 4. Map of quadrangular boards contemporary with the Roman Empire. Large squares are complete boards; small squares are fragmentary boards.
sixteen. Nine of these boards were eight-by-eight, of boards which are contemporary with, but found
the most frequent size in the sample, with multiple outside of, the Roman Empire. These boards have
examples of six-by-seven, seven-by-eight, and eight- been found at sites in Central and Northern Europe,
by-nine boards. The boards with grids larger than well beyond the limes of the Empire, but certainly
this are unique examples of their respective patterns. from areas which were in contact with the Romans.
These boards all appear to have been quadrangular,
Information about the size of boards may also be
like those seen in the Roman Empire. Indeed, the
inferred from the remains of fragmentary boards, in
remains of one of these boards, found at Vimose in
which the grid of squares is incompletely preserved
Denmark, bears the distinct pattern of the Roman
and the exact pattern cannot be known (Figure 3).
game Duodecim Scripta on the opposite side of the
Many of these are too small to make any conclusions
grid,32 demonstrating that Roman games existed
about their original size, but others clearly
outside the Empire (see discussion on globalisation
demonstrate that large boards existed in some places,
in Chapter 1). The preferred material for these
with up to seventeen squares in a row.
boards seems to have been wood, and few of them
Looking at the geographic distribution of the boards, are preserved though there are indications that, at
it becomes apparent that the common board sizes least sometimes, these boards were placed in graves
are found throughout the empire, while the larger, with counters on them, which may provide hints as
uncommon board sizes are further away from the to the original configuration, such as the one found at
core of the Empire, defined as Italy and major urban Leuna in Germany.33 These boards are contemporary
centers (Figure 4).
2.3 Contemporaneous Quadrangular Boards
32 Krüger (1982), 162, 222.
The archaeological record also presents examples 33 Schulz (1953), 29, 63–66, Pl. XXVII–XXVIII.

GAMES IN THE ANCIENT WORLD: PLACES, SPACES, ACCESSORIES, ALESSANDRO PACE, TIM PENN, ULRICH SCHÄDLER (EDS) | 67
with the Roman Empire.34 Only one example is intact: implement them in the Ludii software to calculate
a wooden board with a seventeen-by-seventeen or metrics that can indicate how the difference in board
seventeen-by-eighteen grid found in a 4th century sizes changes the experience of the game. Candidate
CE elite tomb in Poprad, Slovakia.35 A seven-by- rulesets should contain all of the rules that are
fourteen board found at Musawwarat el-Sufra, a known to have been a part of Ludus Latrunculorum
Meriotic site in Sudan,36 also points to quadrangular as described by the ancient sources, and have been
boards of larger size outside the Empire. documented geographically close to the region
where the game was played. Ideally, contemporary
The influence of Rome on the board games of
rules would be chosen over rules which are further
temperate Europe is clear, especially with respect to
distant with regard to time, but since there are no
game pieces, which were often imported from places
completely documented game rules contemporary
within the Empire.37 Indeed, the evidence for these with the Roman Empire in the Ludii database, this
games and the Roman influence that can be traced is impossible.
has been used as evidence for a connection between
Ludus Latrunculorum and the medieval game To identify which games in the Ludii database contain
Hnefatafl and other games which are probably related the rules known for Ludus Latrunculorum, a search
to it, such as Brandubh and Tawlbwrdd.38 Aside was made for the following game concepts: 3.3.2.6
from the material evidence for Roman games and Custodial Capture; 3.3.4.1.3 Orthogonal Direction
Roman-style gaming materials in Northern Europe, (movement); 2.1.1.1.1 Square Shape (board);
the connection between Ludus Latrunculorum and 2.1.1.2.1 Square Tiling (square spaces on the board);
Hnefatafl is made because it is thought that Hnefatafl and 3.2.2 Pieces Placed Outside Board (indicates
also employed the custodial capture mechanism. pieces begin off the board and must be placed on it,
This is inferred from the documentation of the game i.e. there is a placement phase in the game).41 This
search produced five games that contain all of those
of Tablut among the Sámi people in what is now
rules: two types of Gala,42 Kharebga,43 Seega,44 and
Finland during the 18th century CE by Linnaeus,
Shantarad.45 Of these games, the only ones which
which features custodial captures.39 Tablut is thought
were played in places that at one time were part of
to be a game derived from, or somehow related to, the Roman Empire are Kharebga, documented in
Hnefatafl, based on the board and the presence of El Oued, Algeria, and Seega, played in Egypt and
a “King” piece for one player and not the other.40 Sudan. Gala is played on the island of Sulawesi in
However, none of the sources which discuss Hnefatafl Indonesia, and Shantarad is played in Somalia, and
or its contemporary medieval games confirm that thus far beyond the borders of the Roman Empire.
custodial capture was part of these games. Nevertheless, Seega was documented in the 19th
The archaeological evidence of quadrangular century, and Kharebga in the 20th century, so the
game boards found within the Roman Empire chronological distance between these games and
and in adjacent regions clearly demonstrate that that of Ludus Latrunculorum is greater than a
boards ranging from six-by-seven to seventeen-by- millennium. Despite this, Kharebga and Seega are
seventeen or -eighteen were used to play games. the most likely candidates for games that could be
This very wide range of board sizes presents the similar to Ludus Latrunculorum.
question of whether the same game can be played In addition, Tablut is the only game from Europe for
on such disparate boards, or if the different board which the complete set of rules have been documented
sizes could indicate the presence of games other that also contains the custodial capture mechanism,
than Ludus Latrunculorum that were also played on and there is reason to examine it on these boards as
quadrangular boards. well. If Tablut is indeed related to Hnefatafl, as the
evidence seems to suggest, this connects the evidence
3. METHODOLOGY to a period in time much closer to the end of the
Roman Empire, as evidence for Hnefatafl appears in
3.1 Identifying Candidate Rulesets the late 8th or 9th century CE. Nevertheless, Tablut
In order to determine whether a game can be played does not have the placement phase known to exist
on the wide range of quadrangular boards found in in Ludus Latrunculorum, but has custodial capture,
and around the Roman Empire, a reasonable set of orthogonal movement, and a quadrangular board.
rules must be applied to these boards to be able to However, it also has differentiated pieces, with one
player playing with a “King” piece. There are also
34 De Voogt (2019), 92–93. different numbers of pieces per player, a central
35 Staneková (2020), 56–57.
36 Crist et al. (2016a): 140. 41 Piette et al. (2021).
37 Hall, Forsyth (2011), 1326–1330 42 Matthes (1859), 899; Matthes (1874), 71–72.
38 Hall (2019), 204–207, Hall, Forsyth (2011), 1333. 43 Bellin (1964), 53–54.
39 Linnaeus (1732), 147–148. 44 Davies (1925), 138–139; Lane (1836), 356–357.
40 Murray (1913), 445. 45 Marin (1931), 595–596.

