Practical Phylogenetic Methods for Linguistic Typology

Authors

Affiliation

Erich Round

University of Queensland

Martin Schweinberger

University of Queensland

Introduction

A perennial task in typology is the characterisation of frequencies of traits of interest among the world’s languages. The scientific interest of such questions typically lies not merely in the contingent facts of today’s particular languages and language families — rather, the goal is to characterise the nature of human language in general, using today’s empirical data as evidence. One of the key challenges is that languages are historically related to each other. This tutorial is a practical introduction to phylogenetic comparative methods, which meet this challenge in a principled way.

Prerequisite Tutorials

Before working through this tutorial, we recommend familiarity with:

Getting Started with R — R objects, basic syntax, RStudio orientation
Loading and Saving Data — reading files into R

No prior knowledge of phylogenetics or evolutionary biology is required. All necessary concepts are introduced as they arise.

Learning Objectives

By the end of this tutorial you will be able to:

Explain why genealogical relatedness poses a challenge for linguistic typology
Create, manipulate, and plot linguistic phylogenetic trees in R using the Newick format and the ape package
Combine glottolog family trees into composite supertrees using glottoTrees
Modify trees by adding, removing, cloning, and moving tips and nodes
Assign realistic branch lengths using exponential and ultrametric scaling
Calculate genealogically-sensitive proportions and averages using the ACL and BM methods in phyloWeights
Apply the full workflow to a real typological dataset

Citation

Round, Erich & Martin Schweinberger. 2026. Practical Phylogenetic Methods for Linguistic Typology. Brisbane: The Language Technology and Data Analysis Laboratory (LADAL). url: https://ladal.edu.au/tutorials/ladal_phylogentic_showcase/ladal_phylogentic_showcase.html (Version 2026.05.01).

The tutorial is based largely on the Supplementary Materials section S1 of Macklin-Cordes and Round (2021a), and draws substantially on the functionality of glottoTrees (Round 2021a) and phyloWeights (Round 2021b).

Background Reading

This tutorial covers the practical, technical side of phylogenetic methods. For the scientific motivation — why genealogy must be accounted for in typological research and how phylogenetic comparative methods provide the most principled response developed so far — we recommend:

The LADAL seminar by Macklin-Cordes & Round (Macklin-Cordes and Round 2021b), available at https://youtu.be/yJeQ6AGDLK0
The companion journal article Macklin-Cordes and Round (2021a)

The text of this tutorial is taken largely from the Supplementary Materials section S1 of Macklin-Cordes and Round (2021a).

An intuitive introduction to the challenge of genealogy

To get an intuitive sense of the challenge posed by genealogical relatedness, consider the small language family below. It contains four languages, which are either SOV or SVO. What proportion of this family is SOV? If you simply count languages, the answer is 75%. But this may feel wrong: one half of the family is counting three times as much as the other half, just because it happens to contain more languages. The figure of 75% is strongly influenced by contingencies of history.

This document provides a practical guide to calculating genealogically-sensitive proportions and averages — methods that answer the question: when characterising the frequencies of traits among the world’s languages, how can we take genealogy into account? We cover two methods: the ACL method (Altschul, Carroll, and Lipman 1989) and the BM method (Stone and Sidow 2007).

The tutorial is divided into four main sections. Section Trees in R introduces how trees are represented and manipulated in R. Section Genealogically-sensitive averages and proportions covers the calculation itself. Section Using and adapting trees from glottolog explains how to prepare phylogenetic trees from freely available resources at glottolog.org. Section Worked example: Yin (2020) provides a complete worked example from a typological investigation of sonority sequencing.

Installation and Setup

Packages Used in This Tutorial

This tutorial requires two GitHub-only packages — glottoTrees and phyloWeights — which are not currently available on CRAN. Both are actively maintained by Erich Round and must be installed via remotes.

Installing packages

Code

# Install remotes if needed
install.packages("remotes")

# Install from GitHub
remotes::install_github("erichround/glottoTrees",
                        dependencies = TRUE)
remotes::install_github("erichround/phyloWeights",
                        dependencies = TRUE)

# Install CRAN packages
install.packages(c("ape", "dplyr", "flextable", "checkdown"))

Installation Notes

If the GitHub installation fails, try adding INSTALL_opts = c("--no-multiarch") to the remotes::install_github() calls. On Windows this resolves the most common installation errors:

remotes::install_github("erichround/glottoTrees",
                        dependencies = TRUE,
                        INSTALL_opts = c("--no-multiarch"))

The packages require an internet connection on first install. Once installed, all glottolog data is bundled locally and no internet connection is needed to run the tutorial.

Loading packages

Code

library(ape)
library(dplyr)
library(flextable)
library(glottoTrees)
library(phyloWeights)
library(checkdown)

Trees in R

Section Overview

What you will learn: How trees are created from Newick strings; the phylo and multiPhylo object classes; how to access and plot tree components; and why the left-to-right arrangement of tips in a tree carries no linguistic meaning

This section discusses how trees are created, manipulated and plotted in R. Terminology we will use: tips at the ends of trees (usually individual languages or lects); the branches of a tree; interior nodes where branches join; and the root, the deepest node. R represents trees as phylo objects containing tip labels, node labels, branch lengths, and topology.

Creating trees from Newick strings

The simplest way to construct a tree in R begins with a Newick string — a bracketing notation where tips are grouped by parentheses, separated by commas, and terminated with a semicolon:

Code

my_newick <- "(((A,B),C),D);"
my_tree   <- ape::read.tree(text = my_newick)

We can plot the tree using R’s plot() function (which by default plots trees horizontally, following biological convention):

Code

plot(my_tree)

The glottoTrees package provides plot_glotto(), which plots trees in the downward-running format more common in linguistics:

Code

plot_glotto(my_tree)

Adding branch lengths

Branch lengths are written in Newick format with a preceding colon directly after a tip label or closing bracket:

Code

my_newick2 <- "(((A:4,B:4):1,C:5):3,D:8);"
my_tree2   <- ape::read.tree(text = my_newick2)

Code

plot_glotto(my_tree2)

Adding node labels

Internal nodes can be labelled by placing the label directly after a closing parenthesis. In linguistics, node labels may reflect a taxonomic subgroup or a proto-language:

Code

my_newick3 <- "(((A:4,B:4)proto-AB:1,C:5)proto-ABC:3,D:8)proto-ABCD;"
my_tree3   <- ape::read.tree(text = my_newick3)
plot_glotto(my_tree3)

The phylo object

Trees in R are objects of class phylo:

Code

class(my_tree3)

[1] "phylo"

A phylo object stores the tree topology, branch lengths, and tip and node labels, accessible via $:

Code

my_tree3$edge.length   # branch lengths

[1] 3 1 4 4 5 8

Code

my_tree3$tip.label     # tip labels

[1] "A" "B" "C" "D"

Code

my_tree3$node.label    # node labels

[1] "proto-ABCD" "proto-ABC"  "proto-AB"

The multiPhylo class

Multiple trees are stored together in a multiPhylo object. Individual trees within it are accessed using double square brackets [[i]], not $:

Code

newick_a      <- "(((A:4,B:4):1,C:5):3,D:8);"
newick_b      <- "((A:2,B:2):1,(C:1,D:1,E:1):2);"
tree_a        <- ape::read.tree(text = newick_a)
tree_b        <- ape::read.tree(text = newick_b)
my_multiPhylo <- c(tree_a, tree_b)
class(my_multiPhylo)

[1] "multiPhylo"

Code

# Plot the second tree in the multiPhylo object
plot_glotto(my_multiPhylo[[2]])

The irrelevance of left-to-right arrangement

In a phylogenetic tree, there is no meaningful difference between (A,B) and (B,A): in both, A and B are sisters under a shared parent. Likewise, (A,(B,C)), (A,(C,B)), ((B,C),A) and ((C,B),A) are all equivalent. The left-to-right ordering of tips reflects only the plotting convention, never a linguistic claim.