68 |COMPUTATIONAL APPROACHES FOR RECOGNISING AND RECONSTRUCTING ANCIENT GAMES: THE CASE OF LUDUS LATRUNCULORUM
square which affects capturing mechanisms, and for all three games, the rules must be adapted for a
different winning conditions for the two players— board with an even number of spaces. For Kharebga
one attempts to move the King piece to the edge of and Seega, the rule that the central space must be left
the board, while the opponent attempts to capture empty is discarded, and play begins when only two
the King piece. Despite these differences, and in light spaces are left empty on the board; with each player
of the suggestions that Hnefatafl—and, therefore, placing one piece on the final turn of the placement
Tablut—is derived from, inspired by, or in some way phase. It should also be noted that increasing the
connected to Ludus Latrunculorum, it is worthwhile board size for these games implicitly introduces
evaluating the Tablut rules on quadrangular boards more pieces to the board: (n/2)-1 for each player,
as well. with n being the total number of spaces on the board.
For Tablut, the central space is more crucial, as it
3.2 Applying Rulesets to Boards is required for the “King” piece’s starting position
To measure behaviour metrics on the different which, in turn, imposes symmetry on the starting
board sizes of quadrangular board, and thus evaluate position of the pieces. The central space also imposes
whether they all might have been used for the same capturing and movement restrictions. Because of the
set of rules, the rulesets for Kharebga, Seega, and many changes that would be required to adapt Tablut
Tablut were applied to boards with the dimensions rules to a board with an even number of spaces, it was
of the intact quadrangular boards from within the not applied to these boards, as it would effectively be
Roman Empire, as well as the board from Poprad. a different game. Tablut rules were therefore only
These rulesets were implemented in the Ludii applied to the seventeen-by-seventeen board, and it
software so they could be played. was tested with two different versions: one with the
original number of pieces—eight plus the “King” for
Kharebga is played on either a five-by-five board
one player; sixteen for the opponent— and one with
or a seven-by-seven board. Players alternate turns the number of pieces increased in proportion and
placing two pieces on the board, leaving the central
configuration to the larger board size—sixteen plus
space empty. When all the pieces have been placed,
the “King” for one player, forty for the other.
the players alternate turns moving their pieces
over any distance in an orthogonal direction—also 3.3 Artificial Intelligence Agents
known as the slide move. Captures are made when
Once these rules were implemented on the different
an opponent’s piece is between two of the player’s
boards in Ludii, it was necessary to choose the
own pieces (i.e., custodial capture). Captures are not
made during the placement phase. The first player to parameters for the AI to conduct experiments to
capture all of their opponent’s pieces wins. generate behavioural game-play metrics. Two basic
tree search approaches were considered to provide
Seega is very similar to Kharebga; it is played on a the AI engines for automated game-playing; UCT46
five-by-five, seven-by-seven, or nine-by-nine board. and Alpha-Beta47 search. Both of these tree search
The other main difference is that pieces move one approaches perform lookahead searches, in which
space at a time, rather than at a distance—known as they “look ahead” into various different future
the step move. game states that may be reached through different
Tablut is very different from the other two. There
is no placement phase; the pieces start in a fixed
position, with the “King” piece beginning on the
center spot, with its eight allied pieces surrounding
it, and the opponent’s sixteen pieces are arrayed
toward the edge of the board. Players alternate turns
moving a piece orthogonally any distance. Captures
are custodial, but if the “King” is still in the central
space, it can only be captured by surrounding it with
four pieces, and if it is next to the central space, it
must be surrounded on three sides. The central space
cannot be entered by any piece including the “King”
once it has left it. The player with the “King” wins
by moving the “King” to the edge of the board; the
opponent wins by capturing the King. It was played
on a nine-by-nine board.
Adapting the rules to the known quadrangular boards Fig. 5. Example game tree for lookahead tree search algorithms. Thick
arrows correspond to moves that tree search algorithms would end up
contemporary with the Roman Empire immediately playing after sufficient analysis.
presents some issues. Only one of the known board
sizes—seventeen-by-seventeen—provides a central 46 Kocsis, Szepesvári (2006).
space. Since the central space features in the rules 47 Knuth, Moore (1975).

GAMES IN THE ANCIENT WORLD: PLACES, SPACES, ACCESSORIES, ALESSANDRO PACE, TIM PENN, ULRICH SCHÄDLER (EDS) | 69
sequences of moves from the current game state 3.4 Behaviour Metrics
(Figure 5), but they use different strategies for By allowing these AI programs to play multiple
deciding which parts of their search spaces to explore games against each other, it is possible to measure
or prioritise. UCT was found to barely play better certain gameplay properties that may give an
than a random player, if at all, in all but the smallest indication about how the game is experienced by
of boards. This is a common issue for UCT in games humans. These are called behaviour metrics, and
where long sequences of random play—of which can reveal interesting properties about a game which
the algorithm uses many to estimate how valuable may not be apparent from just a casual observance
different game states are—are unlikely to lead to a of the rules—for example, measuring how long a
variety of outcomes. This is the case in particular for game typically takes to play, the percentage of games
the Seega and Kharebga rulesets, where random play which end in a draw, how many moves a player
on large boards is highly unlikely to ever lead to a has to decide between each turn, etc. While there
victory for either player. are over one hundred behaviour metrics currently
Alpha-Beta search was evaluated with a variety implemented in Ludii, this paper is only concerned
of different heuristic state evaluation functions, with three of them which produce interesting results:
which it can use to compare states to each other - Duration: The average number of turns needed to
without relying on such rollouts of random play. complete a game. This metric can be used to tell if
A straightforward “Material” heuristic, which a game finishes in a reasonable number of turns, or
incentivises the program to attempt capturing more will likely take thousands of turns to complete.
opposing pieces than it loses, was found to produce - Completion: The percentage of games which did
an effective player. The program also used iterative not finish before reaching the 2,500 turn limit.
deepening48 to automatically tune its search depth Similar to duration, this metric can be used to tell
for any given time constraint, and a transposition if a game finishes in a reasonable number of turns.
table49 to improve the efficiency of the algorithm.
The Alpha-Beta program can perform deeper tree - Branching Factor: The average number of different
searches if it is given more “thinking time” per move, possible moves that a player can make during their
and deeper searches usually lead to a stronger level of turn. Games with a higher branching factor mean
play. Hence, the amount of thinking time per move that each player has more options to consider on
may be tuned to run different experiments between their turn, often indicating a more complex game.
different players with different levels of playing 3.5 Experiments
strength.
To compute all the metrics of each proposed
One risk with AI-based players is that they will ruleset, one hundred playouts are run between two
never get “bored”; if they fail to find a line of play Alpha-Beta agents, one using even search depths
that improves their position or leads to a win, but do and the other using odd search depth alternately
find a line of play that simply stalls the game, they playing player one and player two after each game.
are likely to prefer indefinitely stalling the game over Ten seconds are allocated to each agent to make a
a more risky, aggressive move. This is arguably not decision at each move. Each playout is limited by
representative of how humans would have played 1,250 moves per player; if that limit is exceeded the
a game. In an attempt to address this, experiments game is considered to be incomplete.
were also run with variants of the Alpha-Beta
Every process was run on a single CPU core @2.2
program that were either restricted to solely even
GHz. 20,480MB of memory was allocated per
search depths, or solely odd search depths. When
process, of which 16,384 MB was made available to
a program with a material heuristic searches only
the Java Virtual Machine (JVM). Each process was
to odd search depths, it may typically be expected
also limited to four days of computation.
to play more aggressively, because it will focus on
evaluating states in which it was allowed to make the Concerning the largest board sizes (seventeen-
final move—without considering a final reply by the by-seventeen, seventeen-by-eighteen for Seega
opponent. Conversely, a program that searches only and Kharebga rulesets), most of the playouts are
to even search depths may be expected to have a more extremely lengthy to run due to the high number of
defensive playstyle. This allows for experiments to be pieces and playable sites (~ 7.4 hours per playout).
run in which it is expected that there will be at least For this reason and due to the limit of four days, the
some variety in playstyles between the two players, decision was taken to run only ten playouts for these
where the presence of at least one player with a more rulesets.
aggressive playstyle can help to reduce the likelihood As a separate process to track the evolution of the
of games that are stalled indefinitely. number of pieces owned by each player after each
48 Korf (1985). move, one single playout for each ruleset was run and
49 Greenblatt, Eastlake, Crocker (1967). the total number of pieces placed on the board and

70 |COMPUTATIONAL APPROACHES FOR RECOGNISING AND RECONSTRUCTING ANCIENT GAMES: THE CASE OF LUDUS LATRUNCULORUM
the number of pieces owned by each player placed on greater than eleven-by-twelve in size at nearly 2,500
the board at each state were stored. turns—the timeout limit. Kharebga rulesets also
increased in duration with respect to board size, but
4. RESULTS the increase was more gradual. It should be noted that
the rate of increase appears to be more drastic up to
4.1 Completion
the eleven-by-sixteen board size, with the seventeen-
The Kharebga, Seega, and both Tablut rulesets by-seventeen and seventeen-by-eighteen boards
demonstrate different trends with respect to board having lower duration than expected in comparison
size. As board size increases, the percentage of to the trend on boards smaller than this. This could
playouts which played to completion before reaching be attributed to sampling error with a sample of only
the turn limit decreased for both Kharebga and ten playouts. The Tablut rulesets were shorter than
Seega rulesets (Figure 6). Completion for Seega any of the others, even though they were played
rulesets plummeted with increasing board size, with on the second-largest board, lasting 26.04 turns for
only the six-by-six, six-by-seven, and six-by-eight the unmodified Tablut rules and 73.17 turns for the
boards completing over 75% of the time. Boards ruleset with added pieces.
larger than eleven-by-twelve completed less than
13% of the time. The seventeen-by-seventeen and These results show that increasing board size has a
seventeen-by-eighteen Seega rulesets never played greater effect on the duration of games played with
to completion. In comparison, for the Kharebga Seega rules than for Kharebga rulesets. Nevertheless,
rulesets, only the seventeen-by-eighteen board though the increase in duration of Kharebga may
completed less than 75% of the time, with a gradual be more gradual, increased board size does increase
decrease with increasing board size. Both Tablut the duration of games to over one thousand turns.
rulesets—only implemented on the seventeen-by- Meanwhile, Tablut rulesets, even though they are
seventeen board— completed 100% of the time. played on a large board, are played in a fraction of
the amount of time as the Seega or Kharebga rulesets.
4.2 Duration
4.3 Branching Factor
The number of turns for Kharebga, Seega, and Tablut
rulesets showed results that mirror those for the Branching factor also increases for the Kharebga and
completion metric (Figure 7). The length of the game Seega rulesets (Figure 8). The Kharebga rulesets all
increased steeply for Seega rulesets, with boards have greater branching factor than Seega rulesets

Fig. 6. Percent completion of playouts for different board sizes.

GAMES IN THE ANCIENT WORLD: PLACES, SPACES, ACCESSORIES, ALESSANDRO PACE, TIM PENN, ULRICH SCHÄDLER (EDS) | 71
Fig. 7. Duration of playouts for different board sizes.

Fig. 8. Branching factor for different board sizes.