Genealogically-Sensitive Averages and Proportions

Section Overview

What you will learn: The two components needed for genealogically-sensitive calculations — a tree and a dataframe; how to use phylo_average() from phyloWeights; how to interpret ACL and BM method results and weights; and how the answer changes as the hypothesised tree changes

We now turn to calculating genealogically-sensitive proportions and averages. Consider these six trees, representing six different genealogical hypotheses for languages A, B, C, D — each shown with word order and consonant phoneme count:

Required components

Calculating genealogically-sensitive averages requires two inputs:

A phylo or multiPhylo object containing one or more trees
A dataframe with a column named tip (matching the tip labels in the trees) and at least one column of numerical data

Our six trees are placed in a multiPhylo object:

Code

multiPhylo_ABCD <-
  c(ape::read.tree(text = "(((A:0.2,B:0.2,C:0.2):1.8,D:2):0.3);"),
    ape::read.tree(text = "(((A:1,B:1,C:1):1,D:2):0.3);"),
    ape::read.tree(text = "(((A:1.8,B:1.8,C:1.8):0.2,D:2):0.3);"),
    ape::read.tree(text = "((((A:1,B:1):0.8,C:1.8):0.2,D:2):0.3);"),
    ape::read.tree(text = "(((A:1.8,(B:1,C:1):0.8):0.2,D:2):0.3);"),
    ape::read.tree(text = "(((A:1,B:1):1,(C:1.8,D:1.8):0.2):0.3);"))

Preparing the dataframe

The dataframe must contain a tip column whose values match the tree’s tip labels, plus numerical columns to be averaged. Use is_X or has_X names for proportion variables and n_X for count variables:

Code

data_ABCD <- data.frame(
  tip          = c("A", "B", "C", "D"),
  is_SOV       = c(1, 1, 1, 0),
  n_consonants = c(18, 20, 22, 40),
  stringsAsFactors = FALSE
)

To calculate a proportion (e.g. proportion of SOV languages), fill the column with 1 (language has the property) and 0 (language does not). To calculate an average (e.g. mean consonant count), fill the column with the values.

In practice, the dataframe will often be read from a CSV file:

Code

data_ABCD <- read.csv("my_data_file.csv", stringsAsFactors = FALSE)

Running phylo_average()

Code

results_ABCD <- phyloWeights::phylo_average(
  phy  = multiPhylo_ABCD,
  data = data_ABCD
)

Computation time

phylo_average() may take up to several minutes for large trees or many input trees. It is normal for it to run silently for a few minutes before returning.

Interpreting the results

The results object contains several components, accessed with $.

The supplied data (as a reference):

Code

results_ABCD$data

  tip is_SOV n_consonants
1   A      1           18
2   B      1           20
3   C      1           22
4   D      0           40

ACL genealogically-sensitive averages — one row per tree, one column per numerical variable. Notice how the answer changes as the genealogical hypothesis changes:

Code

results_ABCD$ACL_averages

   tree is_SOV n_consonants
1 tree1 0.5172        29.66
2 tree2 0.6000        28.00
3 tree3 0.7143        25.71
4 tree4 0.6769        26.64
5 tree5 0.6769        26.29
6 tree6 0.7115        25.92

BM genealogically-sensitive averages:

Code

results_ABCD$BM_averages

   tree is_SOV n_consonants
1 tree1 0.5569        28.86
2 tree2 0.6952        26.10
3 tree3 0.7468        25.06
4 tree4 0.7257        25.56
5 tree5 0.7257        25.41
6 tree6 0.7289        25.51

Both methods work by assigning weights to languages, reflecting how much each language contributes to the final result. The weights are in $ACL_weights and $BM_weights. Comparing these weights to the tree structure illuminates how the methods respond to different topologies:

Code

results_ABCD$ACL_weights

  tip  tree1 tree2  tree3  tree4  tree5  tree6
1   A 0.1724   0.2 0.2381 0.1965 0.2838 0.2115
2   B 0.1724   0.2 0.2381 0.1965 0.1965 0.2115
3   C 0.1724   0.2 0.2381 0.2838 0.1965 0.2885
4   D 0.4828   0.4 0.2857 0.3231 0.3231 0.2885

Code

results_ABCD$BM_weights

  tip  tree1  tree2  tree3  tree4  tree5  tree6
1   A 0.1856 0.2317 0.2489 0.2289 0.2680 0.2289
2   B 0.1856 0.2317 0.2489 0.2289 0.2289 0.2289
3   C 0.1856 0.2317 0.2489 0.2680 0.2289 0.2711
4   D 0.4431 0.3048 0.2532 0.2743 0.2743 0.2711

Results can be saved to CSV:

Code

write.csv(results_ABCD$ACL_averages, file = "my_ACL_averages.csv")

Using and Adapting Trees from Glottolog

Section Overview

What you will learn: How to access glottolog’s genealogical metadata and family trees in R; how to combine multiple family trees into a supertree; how to add, remove, keep, clone and move tips and nodes; and how to assign realistic branch lengths

Glottolog’s genealogical data

Glottolog is a major online resource for quantitative typology, providing metadata about the world’s language varieties and their genealogical relationships. The glottoTrees package bundles this data directly, so no internet connection is needed once the package is installed. The package currently contains glottolog versions 4.0 through 5.0; by default all functions use the most recent bundled version.

Language metadata is accessed with get_glottolog_languages():

Code

language_metadata <- glottoTrees::get_glottolog_languages()
head(language_metadata, n = 10)

   glottocode isocodes       name name_in_tree position tree
1    3adt1234          3Ad-Tekles   3Ad-Tekles      tip  293
2    aala1237              Aalawa       Aalawa      tip  357
3    aant1238          Aantantara   Aantantara      tip  202
4    aari1238     <NA>       <NA>   Aari-Gayil     node   53
5    aari1239      aiw       Aari         Aari      tip   53
6    aari1240      aay     Aariya       Aariya     <NA>   NA
7    aasa1238      aas      Aasax        Aasax      tip  293
8    aasd1234            Aasdring     Aasdring      tip  251
9    aata1238           Aatasaara    Aatasaara      tip  202
10   abaa1238              Rngaba       Rngaba      tip   85
               tree_name
1           Afro-Asiatic
2           Austronesian
3  NuclearTransNewGuinea
4            SouthOmotic
5            SouthOmotic
6                   <NA>
7           Afro-Asiatic
8          Indo-European
9  NuclearTransNewGuinea
10          Sino-Tibetan

To save the full ~26,000-row metadata to a file for browsing in Excel:

Code

write.csv(language_metadata, "language_metadata.csv")

Listed here are glottolog’s languages, dialects, subgroups and families, each identified by a name, an ISO-639-3 code (if available), and a glottocode (four letters + four digits). The table also describes each entity’s representation in the glottolog trees — its position as tip or node, and the tree’s number and name.

To access an older glottolog version, supply the version number:

Code

language_metadata_v4.3 <- glottoTrees::get_glottolog_languages(
  glottolog_version = "4.3"
)
head(language_metadata_v4.3, n = 10)

   glottocode isocodes       name name_in_tree position tree
1    3adt1234          3Ad-Tekles   3Ad-Tekles      tip  186
2    aala1237              Aalawa       Aalawa      tip  205
3    aant1238          Aantantara   Aantantara      tip  145
4    aari1238     <NA>       <NA>   Aari-Gayil     node   85
5    aari1239      aiw       Aari         Aari      tip   85
6    aari1240      aay     Aariya       Aariya     <NA>   NA
7    aasa1238      aas      Aasax        Aasax      tip  186
8    aasd1234            Aasdring     Aasdring      tip  179
9    aata1238           Aatasaara    Aatasaara      tip  145
10   abaa1238              Rngaba       Rngaba      tip  329
               tree_name
1           Afro-Asiatic
2           Austronesian
3  NuclearTransNewGuinea
4            SouthOmotic
5            SouthOmotic
6                   <NA>
7           Afro-Asiatic
8          Indo-European
9  NuclearTransNewGuinea
10          Sino-Tibetan

Briefer metadata about glottolog’s language families is available via get_glottolog_families():

Code

family_metadata <- glottoTrees::get_glottolog_families()
head(family_metadata, n = 10)

   tree     tree_name n_tips n_nodes main_macroarea
1     1  Abkhaz-Adyge     14       7        Eurasia
2     2        Surmic     21      14         Africa
3     3        Tamaic     13       6         Africa
4     4       Yareban      9       4      Papunesia
5     5         Bogia      2       1      Papunesia
6     6       Teberan     14       3      Papunesia
7     7       Saliban      3       2  South America
8     8 Hibito-Cholon      2       1  South America
9     9       Kiwaian     18       6      Papunesia
10   10      Pahoturi      5       3      Papunesia

The current default version divides the world’s languages into over 400 families, with an internal node for each. Geographically, glottolog assigns each language variety to one of six macroareas: Africa, Australia, Eurasia, Papunesia, South America, or North America.

Accessing glottolog’s family trees

Glottolog’s family trees are stored in a multiPhylo object. The current default version object is glottolog_trees_v5.0. Trees can be retrieved by number or by family name.