72 |COMPUTATIONAL APPROACHES FOR RECOGNISING AND RECONSTRUCTING ANCIENT GAMES: THE CASE OF LUDUS LATRUNCULORUM
Fig. 9. Number of pieces per player at each turn for different board sizes.

Fig. 10. Percent of pieces on the board for each player at each turn for different board sizes.

GAMES IN THE ANCIENT WORLD: PLACES, SPACES, ACCESSORIES, ALESSANDRO PACE, TIM PENN, ULRICH SCHÄDLER (EDS) | 73
on the same board, and the difference increases archaeological evidence, may indicate whether all of
with increasing board size. This is logical, since the these boards were likely to have been used for Ludus
sliding move of Kharebga gives more movement Latrunculorum, a similar game, or a completely
options for the players than the step move of Seega. different one.
Furthermore, the values for the Tablut rulesets show
The clear results of the playouts were that both games
that these rules also have large branching factors,
increased in duration as board size was increased.
though they have fewer pieces than the Kharebga or
This is not surprising, because increasing the size
Seega rulesets, but increasing the number of pieces
of the board also increases the number of pieces
(i.e., the difference between the Tablut and Tablut+
on the board, all of which need to be captured to
rulesests) also increases the branching factor.
win. However, the pieces per turn data show that
4.4 Pieces per Turn increasing the board size also makes it more difficult
Examining the number of pieces on the board at every for the AI agents to make captures, particularly in
turn gives a sense of the gameplay for the Kharebga Seega rulesets. This likely happens for a couple of
and Seega rulesets, and how that changes with reasons. For Seega rulesets, which employ the step
increasing board size (Figure 9). The plots corroborate move, the AI may have difficulty detecting a move
the duration metric results—that Kharebga rulesets that brings it closer to an opposing piece in order to
play more quickly than the Seega rulesets. For all make a capture if they are distant from one another
of the board sizes, and for both rulesets, the game on the board because of the time limit imposed
tends to follow a similar pattern. After the initial on the tree search. For Kharebga rulesets, the slide
placement phase, when the number of pieces on move allows the player more movement options, and
the board increases linearly, there is a period where therefore it is easier for an AI agent to avoid capture
captures can be made in fairly rapid succession. After for a long time.
this phase, there is a long tail where one player has For these reasons, it is difficult to exactly translate the
more pieces than the other, but captures happen number of turns played by the AI agents into real-life
with decreased frequency, sometimes with hundreds turns of a game played by humans. Nevertheless, the
of turns between captures. This tail lengthens more trend of increased number of turns and board size
quickly with increasing board size in Seega rulesets, is consistent across both rulesets, and constraints
mirroring the more rapid increase in duration seen which compel the agents to to play in a more
in this ruleset. human-like fashion do not eliminate the problem.
Examining the percentage of pieces owned by each Despite this, the problem is not only with the way
player at each turn throughout the game provides AI agents play, but is a problem with having a large
more detail about gameplay (Figure 10). These results amount of empty space with few pieces remaining
show more clearly that after the initial phase of rapid on the board, requiring the coordinated movement
capturing, one player gains an insurmountable lead of two pieces to make a capture. This is alluded to
over their opponent. In the larger boards, hundreds in previous work refuting the hypothesis that the so-
of turns can be played where one player has twice or called “Doctor’s Game” from Stanway, England was a
thrice the number of pieces of their opponent. In the Ludus Latrunculorum board because there were too
one case where one player does not gain an advantage few pieces included with the board.50 Furthermore,
over the other (eleven-by-sixteen Kharebga), the the difficulty in making the final captures has been
game still continues for hundreds of turns without observed in classroom settings,51 indicating that
captures. this is a real-world problem with the game and not
merely a limitation of the AI agents.
5. DISCUSSION The fact that the Tablut rulesets play much more
quickly on a large board than the Kharebga and
5.1 Playing on Large Boards
Seega rulesets shows that a large board itself is not
The playout data indicate some of the gameplay the problem, nor is it the number of choices that
consequences of increasing board size for particular the player has to make, since the branching factor
sets of rules. Based on what we know about Ludus for the Tablut and Tablut+ rulesets also have high
Latrunculorum, the rulesets for Kharebga and Seega values for this metric. It may be, though, that having
contain all of the rules that are known about the a large number of decisions to make, over hundreds
Roman game, and also are played in places where the of turns without captures, would be exhausting and
Roman Empire once existed. Evaluating the playout indicates a game that would likely not be played. For
results of these rulesets on the different board sizes these reasons, Seega is a less convincing candidate
found within the Roman Empire, as well as the ruleset for the larger boards. Since we know that
board found at Poprad, can help us to recognize
which rulesets may be closer to the original rules of
50 Schädler (2007), 368–369.
Ludus Latrunculorum, and, when coupled with the
51 Marco Tibaldini, personal communication.

74 |COMPUTATIONAL APPROACHES FOR RECOGNISING AND RECONSTRUCTING ANCIENT GAMES: THE CASE OF LUDUS LATRUNCULORUM
Seega was played on nine-by-nine boards, it cannot depicted on an eighteen-by-eighteen board.54 The
be ruled out as a possibility for the smaller boards— playout results for the Tablut rules are relevant here,
but the fact that it was played on boards ranging since it is likely that Hnefatafl and Alea Evangelii had
from five-by-five to nine-by-nine but not larger may similar rules. When applied to larger boards, Tablut
be indicative of the fact that larger boards make the rules play much faster than the Seega or Kharebga
game interminable with these rules. rules, which may be attributable to the fact that
winning the game is focused on the capture or escape
Furthermore, the slide move employed in Kharebga
of the “King” piece. It is expected that Hnefatafl and
is more in line with Ovid’s description of the pieces
Alea Evangelii would play similarly.
moving in a straight line than a step move. It is also
important to note a passage in the Historia Augusta, The game at Stanway may have been a nine-by-
which provides a second-hand account of Proculus thirteen board. Though Roman in date, it is thought to
being declared imperator after winning ten games of have been a Celtic board game.55 The aforementioned
Ludus Latrunculorum in a row,52 indicating the game boards from Vimose include one which has eighteen
must have been of sufficient brevity to play so many squares per row and another with fourteen or more
in succession. For all of these reasons, it appears to per row.
be more likely that Kharebga was a more suitable
Other boards, for which the names of the games are
ruleset because the slide move allows for a game that
unknown, have been found at pre-Roman sites in the
can more frequently be played to a conclusion, and
Mediterranean basin. At Tell Zakariya in the Levant,
which can be played faster on the greatest number of
a complete twelve-by-twelve board was found in the
boards. Neither Seega nor Kharebga appear to play
upper layers of the site, post-Iron Age but pre-Roman
in a reasonable fashion on larger boards. The Tablut
in date, as well as a fragment of a stone gaming table
and Tablut+ playout data indicate that other rules
that had at least ten rows of squares.56 Another, at
can be more amenable to large boards.
nearby Maresha, has at least ten spaces per row, and is
5.2 Games with Large Boards Hellenistic in date.57 Other large pre-Roman boards
It is useful to look at games that were played on large have been found in Greece at Rhamnous, where a
boards to examine whether people played games nine-by-nine and an eight-by-ten board were found
similar to Kharebga and Seega on large boards. The inscribed on a block dating to the 3rd century BCE,58
sample of games documented as part of the Digital as well as an eleven-by-eleven Hellenistic board from
Ludeme Project allows for such comparison. In the Pella.59
sample of 1,007 games at the time of writing, 17 war These boards show that other games were played on
games53 have boards with more than 100 playing larger boards before and after the Roman Empire.
sites. These can largely be divided into four separate The Hellenistic boards above could be for the game
categories: Draughts games, multiplayer games, Polis, which is mentioned in Greek sources but not
enlarged versions of existing games with added described in great detail. It is thought to be similar
pieces with new movement properties, and Konane. in some ways to Ludus Latrunculorum, but the
Each of these categories features something about candidate boards are larger than the commonly-
their rules that either requires a large board or speeds found Ludus Latrunculorum boards and, perhaps
up the game. Draughts games only use half of the indicatively, tend to have an odd number of spaces
sites. Multiplayer games have three or more players, while Ludus Latrunculorum strictly has an even
requiring more space than a game with two players. number. Less is known about the rules for Polis, so
Enlarged games, which are largely versions of Chess there is likely something missing from these rules to
and Shogi, add pieces which both necessitate a make a playable game on a larger board.
larger board for the starting position and introduce
more powerful movement for these pieces. Finally, 5.3 Archaeological Context of Boards
the number of moves in Konane is capped by the The archaeological context of the boards provides
number of spaces on the board because each move further evidence for the interpretation of
must involve a capture. quadrangular grids. Looking at the geographical
distribution of board size, it is clear that the most
In addition, there is archaeological and historical
common board sizes, six-by-seven, seven-by-eight,
evidence for other large games, for which the rules
have been lost. Hnefatafl, and games which may
be related to it such as Alea Evangelii, fall into this 54 Duggan (2021); Schulte (2017) for recent surveys of the
group. Hnefatafl could apparently be played on a evidence for these games.
thirteen-by-thirteen board, while Alea Evangelii was 55 Schädler (2007), 368–369.
56 Bliss, Macallister (1902), 144.
52 Flavius Vopiscus Syracusanus, Four Tyrants, 12.1–3. 57 Stern (2019), 127–128; Stern this volume.
53 Defined as games in which capturing the opponent’s pieces is the 58 Fachard (2021).
primary criterion for determining a win. 59 Ignatiadou (2019), 142, 152.

GAMES IN THE ANCIENT WORLD: PLACES, SPACES, ACCESSORIES, ALESSANDRO PACE, TIM PENN, ULRICH SCHÄDLER (EDS) | 75
and eight-by-eight, are widely spread throughout the function as social lubricants which can be used to
empire. These are also found in major cities of the acquaint people with one another and to form the
empire, such as Athens, Rome itself, and the cities basis for other social interactions.64
of western Asia Minor. These games exist as graffiti
in urban contexts, such as at the Basilica Iulia and 6. CONCLUSIONS AND FUTURE WORK
the Parthenon, but others were also found in military Computational methods can add another line
contexts, particularly in Britain, Spain, and Egypt
of evidence to traditional archaeological and
(e.g., Chesters, Monte de Santa Tegra, Abu Sha’ar).
philological methods to help identify board
The widespread use of these particular grid patterns,
games and test plausible rulesets. For Ludus
and the fact that the seven-by-eight and eight-by-
Latrunculorum, playout analysis shows that a game
eight boards are the only ones found intact in Italy,
follows the pattern that would be expected in the in which pieces move with a slide movement such as
Roman Empire—that the game would appear both Kharebga fits the description of the Roman authors
in the central core of the Empire as well as in the and is reasonably playable on the boards known
places the Romans occupied. from the archaeological record. Furthermore, larger
boards seem to be less amenable to these rules,
The fact that the larger boards are not found in Italy, and traditional games with similar rules on large
but typically in forts on the fringes of the empire, boards have not been recorded in human history.
could be explained in two ways. The simplest This leads to one of two conclusions: that there are
argument is that troops stationed at forts have a lot of key rules missing from the written record of Ludus
idle time, and therefore the presence of larger boards Latrunculorum that would allow these rules to work
at these places might explain an attempt to take up on large boards, or that there are other games with
this time. The playout data indicates, though, that
unknown rules that were played on large boards. The
the rules known for Ludus Latrunculorum played
archaeological record supports the presence of pre-
on these boards are probably not amenable even
and non-Roman games on large boards, so the latter
for people who are trying to waste a considerable
argument is favoured.
amount of time.
Perhaps a more compelling explanation is that these Future work can improve the utility of computational
larger game boards are not for Ludus Latrunculorum, methods in the study of games. Developing improved
but other games about which we know nothing AI agents by introducing heuristics and features to
concerning the rules. Pre-Roman quadrangular the agents can help them to play more effectively.
boards indicate that Ludus Latrunculorum was not The development of game distance metrics can
the first such game to exist in the Mediterranean help to quantitatively analyze the ludemic similarity
basin. Indeed, Polis, which was played in the Greek of games to one another, allowing for better
world, seems to have been played on larger boards, identification of candidate rulesets. In addition,
and it is possible that the board from Samos is this a social network approach can add the cultural
game, rather than Ludus Latrunculorum. But other dimension to measurement of game distance. At
boards, such as one from the Late Period Sacred press, this is being developed by the Digital Ludeme
Animal Necropolis at Saqqara in Egypt,60 as well as Project to potentially automate the identification of
several in northwest Spain61 cannot be tied to any candidate games for analyzing and reconstructing
known games. Coupled with the conclusion that the traditional games of the past. These innovations can
Stanway game was a Celtic board-game of unknown be used not only to bring new insight into antiquity,
type, and that other larger boards found at Vimose, but also to make the past interactive and to preserve
Leuna, and Poprad were also probably indigenous this intangible cultural heritage by providing playable
rather than Roman games,62 it is plausible that these games that more closely replicate the way ancient
larger boards were not Ludus Latrunculorum, but peoples played board games.
other games which existed in a suite of games played
on quadrangular boards stretching from Britain and ACKNOWLEDGEMENTS
Scandinavia to the Meroitic Kingdom in Sudan.
We wish to thank Véronique Dasen for the invitation
Auxiliary troops were drawn from the indigenous
to contribute to the volume, as well as the ERC Locus
populations of the Empire and beyond,63 so the
Ludi Project and Marco Tibaldini for pointing us to
presence of indigenous games at Roman military
more evidence for Ludus Latrunculorum and Polis.
sites is not surprising, particularly since games
This research was funded by the European Research
60 Martin (1981), 30, Pl. 39. Council (ERC Consolidator Grant #771292).
61 Berrocal-Ranger et al. (2003), 103–104; Llanos Ortiz de
Landaluze (2002), 191–193.
62 Schädler (2007), 369–373.
63 Haynes (2013), 95–142. 64 Malaby (2003), 59–74; Crist et al. (2016b).