By name, using get_glottolog_trees():

Code

tree_GA <- glottoTrees::get_glottolog_trees("GreatAndamanese")
plot(tree_GA, x.lim = c(-0.3, 14))

To find the number of any family’s tree, use which_tree():

Code

glottoTrees::which_tree("GreatAndamanese")

GreatAndamanese 
            190

Code

glottoTrees::which_tree(c("Turkic", "Tupian", "Tuu"))

Turkic Tupian    Tuu 
   112    152    303

In glottolog’s trees, tip and node labels are long strings including the language name, glottocode, and ISO code. The function abridge_labels() shortens them to just the glottocode:

Code

tree_GA_abr <- glottoTrees::abridge_labels(tree_GA)
plot_glotto(tree_GA_abr)

Equal branch lengths in glottolog

In glottolog’s trees, all branches are of equal length. This is an unrealistic assumption — see Section How to add branch lengths for how to assign more realistic lengths.

How to combine trees

Comparison of languages across families requires a commitment to a genealogical hypothesis. glottoTrees provides tools for making such hypotheses explicit by combining multiple glottolog trees into a single composite tree.

As a small example, here we combine five glottolog families to represent the hypothesised Arnhem group in northern Australia (Green 2003):

Code

arnhem_family_names <- c("Gunwinyguan", "Mangarrayi-Maran",
                         "Maningrida", "Kungarakany", "Gaagudju")
multiPhylo_arnhem   <- glottoTrees::get_glottolog_trees(arnhem_family_names)

assemble_rake() joins the trees in a multiPhylo object into a single tree with a rake structure at the root (all families joining directly to a shared root with no additional subgrouping):

Code

tree_arnhem <- glottoTrees::assemble_rake(multiPhylo_arnhem)

Code

tree_arnhem_abr <- glottoTrees::abridge_labels(tree_arnhem)
plot_glotto(tree_arnhem_abr, nodelabels = FALSE)

More structure can be added by using assemble_rake() iteratively. Here we hypothesise that Gunwinyguan, Mangarrayi-Maran and Maningrida form a subgroup before joining the isolates:

Code

multiPhylo_A  <- glottoTrees::get_glottolog_trees(
  c("Gunwinyguan", "Mangarrayi-Maran", "Maningrida")
)
tree_A        <- glottoTrees::assemble_rake(multiPhylo_A)
multiPhylo_arnhem2 <- c(tree_A,
                        glottoTrees::get_glottolog_trees(c("Kungarakany", "Gaagudju")))
tree_arnhem2       <- glottoTrees::assemble_rake(multiPhylo_arnhem2)
tree_arnhem2_abr   <- glottoTrees::abridge_labels(tree_arnhem2)
plot_glotto(tree_arnhem2_abr, nodelabels = FALSE)

For studies covering many families, assemble_supertree() joins all families in the default version of glottolog into a single supertree, organised first by macroarea. The tree is large, so we do not plot it here:

Code

my_supertree <- glottoTrees::assemble_supertree()

The macro-level groupings can be customised. To join all families in a flat 420-pronged rake (no macroarea grouping):

Code

my_supertree <- glottoTrees::assemble_supertree(macro_groups = NULL)

To merge North and South America into a single group:

Code

my_list <- list("Africa", "Australia", "Eurasia", "Papunesia",
                c("South America", "North America"))
my_supertree <- glottoTrees::assemble_supertree(macro_groups = my_list)

To include only families from Africa and Eurasia (as separate nodes):

Code

my_supertree <- glottoTrees::assemble_supertree(
  macro_groups = list("Africa", "Eurasia")
)

How to modify trees

Typological studies often require trees whose tip sets differ from glottolog’s. The glottoTrees package supplies a complete toolkit for modifying tree topology. The following examples all use the Great Andamanese tree with abridged labels:

Code

tree_GA     <- glottoTrees::get_glottolog_trees("GreatAndamanese")
tree_GA_abr <- glottoTrees::abridge_labels(tree_GA)
plot_glotto(tree_GA_abr)

How to remove tips

remove_tip() removes specified tips; if all tips below a node are removed, that node is also removed automatically:

Code

tree_GAa <- glottoTrees::remove_tip(tree_GA_abr,
                                    label = c("akab1249", "akak1251", "apuc1241"))
plot_glotto(tree_GAa)

Code

# Removing both tips under a node also removes the node
tree_GAb <- glottoTrees::remove_tip(tree_GA_abr,
                                    label = c("akab1249", "akar1243"))
plot_glotto(tree_GAb)

keep_tip() specifies which tips to retain (the complement of remove_tip()):

Code

tree_GAc <- glottoTrees::keep_tip(tree_GA_abr,
             label = c("akar1243", "akak1251", "akac1240",
                       "akak1252", "apuc1241", "okoj1239"))
plot_glotto(tree_GAc)

Code

# Node boca1235 is removed because neither of its tips are kept
tree_GAd <- glottoTrees::keep_tip(tree_GA_abr,
             label = c("akar1243", "akak1251", "akak1252",
                       "apuc1241", "okoj1239"))
plot_glotto(tree_GAd)

How to remove tips and convert nodes to tips

keep_as_tip() accepts both tip and node labels. Tips are kept; nodes are converted into tips (all structure below them is removed). This is useful when the typologist has data at the language level but glottolog represents languages as internal nodes above dialect tips:

Code

tree_GAe <- glottoTrees::keep_as_tip(
  tree_GA_abr,
  label = c("akar1243", "akak1251", "akak1252",
            "apuc1241", "okoj1239", "boca1235")
)
plot_glotto(tree_GAe)

A typical workflow: read a CSV with a tip column and pass it directly to keep_as_tip():

Code

my_dataframe <- read.csv("my_data_file.csv", stringsAsFactors = FALSE)
my_new_tree  <- glottoTrees::keep_as_tip(my_old_tree,
                                         label = my_dataframe$tip)

To convert specific nodes to tips without removing anything else:

Code

tree_GAf <- glottoTrees::convert_to_tip(tree_GA_abr,
                                        label = c("okol1242", "sout2683"))
plot_glotto(tree_GAf)

How to remove internal nodes

After removing tips, some remaining tips may sit under a node that dominates only them (a non-branching node). collapse_node() removes such nodes, reducing tip depth:

Code

tree_GAg <- glottoTrees::collapse_node(tree_GAc,
                                       label = c("boca1235", "okol1242"))
plot_glotto(tree_GAg)

nonbranching_nodes() identifies which nodes have only one child:

Code

glottoTrees::nonbranching_nodes(tree_GAc)

[1] "okol1242" "boca1235" "jeru1239" "sout2683"

Code

glottoTrees::nonbranching_nodes(tree_GAg)

[1] "jeru1239" "sout2683"

collapse_node() can also alter subgrouping by converting a nested structure ((A,B),C) into a flat one (A,B,C):

Code

tree_GAh <- glottoTrees::collapse_node(tree_GA_abr, label = "okol1242")
plot_glotto(tree_GAh)

How to add tips

add_tip() adds a new tip below a specified parent node:

Code

tree_GAi <- glottoTrees::add_tip(tree_GA_abr,
                                 label       = "xxxx1234",
                                 parent_label = "sout2683")
plot_glotto(tree_GAi)

How to clone tips

clone_tip() duplicates one or more tips — useful when glottolog provides one glottocode for multiple lects in the typologist’s sample:

Code

tree_GAj <- glottoTrees::clone_tip(tree_GA_abr,
                                   label = c("akar1243", "akak1252"))
plot_glotto(tree_GAj)

Setting subgroup = TRUE places clones in a new subgroup node:

Code

tree_GAk <- glottoTrees::clone_tip(tree_GA_abr,
                                   label    = c("akar1243", "akak1252"),
                                   subgroup = TRUE)
plot_glotto(tree_GAk)

Multiple clones can be created with the n argument:

Code

tree_GAl <- glottoTrees::clone_tip(tree_GA_abr,
                                   label    = "akab1248",
                                   n        = 3,
                                   subgroup = TRUE)
plot_glotto(tree_GAl)

After cloning, tip labels may no longer be unique. apply_duplicate_suffixes() adds a hyphen-number suffix to duplicate labels:

Code

tree_GAm <- glottoTrees::apply_duplicate_suffixes(tree_GAj)
plot_glotto(tree_GAm)

How to move a tip

move_tip() moves a tip to a new position beneath a specified parent node:

Code

tree_GAn <- glottoTrees::move_tip(tree_GA_abr,
                                  label        = "apuc1241",
                                  parent_label  = "jeru1239")
plot_glotto(tree_GAn)

How to move a node and its descendants

move_node() moves an internal node — along with all structure below it — to a new parent:

Code

tree_GAo <- glottoTrees::move_node(tree_GA_abr,
                                   label        = "jeru1239",
                                   parent_label  = "okol1242")
plot_glotto(tree_GAo)

Summary: toolkit for curating tree topology

The functions remove_tip(), keep_tip(), keep_as_tip(), convert_to_tip(), collapse_node(), add_tip(), clone_tip(), move_tip() and move_node() provide a general-purpose toolkit for modifying a tree’s set of tips and its subgrouping structure to match any typological sample.