76 |COMPUTATIONAL APPROACHES FOR RECOGNISING AND RECONSTRUCTING ANCIENT GAMES: THE CASE OF LUDUS LATRUNCULORUM
BIBLIOGRAPHY Crist, Walter, Dunn-Vaturi, Anne-Elizabeth, De Voogt,
Austin, Roland, “Roman Board Games I”, Greece and Alex, “Facilitating Interaction: Board Games as Social
Rome, 4 (1934), 24–34. Lubricants in the Ancient Near East”, Oxford Journal of
Archaeology, 35 (2016), 179–196.
Avelaira, Tomás, “Aquae Querquennae (Porto Quintela,
Ourense, España): Un campamento Romano en el NW de Culin, Stewart, Games of the North American Indians.
Hispania”, Ephemeris Napocensis, 25 (2015), 43–80. Lincoln (1907).

Baddesley, St. Clair, “A Roman Draught-Board”, Da Ponte, Sálete, “Jogos romanos de Conimbriga”,
Transactions of the Bristol and Gloucestershire Conimbriga: Revista de Arqueologia, 25 (1986), 131–141.
Archaeological Society, 45 (1923), 285. Daniaux, Thomas, “Plateaux et jetons de jeu d’Aventicum
Becq de Fouquières, Louis. Les jeux des anciens, Paris (Suisse)”, in Véronique Dasen (ed.), Ludique. Jouer dans
(1869). l’antiquité, catalogue de l’exposition, Lugdunum, musée
et théâtres romains, 20 juin-1er décembre 2019, Gent
Bell, Robert, Board and Table Games from Many (2019), 88–89.
Civilizations, New York (1979).
Davies, Robert, “Some Arab Games and Puzzles”, Sudan
Bell, Robert, “Notes on Pavement Games of Greece and Notes and Records, 8 (1925), 137–152.
Rome”, in Irving Finkel (ed.), Ancient Board Games in
De Voogt, Alex, “Traces of Appropriation: Roman Board
Perspective, London (2007), 98–99.
Games in Egypt and Sudan”, in Véronique Dasen, Ulrich
Bellin, Paul, “L’enfant saharien à travers ses jeux”, Journal Schädler (eds), Dossier Jouer dans l’Antiquité. Identité et
des Africanistes, 33 (1964), 47–104. multiculturalité, Archimède, 6 (2019), 89–99.
Berrocal-Ranger, Luis, Martínez Seco, Paz, Ruiz De Voogt, Alex, Nilsson, Maria, Ward, John, “The
Treviño, Carmen. Castiellu de Llagú (Latores, Oviedo): Role of Graffiti Game Boards in the Understanding of an
Un castro Astur en los orígenes de Oviedo, Bibliotheca Archaeological Site: The Gebel el-Silsila Quarries”, Journal
Archaeologica Hispana 13, Madrid (2002). of Egyptian Archaeology, 106 (2020), 123–132.
Bliss, Frederick, Macallister, Robert. Excavations in Duggan, Eddie, “A Game on the Edge: An Attempt to
Palestine in the Years 1898–1900, London (1902). Unravel the Gordian Knot of Tafl Games”, Board Game
Studies, 15, 99–132.
Brezee, David, Bearsden: A Roman Fort on the Antonine
Wall, Edinburgh (2016). Dunwell, Andrew, Ralston, Ian, “Excavations at
Inverarvon on the Antonine Wall, 1991”, Proceedings of
Browne, Cameron, et al., “Foundations of Digital
the Society of Antiquaries of Scotland, 125 (1995), 521–576.
Archaeoludology. Report on Dagstuhl Research Meeting
19153”, ArXhiv 1905.13516v1 (2019). Ersoy, Akın, Erdin, Özer, “Antik Dönem Smyrna
Agorası’nda Bulunan Roma Dönemi Ludus Latrunculorum
Brun, Jean-Pierre, “Le dépotoir”, in Hélène Cuvigny (ed), ve Mankala Oyun Tablaları Üzerine İnceleme”, in Emre
Didymoi: Une garnison romaine dans le désert Orientale Okan, Cenker Atila (eds), Prof. Dr. Ömer Özyiğit’e
d’Égypte, Institut Français d’Archéologie Orientale 64/1, Armağan, İstanbul (2015), 141–156.
Cairo (2011), 115–155.
Fachard, Sylvain, “Games in the Garrison Forts of
Bushe-Fox, Joscelyn, Second Report on the Excavation Attica”, paper given at The Archaeology of Play and Games
of the Roman Fort at Richborough, Kent, Reports of the Workshop, ERC Locus Ludi (2021).
Research Committee of the Society of Antiquaries of
London 7, Oxford (1928). Fernández Pintos, Julio, “Tableros de juego rupestres
en el SO. de Galicia”, Unpublished manuscript (2017).
Busuladžić, Adnan, “Roman Gaming Boards and
Pieces: Unpublished Astragals, Talus and Calculi from the Greenblatt, Richard D., Eastlake III, Donald E.,
Antiquities Collection of the National Museum of Bosnia Crocker, Stephen D., “The Greenblatt chess program”,
and Herzegovina”, Glasnik Zemaljskog Muzeja Bosne i in Proceedings of the November 14-16, 1967, Fall Joint
Hercegovine u Sarajevu: Arheologija, 54 (2017), 185–208. Computer Conference, London (1967), 801–810.