How to add branch lengths

Branch lengths convey information and most phylogenetic comparative methods — including genealogically-sensitive averages — are sensitive to relative branch lengths. Glottolog’s trees contain equal branch lengths, which is almost certainly unrealistic.

A good approximation to the most likely distribution of branch lengths under a variety of assumptions is exponential (Venditti, Meade, and Pagel 2010): very long branches are rare; very short ones are frequent. rescale_branches_exp() implements this by setting the deepest branches to length 1/2, the next layer to 1/4, and so on:

Code

tree_GAp <- glottoTrees::rescale_branches_exp(tree_GA_abr)
plot_glotto(tree_GAp)

Code

tree_EA     <- glottoTrees::get_glottolog_trees("Eskimo-Aleut")
tree_EA_abr <- glottoTrees::abridge_labels(tree_EA)
tree_EAa    <- glottoTrees::rescale_branches_exp(tree_EA_abr)
plot_glotto(tree_EAa, nodelabels = FALSE)

ultrametricize() stretches terminal branches so that all tips are equidistant from the root:

Code

tree_EAb <- glottoTrees::ultrametricize(tree_EAa)
plot_glotto(tree_EAb, nodelabels = FALSE)

rescale_deepest_branches() adjusts only the deepest layer of branches — useful when multiple family trees have been joined and you wish to control the implied distance between first-order branches. Here we triple the deepest branch length and then ultrametricise:

Code

tree_arnhem_a <- glottoTrees::rescale_branches_exp(tree_arnhem_abr)
tree_arnhem_b <- glottoTrees::rescale_deepest_branches(tree_arnhem_a, 1.5)
tree_arnhem_c <- glottoTrees::ultrametricize(tree_arnhem_b)
plot_glotto(tree_arnhem_c, nodelabels = FALSE)

Exporting trees

Trees can be saved in Newick format using write.tree() from ape, for use with other software such as FigTree:

Code

ape::write.tree(tree_arnhem_c, "my_arnhem_tree.tree")

relabel_with_names() replaces glottocodes with glottolog’s full language names before plotting or exporting:

Code

tree_arnhem_c_namelabels <- glottoTrees::relabel_with_names(tree_arnhem_c)
plot_glotto(tree_arnhem_c_namelabels, nodelabels = FALSE)

Worked Example: Yin (2020)

Section Overview

What you will learn: How to apply the full phylogenetic workflow — supertree construction, tip curation, branch length assignment, and genealogically-sensitive averaging — to a real published typological dataset. This section reproduces the phylogenetic analysis of Yin (2020).

Yin (2020) examined violations of the sonority sequencing principle in 496 languages and calculated the genealogically-sensitive proportions of languages in which various violations occurred. The language sample consisted of 496 languages from the CLICS2 database (Anderson et al. 2018) and the AusPhon-Lexicon database (Round 2017).

Computation Time

The supertree construction in this section (assemble_supertree()) may take several minutes to run, depending on your hardware. This is normal. We recommend running the code chunks in this section interactively rather than knitting the full document.

The data

Yin’s raw data consists of language names, glottocodes, and binary indicators of whether each language has consonant clusters with sonority reversals in word-initial onsets or word-final codas (coded 1 for yes, 0 for no). This dataset is provided with phyloWeights as the dataframe yin_2020_data, with columns name, tip, has_onset_violation and has_coda_violation. The first ten rows:

Code

head(yin_2020_data, n = 10)

               name      tip has_onset_violation has_coda_violation
1            Abkhaz abkh1244                   1                  1
2              Abui abui1241                   0                  0
3           Achagua acha1250                   0                  1
4             Adang adan1251                   0                  1
5     Adnyamathanha adny1235                   0                  0
6            Adyghe adyg1241                   1                  1
7      Hokkaidoainu ainu1240                   0                  0
8             Alawa alaw1244                   1                  0
9  Standardalbanian alba1267                   1                  1
10            Aleut aleu1260                   1                  1

Preparing the supertree

The tree for Yin’s study was constructed from a glottolog supertree. The language sample covered relatively few families in the Americas (so North and South America were merged) and only one African language (Arabic), so Africa and Eurasia were also merged. We use glottolog v4.2, as in the original study, to ensure exact reproducibility:

Code

yin_macro <- list(
  c("South America", "North America"),
  c("Africa", "Eurasia"),
  "Papunesia",
  "Australia"
)
supertree     <- glottoTrees::assemble_supertree(
  macro_groups      = yin_macro,
  glottolog_version = "4.2"
)
supertree_abr <- glottoTrees::abridge_labels(supertree)

Five tips were cloned, in cases where Yin had data for two varieties corresponding to just one tip in the glottolog supertree:

Code

supertree_a <- glottoTrees::clone_tip(
  supertree_abr,
  subgroup = TRUE,
  label    = c("ayab1239", "basu1242", "biri1256", "ikar1243", "peri1265")
)
supertree_b <- glottoTrees::apply_duplicate_suffixes(supertree_a)

Eight tips were added, in cases where for sister lects (A, B), glottolog placed A as a node above B. In such cases, a new tip A was placed below the existing glottolog node A:

Code

supertree_c <- supertree_b
nodes_to_add_as_tips <- c("alor1249", "gami1243", "guri1247", "mand1415",
                           "sins1241", "wang1291", "warl1254", "yand1253")
for (node_i in nodes_to_add_as_tips) {
  supertree_c <- glottoTrees::add_tip(supertree_c,
                                      label        = node_i,
                                      parent_label  = node_i)
}

From this supertree, only the 496 languages in Yin’s dataset were kept. The internal node mada1298 and all non-branching internal nodes were then collapsed:

Code

supertree_d <- glottoTrees::keep_as_tip(supertree_c,
                                        label = yin_2020_data$tip)
supertree_e <- glottoTrees::collapse_node(supertree_d,
                                          label = "mada1298")
supertree_f <- glottoTrees::collapse_node(
  supertree_e,
  label = glottoTrees::nonbranching_nodes(supertree_e)
)

Finally, branch lengths were assigned. Branches were first assigned exponential lengths. Then, to diminish the importance of macro groups, the branches above them were shortened to 1/40. The effect is that the implied distance between families in different macro groups is only marginally greater than between families within a single macro group:

Code

supertree_g  <- glottoTrees::rescale_branches_exp(supertree_f)
yin_2020_tree <- glottoTrees::rescale_deepest_branches(supertree_g, 1/40)

The resulting tree, plotted as a fan with full language names:

Code

full_names          <- yin_2020_data$name[
  match(yin_2020_tree$tip.label, yin_2020_data$tip)
]
name_tree           <- yin_2020_tree
name_tree$tip.label <- full_names
plot(ape::ladderize(name_tree, right = FALSE),
     type         = "fan",
     cex          = 0.3,
     label.offset = 0.002,
     edge.width   = 0.5)

Preparing the dataframe

The dataframe yin_2020_data already has a tip column and two numerical columns (has_onset_violation and has_coda_violation), so it meets the requirements of phylo_average() directly. Non-numeric columns (here, name) are ignored automatically.

Calculating genealogically-sensitive proportions

Code

yin_2020_results <- phyloWeights::phylo_average(
  phy  = yin_2020_tree,
  data = yin_2020_data
)

Since we are using a single tree, the results are compact. The genealogically-sensitive proportions according to the ACL and BM methods:

Code

yin_2020_results$ACL_averages

   tree has_onset_violation has_coda_violation
1 tree1              0.3711             0.4098

Code

yin_2020_results$BM_averages

   tree has_onset_violation has_coda_violation
1 tree1              0.4015             0.3848

The first ten rows of phylogenetic weights from each method:

Code

head(yin_2020_results$ACL_weights, n = 10)

               name      tip      tree1
1            Abkhaz abkh1244 0.00698583
2              Abui abui1241 0.00126372
3           Achagua acha1250 0.00024567
4             Adang adan1251 0.00021978
5     Adnyamathanha adny1235 0.00006377
6            Adyghe adyg1241 0.00698583
7      Hokkaidoainu ainu1240 0.01746458
8             Alawa alaw1244 0.00196527
9  Standardalbanian alba1267 0.00236997
10            Aleut aleu1260 0.00852669

Code

head(yin_2020_results$BM_weights,  n = 10)

               name      tip     tree1
1            Abkhaz abkh1244 0.0049475
2              Abui abui1241 0.0021424
3           Achagua acha1250 0.0009110
4             Adang adan1251 0.0009392
5     Adnyamathanha adny1235 0.0008929
6            Adyghe adyg1241 0.0049475
7      Hokkaidoainu ainu1240 0.0064305
8             Alawa alaw1244 0.0027424
9  Standardalbanian alba1267 0.0040995
10            Aleut aleu1260 0.0055528

As a point of comparison, the raw (genealogically unweighted) proportions are the simple column means:

Code

mean(yin_2020_data$has_onset_violation)

[1] 0.3649

Code

mean(yin_2020_data$has_coda_violation)

[1] 0.3145

The difference between the raw proportions and the phylogenetically-weighted proportions reflects the correction for genealogical relatedness. Which direction the correction goes, and how large it is, depends on the distribution of the trait across the tree.

Using These Methods in Typological Research

As we seek to analyse the empirical diversity of attested languages, there are fundamental reasons why genealogy must be part of the picture (see Macklin-Cordes and Round 2021a). And since the genealogies of human languages are still incompletely known, it is imperative to make our phylogenetic hypotheses and assumptions as explicit and as testable as possible. Through glottoTrees and phyloWeights, linguistic trees and the code used to produce them can be published together with typological studies. This enables subsequent researchers to replicate findings and — crucially — to modify the phylogenetic assumptions and thereby test further hypotheses.

Citing these packages

If you find glottoTrees and phyloWeights useful in your research, please cite them as Round (2021a) and Round (2021b) respectively. To cite the notion of genealogically-sensitive averages and proportions, cite Macklin-Cordes and Round (2021a) and/or the more specific references therein.

If you encounter a bug or anomalous behaviour, please use the GitHub Issues pages or contact Erich Round directly at e.round@uq.edu.au.

Citation & Session Info

@manual{round2026phylo,
  author       = {Round, Erich and Schweinberger, Martin},
  title        = {Practical Phylogenetic Methods for Linguistic Typology},
  note         = {tutorials/ladal_phylogentic_showcase/ladal_phylogentic_showcase.html},
  year         = {2026},
  organization = {The University of Queensland, Australia. School of Languages and Cultures},
  address      = {Brisbane},
  edition      = {2026.05.01}
}

AI Transparency Statement

This tutorial was adapted for LADAL by Martin Schweinberger with the assistance of Claude (claude.ai), a large language model created by Anthropic. The original tutorial text and all R code were authored by Erich Round and taken from the Supplementary Materials of Macklin-Cordes and Round (2021a). The adaptation involved converting the document to Quarto format, updating all div blocks to Quarto callouts, updating package installation code and glottolog version references, and adding LADAL-style section overviews. Erich Round and Martin Schweinberger take full responsibility for the accuracy of the content.

Code

sessionInfo()

R version 4.4.2 (2024-10-31 ucrt)
Platform: x86_64-w64-mingw32/x64
Running under: Windows 11 x64 (build 26200)

Matrix products: default


locale:
[1] LC_COLLATE=English_United States.utf8 
[2] LC_CTYPE=English_United States.utf8   
[3] LC_MONETARY=English_United States.utf8
[4] LC_NUMERIC=C                          
[5] LC_TIME=English_United States.utf8    

time zone: Australia/Brisbane
tzcode source: internal

attached base packages:
[1] stats     graphics  grDevices datasets  utils     methods   base     

other attached packages:
[1] phyloWeights_0.4   glottoTrees_0.1.13 flextable_0.9.11   dplyr_1.2.0       
[5] ape_5.8-1          checkdown_0.0.13  

loaded via a namespace (and not attached):
 [1] tidyselect_1.2.1        farver_2.1.2            phytools_2.5-2         
 [4] S7_0.2.1                optimParallel_1.0-2     fastmap_1.2.0          
 [7] combinat_0.0-8          fontquiver_0.2.1        digest_0.6.39          
[10] lifecycle_1.0.5         magrittr_2.0.4          compiler_4.4.2         
[13] rlang_1.1.7             tools_4.4.2             igraph_2.2.2           
[16] yaml_2.3.10             data.table_1.17.0       knitr_1.51             
[19] phangorn_2.12.1         clusterGeneration_1.3.8 askpass_1.2.1          
[22] htmlwidgets_1.6.4       mnormt_2.1.2            scatterplot3d_0.3-45   
[25] xml2_1.3.6              RColorBrewer_1.1-3      expm_1.0-0             
[28] withr_3.0.2             purrr_1.2.1             numDeriv_2016.8-1.1    
[31] grid_4.4.2              gdtools_0.5.0           ggplot2_4.0.2          
[34] scales_1.4.0            iterators_1.0.14        MASS_7.3-61            
[37] cli_3.6.5               rmarkdown_2.30          ragg_1.5.1             
[40] generics_0.1.4          rstudioapi_0.17.1       stringr_1.6.0          
[43] maps_3.4.3              parallel_4.4.2          BiocManager_1.30.27    
[46] vctrs_0.7.2             Matrix_1.7-2            jsonlite_2.0.0         
[49] fontBitstreamVera_0.1.1 patchwork_1.3.0         systemfonts_1.3.1      
[52] foreach_1.5.2           tidyr_1.3.2             glue_1.8.0             
[55] codetools_0.2-20        DEoptim_2.2-8           stringi_1.8.7          
[58] gtable_0.3.6            quadprog_1.5-8          tibble_3.3.1           
[61] pillar_1.11.1           htmltools_0.5.9         openssl_2.3.2          
[64] R6_2.6.1                textshaping_1.0.0       doParallel_1.0.17      
[67] evaluate_1.0.5          lattice_0.22-6          markdown_2.0           
[70] renv_1.1.7              fontLiberation_0.1.0    Rcpp_1.1.1             
[73] zip_2.3.2               uuid_1.2-1              fastmatch_1.1-8        
[76] coda_0.19-4.1           nlme_3.1-166            officer_0.7.3          
[79] xfun_0.56               pkgconfig_2.0.3

Back to LADAL home

References

Altschul, Stephen F., Raymond J. Carroll, and David J. Lipman. 1989. “Weights for Data Related by a Tree.” Journal of Molecular Biology 207 (4): 647–53. https://doi.org/https://doi.org/10.1016/0022-2836(89)90234-9.

Anderson, Cormac, Robert Forkel, Simon J Greenhill, Johann-Mattis List, Christoph Rzymski, and Tiago Tresoldi. 2018. “CLICS2: An Improved Database of Cross-Linguistic Colexifications Assembling Lexical Data with the Help of Cross-Linguistic Data Formats.” Linguistic Typology 22 (2): 277–306. https://doi.org/https://doi.org/10.1515/lingty-2018-0010.

Green, Rebecca. 2003. “Proto Maningrida with Proto Arnhem: Evidence from Verbal Inflectional Suffixes.” In The Non-Pama-Nyungan Languages of Northern Australia : Comparative Studies of the Continent’s Most Linguistically Complex Region, edited by Nicholas Evans, 369–421. Pacific Linguistics.

Macklin-Cordes, Jayden, and Erich R Round. 2021a. “Challenges of Sampling and How Phylogenetic Comparative Methods Help: With a Case Study of the Pama-Nyungan Laminal Contrast.”

———. 2021b. “Phylogenetic Comparative Methods: What All the Fuss Is about, and How to Use Them in Everyday Research.” Brisbane.

Round, Erich R. 2017. “The AusPhon-Lexicon Project: 2 Million Normalized Segments Across 300 Australian Languages.”

———. 2021a. glottoTrees: Phylogenetic Trees in Linguistics. https://github.com/erichround/glottoTrees.

———. 2021b. phyloWeights: Calculation of Genealogically-Sensitive Proportions and Averages. https://github.com/erichround/phyloWeights.

Stone, Eric A., and Arend Sidow. 2007. “Constructing a Meaningful Evolutionary Average at the Phylogenetic Center of Mass.” BMC Bioinformatics 8 (1): 222. https://doi.org/https://doi.org/10.1186/1471-2105-8-222.

Venditti, Chris, Andrew Meade, and Mark Pagel. 2010. “Phylogenies Reveal New Interpretation of Speciation and the Red Queen.” Nature 463 (7279): 349–52. https://doi.org/https://doi.org/10.1038/nature08630.