Carretero Vaquero, Santiago, “El Ludus Gregory, Tony, Gurney, David, Excavations at Thornham,
Latrunculorum, un juego de estrategia practicado por los Warham, Wighton and Calstor, Norfolk, Early Anglian
equites del Ala II Flavia”, Boletín del Seminario de Estudios Archaeology 175, Norwich (1986).
de Arte y Arqueología, 64 (1998), 117–140. Gudea, Nicolae, “Ceramica”, in Eugen Chirilă (ed.),
Castrul roman de la Buciumi, Cluj (1972).
Courts, Summer, Penn, Timothy, “A Corpus of Gaming
Boards from Roman Britain”, Lucerna, 57 (2019), 4–12. Gudea, Nicolae, Pop, Ioan, Castrul roman de la Risnoc
Cumidava, Brasov (1971).
Crist, Walter, De Voogt, Alex, Dunn-Vaturi, Anne-
Elizabeth, Ancient Egyptians at Play: Board Games across Hall, Mark, “Whose Game Is It Anyway? Board and Dice
Borders, London (2016a). Games as an Example of Cultural Transfer and Hybridity”,

GAMES IN THE ANCIENT WORLD: PLACES, SPACES, ACCESSORIES, ALESSANDRO PACE, TIM PENN, ULRICH SCHÄDLER (EDS) | 77
in Véronique Dasen, Ulrich Schädler (eds), Dossier Jouer Martin, Geoffrey, The Sacred Animal Necropolis at North
dans l’Antiquité. Identité et multiculturalité, Archimède, 6 Saqqara: The Southern Dependencies of the Main Temple
(2019), 199–212. Complex, London (1981).
Hall, Mark, Forsyth, Katherine. “Roman Rules? The Matelly, Marie-Angès, “Les petits objets”, in Hélène
Introduction of Board Games to Britain and Ireland”, Cuvigny (ed), La route de Myos Hormos: L’armée romaine
Antiquity 85 (2011), 1325–1338. dans le désert Oriental d’Égypte, Institut Français
Haynes, Ian, Blood of the Provinces: The Roman Auxilia d’Archéologie Orientale 48/II, Cairo (2003), 589–617.
and the Making of Provincial Society from Augustus to the Matthes, Benjamin, Makassaarsch-Hollandsch
Severans, Oxford (2013). Woordenboek, Amsterdam (1859).
Holbrook, Neil, Bidwell, Paul, Roman Finds from Matthes, Benjamin, Boegineesch-Hollandsch
Exeter, Exeter Archaeological Reports 4, Exeter (1991). Woordenboek, Amsterdam (1874).
Ignatiadou, Despina, “Luxury Board Games for the May, Roland. Jouer dans l’Antiquité, Marseille (1991).
Northern Greek Elite”, in Véronique Dasen, Ulrich
Michaelis, Adolf, “Terrakottagruppe aus Athen”,
Schädler (eds), Dossier Jouer dans l’Antiquité. Identité et
multiculturalité, Archimède, 6 (2019), 144–159. Archäologischer Anzeiger, 13 (1863), 37–43.

Jankovic, Marko, “Rimske igre na table u Singidunumu i Mosca, Fabio, Puppo, Paola, “Riflessioni su dadi e giochi
okolini”, Godišnjak grada Beograda, 50-51 (2009), 55–68. da tavolo nel mondo Romano”, Histria Antiqua, 21 (2012),
271–280.
Jeremic, Gordana, Saldum: Roman and Early Byzantine
Fortification, Cahiers des Portes de Fer Monographies 6, Mulvin, Lynda, Sidebotham, Steven, “Roman Game
Belgrade (2009). Boards from Abu Sha’ar (Red Sea Coast, Egypt)”, Antiquity,
78 (2004), 602–617.
Karakitsou, Elena, “Games in the Parthenon”, The
Acropolis Restoration News, 9 (2009), 23–25. Murray, Harold, A History of Chess, London (1913).

Knuth, Donald E., Moore, Ronald W., “An Analysis of Murray, Harold, A History of Board-Games other than
Alpha-Beta Pruning”, Artificial Intelligence, 6 (1975), 293– Chess, Oxford (1951).
326. Pace, Alesssandro, “Miles ludens: il gioco e i soldati
Kocsis, Levente, Szepesvári, Csaba, “Bandit Based nella Britannia romana”, in Claudia Lambrugo, Fabrizio
Monte-Carlo Planning”, in Johannes Fürnkranz, Tobias Slavazzi, Anna Maria Fedeli (eds), I materiali della
Scheffer, Myra Spiliopoulou (eds), Machine Learning: Collezione Archeologica “Giulio Sambon” di Milano, 1. Tra
ECML 2006, Vol. 4212 of Lecture Notes in Computer alea e agòn: giochi di abilità e azzardo, Materia e Arte 1,
Science, Berlin (2006), 282–293. Firenze (2015), 43-49
Korf, Richard E., “Depth-first iterative-deepening: An Pace, Alesssandro, “Playing with Batavians. Game as
optimal admissible tree search”, Artificial Intelligence, 27 educational tool for a romano more vivere”, Archimède, 7
(1985), 97–109. (2020), 317-326.
Krüger, Thomas, Das Brett- und Würfelspiel der Pace, Alesssandro, “Play at being Romans. Ludus e
Spätlatènezeit und römischen Kaiserzeit im freien omologazione culturale nei contesti militari del mondo
Germanien, Hildesheim (1982). romano”, in Raimon Graells i Fabregat, Alessandro Pace,
Lane, Edward, An Account of the Manners and Customs of Miguel F. Pérez Blasco (eds), Warrios@Play, Proceedings
the Modern Egyptians, London (1836). of the International Congress held at the Museum of
History and Archaeology of Elche, 28th May 2021, Alacant
Linnaeus, Carolus, Iter Lapponicum, Stockholm (1732). (2022), 155-167.
Llanos Ortiz de Landaluze, Armando, “Tableros de Parlett, David, The Oxford History of Board Games,
juego en el patrimonio arqueológico de Álava”, Estudios de Oxford (1999).
Arqueología Alavesa, 19 (2002), 191–196.
Parlett, David, “What’s a Ludeme?” Game Puzzle Design
Malaby, Thomas, Gambling Life: Dealing in Contingency
2 (2016), 83–86.
in a Greek City, Urbana (2003).
Penn, Tim, Courts, Summer. “Playing Games at
Manderscheid, Hubertus, Carboni, Francesca,
Vindolanda”, https://www.vindolanda.com/blog/playing-
Bruno, Matthias, “Tabulae lusoriae nel mondo romano:
games-at-vindolanda (n.d.).
Il tavoliere dei muratori di Villa Adriana, tabulae dalle
Terme di Traiano a Roma e da Leptis Magna”, Archeologia Pérez Losada, Fermin, “Una pieza epigrafica excepcional
Classica, 62 (2011), 513–535. procedente de Noville (Mugardos, A Coruña, Galicia”, II
Congreso Peninsular de História Antiga, Actas, Coimbra
Marin, G., “Somali Games”, Journal of the Royal
(1993), 1049–1066.
Anthropological Institute of Great Britain and Ireland, 61
(1931): 499–511. Petrie, Flinders, Objects of Daily Use, London (1927).

78 |COMPUTATIONAL APPROACHES FOR RECOGNISING AND RECONSTRUCTING ANCIENT GAMES: THE CASE OF LUDUS LATRUNCULORUM
Philip, Brian, Excavation of the Roman Forts of the Classis
Britannica Dover 1970–1977. Kent Monograph Series 3,
Dover (1981).
Piette, Éric, Stephenson, Matthew, Soemers, Dennis J.
N. J., Browne, Cameron. “General Board Game Concepts”,
2021 IEEE Conference on Games (CoG 2021), (2021), 1-8.
doi: 10.1109/CoG52621.2021.9618990.
Robertson, Anne. Birrens (Blatobulgium). Edinburgh
(1975).
Rodrigo Requena, Esther, Romaní Sala, Núria. “Una
tabula lusoria hallada en el castellum republicano de Puig
Castellar de Biosca en Lleida, Catalunya (180–120 a. C.)”,
Archivo Español de Arqueología, 94 (2021), 1–12.
Rose, Pamela, Qasr Ibrim: The Hinterland Survey.
Excavation Memoirs 62, London, (1996).
Schädler, Ulrich, “Latrunculi—ein verlorenes
strategisches Brettspiel der Römer”, Homo Ludens: Der
spielende Mensch, 4 (1994), 47–67.
Schädler, Ulrich, “XII Scripta, Alea, Tabula—New
Evidence for the Roman History of ‘Backgammon’”, in
Alex de Voogt (ed), New Approaches to Board Games
Research: Asian Origins and Future Perspectives, Leiden
(1995), 73–98.
Schädler, Ulrich, “Latrunculi: A Forgotten Roman
Game of Strategy Reconstructed”, Abstract Games 7
(2001), 10–11.
Schädler, Ulrich, “The Doctor’s Game—New Light on
the History of Ancient Board Games,” in Philip Crummy
et al. (eds), Stanway: An Elite Burial Site at Camulodunum,
Britannia Monograph Series 24, London (2007), 359–375.
Schulte, Michael, “”Board Games of the Vikings—From
Hnefatafl to Chess”, Maal og Minne, 2 (2017), 1–42.
Schulz, Walther, Leuna: Ein germanischer
Bestattungsplatz der spätrömischen Kaiserzeit, Berlin
(1953).
Staneková, Zuzana, “Stolové hry v staroveku.
Archeologické doklady zo severozápadného slovenska”,
Zborník Slovenského Národného Múzea v Martine, 15
(2020), 48–67.
Stephenson, Matthew, Crist, Walter, Browne, Cameron,
“Digital Ludeme Project Database Guide,” unpublished
manuscript (2020).
Stern, Ian, “Game Boards,” in Ian Stern et al. (eds),
Excavations at Maresha Subterranean Complex 169: Final
Report Seasons 2000–2016, Annual of the Hebrew Union
College 11, Jerusalem (2019), 127–131.
Villa Valdés, Angel, “Descripción de estructuras
defensivas e trazado urbano no castro do Chao de San
Martín (Grandas de Salime, Asturias)”, Boletin do Museo
Provincial de Lugo, 9 (2000), 367–419.
Villa Valdés, Angel, “Grabados zoomorfos sobre pizarra
y otros epígrafes inéditos en castros asturianos”, Sautuola,
16–17 (2010–2012), 97–112.

GAMES IN THE ANCIENT WORLD: PLACES, SPACES, ACCESSORIES, ALESSANDRO PACE, TIM PENN, ULRICH SCHÄDLER (EDS) | 79
80 |COMPUTATIONAL APPROACHES FOR RECOGNISING AND RECONSTRUCTING ANCIENT GAMES: THE CASE OF LUDUS LATRUNCULORUM