Yin, Ruihua. 2020. “Violations of the Sonority Sequencing Principle: How, and How Often?” https://als.asn.au/Conference/Program.

--- title: "Practical Phylogenetic Methods for Linguistic Typology" author: - name: "Erich Round" affiliation: "University of Queensland" - name: "Martin Schweinberger" affiliation: "University of Queensland" format: html: toc: true toc-depth: 4 code-fold: show code-tools: true theme: cosmo --- ```{r setup, echo=FALSE, message=FALSE, warning=FALSE} options(stringsAsFactors = FALSE) options("scipen" = 100, "digits" = 4) library(checkdown) ``` ![](/images/uq1.jpg){ width=100% } # Introduction {#intro} A perennial task in typology is the characterisation of frequencies of traits of interest among the world's languages. The scientific interest of such questions typically lies not merely in the contingent facts of today's particular languages and language families — rather, the goal is to characterise the nature of human language in general, using today's empirical data as evidence. One of the key challenges is that languages are historically related to each other. This tutorial is a practical introduction to **phylogenetic comparative methods**, which meet this challenge in a principled way. ::: {.callout-note} ## Prerequisite Tutorials Before working through this tutorial, we recommend familiarity with: - [Getting Started with R](/tutorials/intror/intror.html) — R objects, basic syntax, RStudio orientation - [Loading and Saving Data](/tutorials/load/load.html) — reading files into R No prior knowledge of phylogenetics or evolutionary biology is required. All necessary concepts are introduced as they arise. ::: ::: {.callout-note} ## Learning Objectives By the end of this tutorial you will be able to: 1. Explain why genealogical relatedness poses a challenge for linguistic typology 2. Create, manipulate, and plot linguistic phylogenetic trees in R using the Newick format and the `ape` package 3. Combine glottolog family trees into composite supertrees using `glottoTrees` 4. Modify trees by adding, removing, cloning, and moving tips and nodes 5. Assign realistic branch lengths using exponential and ultrametric scaling 6. Calculate genealogically-sensitive proportions and averages using the ACL and BM methods in `phyloWeights` 7. Apply the full workflow to a real typological dataset ::: ::: {.callout-note} ## Citation Round, Erich & Martin Schweinberger. 2026. *Practical Phylogenetic Methods for Linguistic Typology*. Brisbane: The Language Technology and Data Analysis Laboratory (LADAL). url: https://ladal.edu.au/tutorials/ladal_phylogentic_showcase/ladal_phylogentic_showcase.html (Version 2026.05.01). The tutorial is based largely on the Supplementary Materials section S1 of @macklin-cordes_challenges_nodate, and draws substantially on the functionality of `glottoTrees` [@glottoTrees] and `phyloWeights` [@phyloWeights]. ::: ::: {.callout-note} ## Background Reading This tutorial covers the practical, technical side of phylogenetic methods. For the scientific motivation — why genealogy must be accounted for in typological research and how phylogenetic comparative methods provide the most principled response developed so far — we recommend: - The LADAL seminar by Macklin-Cordes & Round [@macklin-cordes_phylogenetic_2021], available at <https://youtu.be/yJeQ6AGDLK0> - The companion journal article @macklin-cordes_challenges_nodate The text of this tutorial is taken largely from the Supplementary Materials section S1 of @macklin-cordes_challenges_nodate. ::: ## An intuitive introduction to the challenge of genealogy {-} To get an intuitive sense of the challenge posed by genealogical relatedness, consider the small language family below. It contains four languages, which are either SOV or SVO. What proportion of this family is SOV? If you simply count languages, the answer is 75%. But this may feel wrong: one half of the family is counting three times as much as the other half, just because it happens to contain more languages. The figure of 75% is strongly influenced by contingencies of history. ```{r SOV-tree, echo=FALSE, message=FALSE, warning=FALSE} library(ape) plot(read.tree(text = "(((SOV:1,SOV:1,SOV:1):1,SVO:2):0.3);"), direction = "downwards", srt = 90, adj = 0.5, label.offset = 0.1, y.lim = c(0, 3), x.lim = c(-0.5, 4.5), no.margin = TRUE, cex = 1.2) ``` This document provides a practical guide to calculating **genealogically-sensitive proportions and averages** — methods that answer the question: when characterising the frequencies of traits among the world's languages, how can we take genealogy into account? We cover two methods: the **ACL** method [@altschul_weights_1989] and the **BM** method [@stone_constructing_2007]. The tutorial is divided into four main sections. Section [Trees in R](#trees-in-r) introduces how trees are represented and manipulated in R. Section [Genealogically-sensitive averages and proportions](#averages) covers the calculation itself. Section [Using and adapting trees from glottolog](#glottolog) explains how to prepare phylogenetic trees from freely available resources at glottolog.org. Section [Worked example: Yin (2020)](#example) provides a complete worked example from a typological investigation of sonority sequencing. --- # Installation and Setup {#setup} ::: {.callout-note} ## Packages Used in This Tutorial This tutorial requires two GitHub-only packages — `glottoTrees` and `phyloWeights` — which are not currently available on CRAN. Both are actively maintained by Erich Round and must be installed via `remotes`. ::: ## Installing packages {-} ```{r install, eval=FALSE, message=FALSE, warning=FALSE} # Install remotes if needed install.packages("remotes") # Install from GitHub remotes::install_github("erichround/glottoTrees", dependencies = TRUE) remotes::install_github("erichround/phyloWeights", dependencies = TRUE) # Install CRAN packages install.packages(c("ape", "dplyr", "flextable", "checkdown")) ``` ::: {.callout-warning} ## Installation Notes If the GitHub installation fails, try adding `INSTALL_opts = c("--no-multiarch")` to the `remotes::install_github()` calls. On Windows this resolves the most common installation errors: ```r remotes::install_github("erichround/glottoTrees", dependencies = TRUE, INSTALL_opts = c("--no-multiarch")) ``` The packages require an internet connection on first install. Once installed, all glottolog data is bundled locally and no internet connection is needed to run the tutorial. ::: ## Loading packages {-} ```{r load-pkgs, message=FALSE, warning=FALSE} library(ape) library(dplyr) library(flextable) library(glottoTrees) library(phyloWeights) library(checkdown) ``` --- # Trees in R {#trees-in-r} ::: {.callout-note} ## Section Overview **What you will learn:** How trees are created from Newick strings; the `phylo` and `multiPhylo` object classes; how to access and plot tree components; and why the left-to-right arrangement of tips in a tree carries no linguistic meaning ::: This section discusses how trees are created, manipulated and plotted in R. Terminology we will use: _tips_ at the ends of trees (usually individual languages or lects); the _branches_ of a tree; _interior nodes_ where branches join; and the _root_, the deepest node. R represents trees as `phylo` objects containing tip labels, node labels, branch lengths, and topology. ## Creating trees from Newick strings {-} The simplest way to construct a tree in R begins with a **Newick string** — a bracketing notation where tips are grouped by parentheses, separated by commas, and terminated with a semicolon: ```{r newick-basic, message=FALSE, warning=FALSE} my_newick <- "(((A,B),C),D);" my_tree <- ape::read.tree(text = my_newick) ``` We can plot the tree using R's `plot()` function (which by default plots trees horizontally, following biological convention): ```{r plot-basic, out.width="70%", fig.align="center", message=FALSE, warning=FALSE} plot(my_tree) ``` The `glottoTrees` package provides `plot_glotto()`, which plots trees in the downward-running format more common in linguistics: ```{r plot-glotto, out.width="70%", fig.align="center", message=FALSE, warning=FALSE} plot_glotto(my_tree) ``` ## Adding branch lengths {-} Branch lengths are written in Newick format with a preceding colon directly after a tip label or closing bracket: ```{r newick-lengths, message=FALSE, warning=FALSE} my_newick2 <- "(((A:4,B:4):1,C:5):3,D:8);" my_tree2 <- ape::read.tree(text = my_newick2) ``` ```{r plot-lengths, out.width="70%", fig.align="center", message=FALSE, warning=FALSE} plot_glotto(my_tree2) ``` ## Adding node labels {-} Internal nodes can be labelled by placing the label directly after a closing parenthesis. In linguistics, node labels may reflect a taxonomic subgroup or a proto-language: ```{r newick-nodes, out.width="70%", fig.align="center", message=FALSE, warning=FALSE} my_newick3 <- "(((A:4,B:4)proto-AB:1,C:5)proto-ABC:3,D:8)proto-ABCD;" my_tree3 <- ape::read.tree(text = my_newick3) plot_glotto(my_tree3) ``` ## The phylo object {-} Trees in R are objects of class `phylo`: ```{r class-check, message=FALSE, warning=FALSE} class(my_tree3) ``` A `phylo` object stores the tree topology, branch lengths, and tip and node labels, accessible via `$`: ```{r phylo-components, message=FALSE, warning=FALSE} my_tree3$edge.length # branch lengths my_tree3$tip.label # tip labels my_tree3$node.label # node labels ``` ## The multiPhylo class {-} Multiple trees are stored together in a `multiPhylo` object. Individual trees within it are accessed using double square brackets `[[i]]`, not `$`: ```{r multiphylo, message=FALSE, warning=FALSE} newick_a <- "(((A:4,B:4):1,C:5):3,D:8);" newick_b <- "((A:2,B:2):1,(C:1,D:1,E:1):2);" tree_a <- ape::read.tree(text = newick_a) tree_b <- ape::read.tree(text = newick_b) my_multiPhylo <- c(tree_a, tree_b) class(my_multiPhylo) ``` ```{r plot-multiphylo, out.width="85%", fig.align="center", message=FALSE, warning=FALSE} # Plot the second tree in the multiPhylo object plot_glotto(my_multiPhylo[[2]]) ``` ::: {.callout-note} ## The irrelevance of left-to-right arrangement In a phylogenetic tree, there is no meaningful difference between (A,B) and (B,A): in both, A and B are sisters under a shared parent. Likewise, (A,(B,C)), (A,(C,B)), ((B,C),A) and ((C,B),A) are all equivalent. The left-to-right ordering of tips reflects only the plotting convention, never a linguistic claim. ::: --- # Genealogically-Sensitive Averages and Proportions {#averages} ::: {.callout-note} ## Section Overview **What you will learn:** The two components needed for genealogically-sensitive calculations — a tree and a dataframe; how to use `phylo_average()` from `phyloWeights`; how to interpret ACL and BM method results and weights; and how the answer changes as the hypothesised tree changes ::: We now turn to calculating genealogically-sensitive proportions and averages. Consider these six trees, representing six different genealogical hypotheses for languages A, B, C, D — each shown with word order and consonant phoneme count: ```{r six-trees, echo=FALSE, out.height="20%", message=FALSE, warning=FALSE} t1 <- ape::read.tree(text = "(((A:0.2,B:0.2,C:0.2):1.8,D:2):0.3);") t2 <- ape::read.tree(text = "(((A:1,B:1,C:1):1,D:2):0.3);") t3 <- ape::read.tree(text = "(((A:1.8,B:1.8,C:1.8):0.2,D:2):0.3);") t4 <- ape::read.tree(text = "((((A:1,B:1):0.8,C:1.8):0.2,D:2):0.3);") t5 <- ape::read.tree(text = "(((A:1.8,(B:1,C:1):0.8):0.2,D:2):0.3);") t6 <- ape::read.tree(text = "(((A:1,B:1):1,(C:1.8,D:1.8):0.2):0.3);") t1$tip.label <- t2$tip.label <- t3$tip.label <- t4$tip.label <- t5$tip.label <- t6$tip.label <- c("A\n\nSOV\n18", "B\n\nSOV\n20", "C\n\nSOV\n22", "D\n\nSVO\n40") par(mfrow = c(2, 3), mar = c(1, 1, 1, 1)) for (t in list(t1, t2, t3, t4, t5, t6)) { plot(t, direction = "downwards", srt = 90, adj = 0.5, label.offset = 0.2, y.lim = c(0, 3.5), x.lim = c(-0.5, 4.5), no.margin = TRUE, cex = 1.2) } par(mfrow = c(1, 1)) ``` ## Required components {-} Calculating genealogically-sensitive averages requires two inputs: 1. A `phylo` or `multiPhylo` object containing one or more trees 2. A dataframe with a column named `tip` (matching the tip labels in the trees) and at least one column of numerical data Our six trees are placed in a `multiPhylo` object: ```{r multi-abcd, message=FALSE, warning=FALSE} multiPhylo_ABCD <- c(ape::read.tree(text = "(((A:0.2,B:0.2,C:0.2):1.8,D:2):0.3);"), ape::read.tree(text = "(((A:1,B:1,C:1):1,D:2):0.3);"), ape::read.tree(text = "(((A:1.8,B:1.8,C:1.8):0.2,D:2):0.3);"), ape::read.tree(text = "((((A:1,B:1):0.8,C:1.8):0.2,D:2):0.3);"), ape::read.tree(text = "(((A:1.8,(B:1,C:1):0.8):0.2,D:2):0.3);"), ape::read.tree(text = "(((A:1,B:1):1,(C:1.8,D:1.8):0.2):0.3);")) ``` ## Preparing the dataframe {-} The dataframe must contain a `tip` column whose values match the tree's tip labels, plus numerical columns to be averaged. Use `is_X` or `has_X` names for proportion variables and `n_X` for count variables: ```{r data-abcd, message=FALSE, warning=FALSE} data_ABCD <- data.frame( tip = c("A", "B", "C", "D"), is_SOV = c(1, 1, 1, 0), n_consonants = c(18, 20, 22, 40), stringsAsFactors = FALSE ) ``` To calculate a **proportion** (e.g. proportion of SOV languages), fill the column with 1 (language has the property) and 0 (language does not). To calculate an **average** (e.g. mean consonant count), fill the column with the values. In practice, the dataframe will often be read from a CSV file: ```{r read-csv, eval=FALSE, message=FALSE, warning=FALSE} data_ABCD <- read.csv("my_data_file.csv", stringsAsFactors = FALSE) ``` ## Running phylo_average() {-} ```{r phylo-avg, message=FALSE, warning=FALSE} results_ABCD <- phyloWeights::phylo_average( phy = multiPhylo_ABCD, data = data_ABCD ) ``` ::: {.callout-note} ## Computation time `phylo_average()` may take up to several minutes for large trees or many input trees. It is normal for it to run silently for a few minutes before returning. ::: ## Interpreting the results {-} The results object contains several components, accessed with `$`. The **supplied data** (as a reference): ```{r results-data, message=FALSE, warning=FALSE} results_ABCD$data ``` **ACL genealogically-sensitive averages** — one row per tree, one column per numerical variable. Notice how the answer changes as the genealogical hypothesis changes: ```{r acl-avgs, message=FALSE, warning=FALSE} results_ABCD$ACL_averages ``` **BM genealogically-sensitive averages**: ```{r bm-avgs, message=FALSE, warning=FALSE} results_ABCD$BM_averages ``` Both methods work by assigning **weights** to languages, reflecting how much each language contributes to the final result. The weights are in `$ACL_weights` and `$BM_weights`. Comparing these weights to the tree structure illuminates how the methods respond to different topologies: ```{r weights, message=FALSE, warning=FALSE} results_ABCD$ACL_weights results_ABCD$BM_weights ``` Results can be saved to CSV: ```{r save-csv, eval=FALSE, message=FALSE, warning=FALSE} write.csv(results_ABCD$ACL_averages, file = "my_ACL_averages.csv") ``` --- # Using and Adapting Trees from Glottolog {#glottolog} ::: {.callout-note} ## Section Overview **What you will learn:** How to access glottolog's genealogical metadata and family trees in R; how to combine multiple family trees into a supertree; how to add, remove, keep, clone and move tips and nodes; and how to assign realistic branch lengths ::: ## Glottolog's genealogical data {#glottolog-data} [Glottolog](https://glottolog.org) is a major online resource for quantitative typology, providing metadata about the world's language varieties and their genealogical relationships. The `glottoTrees` package bundles this data directly, so no internet connection is needed once the package is installed. The package currently contains glottolog versions 4.0 through 5.0; by default all functions use the most recent bundled version. Language metadata is accessed with `get_glottolog_languages()`: ```{r glottolog-languages, message=FALSE, warning=FALSE} language_metadata <- glottoTrees::get_glottolog_languages() head(language_metadata, n = 10) ``` To save the full ~26,000-row metadata to a file for browsing in Excel: ```{r save-metadata, eval=FALSE, message=FALSE, warning=FALSE} write.csv(language_metadata, "language_metadata.csv") ``` Listed here are glottolog's languages, dialects, subgroups and families, each identified by a name, an ISO-639-3 code (if available), and a glottocode (four letters + four digits). The table also describes each entity's representation in the glottolog trees — its position as tip or node, and the tree's number and name. To access an older glottolog version, supply the version number: ```{r glottolog-older, message=FALSE, warning=FALSE} language_metadata_v4.3 <- glottoTrees::get_glottolog_languages( glottolog_version = "4.3" ) head(language_metadata_v4.3, n = 10) ``` Briefer metadata about glottolog's language families is available via `get_glottolog_families()`: ```{r glottolog-families, message=FALSE, warning=FALSE} family_metadata <- glottoTrees::get_glottolog_families() head(family_metadata, n = 10) ``` The current default version divides the world's languages into over 400 families, with an internal node for each. Geographically, glottolog assigns each language variety to one of six macroareas: Africa, Australia, Eurasia, Papunesia, South America, or North America. ## Accessing glottolog's family trees {-} Glottolog's family trees are stored in a `multiPhylo` object. The current default version object is `glottolog_trees_v5.0`. Trees can be retrieved by number or by family name. By name, using `get_glottolog_trees()`: ```{r ga-tree-name, out.width="60%", fig.align="center", message=FALSE, warning=FALSE} tree_GA <- glottoTrees::get_glottolog_trees("GreatAndamanese") plot(tree_GA, x.lim = c(-0.3, 14)) ``` To find the number of any family's tree, use `which_tree()`: ```{r which-tree, message=FALSE, warning=FALSE} glottoTrees::which_tree("GreatAndamanese") glottoTrees::which_tree(c("Turkic", "Tupian", "Tuu")) ``` In glottolog's trees, tip and node labels are long strings including the language name, glottocode, and ISO code. The function `abridge_labels()` shortens them to just the glottocode: ```{r abridge, out.width="70%", fig.align="center", message=FALSE, warning=FALSE} tree_GA_abr <- glottoTrees::abridge_labels(tree_GA) plot_glotto(tree_GA_abr) ``` ::: {.callout-note} ## Equal branch lengths in glottolog In glottolog's trees, all branches are of equal length. This is an unrealistic assumption — see Section [How to add branch lengths](#add-lengths) for how to assign more realistic lengths. ::: ## How to combine trees {#combining-trees} Comparison of languages across families requires a commitment to a genealogical hypothesis. `glottoTrees` provides tools for making such hypotheses explicit by combining multiple glottolog trees into a single composite tree. As a small example, here we combine five glottolog families to represent the hypothesised Arnhem group in northern Australia [@green_proto_2003]: ```{r arnhem-multiphylo, message=FALSE, warning=FALSE} arnhem_family_names <- c("Gunwinyguan", "Mangarrayi-Maran", "Maningrida", "Kungarakany", "Gaagudju") multiPhylo_arnhem <- glottoTrees::get_glottolog_trees(arnhem_family_names) ``` `assemble_rake()` joins the trees in a `multiPhylo` object into a single tree with a rake structure at the root (all families joining directly to a shared root with no additional subgrouping): ```{r arnhem-rake, message=FALSE, warning=FALSE} tree_arnhem <- glottoTrees::assemble_rake(multiPhylo_arnhem) ``` ```{r arnhem-plot, out.width="75%", fig.align="center", message=FALSE, warning=FALSE} tree_arnhem_abr <- glottoTrees::abridge_labels(tree_arnhem) plot_glotto(tree_arnhem_abr, nodelabels = FALSE) ``` More structure can be added by using `assemble_rake()` iteratively. Here we hypothesise that Gunwinyguan, Mangarrayi-Maran and Maningrida form a subgroup before joining the isolates: ```{r arnhem-structured, out.width="75%", fig.align="center", message=FALSE, warning=FALSE} multiPhylo_A <- glottoTrees::get_glottolog_trees( c("Gunwinyguan", "Mangarrayi-Maran", "Maningrida") ) tree_A <- glottoTrees::assemble_rake(multiPhylo_A) multiPhylo_arnhem2 <- c(tree_A, glottoTrees::get_glottolog_trees(c("Kungarakany", "Gaagudju"))) tree_arnhem2 <- glottoTrees::assemble_rake(multiPhylo_arnhem2) tree_arnhem2_abr <- glottoTrees::abridge_labels(tree_arnhem2) plot_glotto(tree_arnhem2_abr, nodelabels = FALSE) ``` For studies covering many families, `assemble_supertree()` joins all families in the default version of glottolog into a single supertree, organised first by macroarea. The tree is large, so we do not plot it here: ```{r supertree-default, message=FALSE, warning=FALSE} my_supertree <- glottoTrees::assemble_supertree() ``` The macro-level groupings can be customised. To join all families in a flat 420-pronged rake (no macroarea grouping): ```{r supertree-null, message=FALSE, warning=FALSE} my_supertree <- glottoTrees::assemble_supertree(macro_groups = NULL) ``` To merge North and South America into a single group: ```{r supertree-custom, message=FALSE, warning=FALSE} my_list <- list("Africa", "Australia", "Eurasia", "Papunesia", c("South America", "North America")) my_supertree <- glottoTrees::assemble_supertree(macro_groups = my_list) ``` To include only families from Africa and Eurasia (as separate nodes): ```{r supertree-afeur, message=FALSE, warning=FALSE} my_supertree <- glottoTrees::assemble_supertree( macro_groups = list("Africa", "Eurasia") ) ``` ## How to modify trees {#modify-trees} Typological studies often require trees whose tip sets differ from glottolog's. The `glottoTrees` package supplies a complete toolkit for modifying tree topology. The following examples all use the Great Andamanese tree with abridged labels: ```{r ga-for-mods, out.width="70%", fig.align="center", message=FALSE, warning=FALSE} tree_GA <- glottoTrees::get_glottolog_trees("GreatAndamanese") tree_GA_abr <- glottoTrees::abridge_labels(tree_GA) plot_glotto(tree_GA_abr) ``` ### How to remove tips {-} `remove_tip()` removes specified tips; if all tips below a node are removed, that node is also removed automatically: ```{r remove-tip, out.width="70%", fig.align="center", message=FALSE, warning=FALSE} tree_GAa <- glottoTrees::remove_tip(tree_GA_abr, label = c("akab1249", "akak1251", "apuc1241")) plot_glotto(tree_GAa) ``` ```{r remove-tip-node, out.width="70%", fig.align="center", message=FALSE, warning=FALSE} # Removing both tips under a node also removes the node tree_GAb <- glottoTrees::remove_tip(tree_GA_abr, label = c("akab1249", "akar1243")) plot_glotto(tree_GAb) ``` `keep_tip()` specifies which tips to retain (the complement of `remove_tip()`): ```{r keep-tip, out.width="70%", fig.align="center", message=FALSE, warning=FALSE} tree_GAc <- glottoTrees::keep_tip(tree_GA_abr, label = c("akar1243", "akak1251", "akac1240", "akak1252", "apuc1241", "okoj1239")) plot_glotto(tree_GAc) ``` ```{r keep-tip-node-removal, out.width="70%", fig.align="center", message=FALSE, warning=FALSE} # Node boca1235 is removed because neither of its tips are kept tree_GAd <- glottoTrees::keep_tip(tree_GA_abr, label = c("akar1243", "akak1251", "akak1252", "apuc1241", "okoj1239")) plot_glotto(tree_GAd) ``` ### How to remove tips and convert nodes to tips {-} `keep_as_tip()` accepts both tip and node labels. Tips are kept; nodes are converted into tips (all structure below them is removed). This is useful when the typologist has data at the language level but glottolog represents languages as internal nodes above dialect tips: ```{r keep-as-tip, out.width="70%", fig.align="center", message=FALSE, warning=FALSE} tree_GAe <- glottoTrees::keep_as_tip( tree_GA_abr, label = c("akar1243", "akak1251", "akak1252", "apuc1241", "okoj1239", "boca1235") ) plot_glotto(tree_GAe) ``` A typical workflow: read a CSV with a `tip` column and pass it directly to `keep_as_tip()`: ```{r keep-as-tip-csv, eval=FALSE, message=FALSE, warning=FALSE} my_dataframe <- read.csv("my_data_file.csv", stringsAsFactors = FALSE) my_new_tree <- glottoTrees::keep_as_tip(my_old_tree, label = my_dataframe$tip) ``` To convert specific nodes to tips without removing anything else: ```{r convert-to-tip, out.width="70%", fig.align="center", message=FALSE, warning=FALSE} tree_GAf <- glottoTrees::convert_to_tip(tree_GA_abr, label = c("okol1242", "sout2683")) plot_glotto(tree_GAf) ``` ### How to remove internal nodes {-} After removing tips, some remaining tips may sit under a node that dominates only them (a non-branching node). `collapse_node()` removes such nodes, reducing tip depth: ```{r collapse-node, out.width="55%", fig.align="center", message=FALSE, warning=FALSE} tree_GAg <- glottoTrees::collapse_node(tree_GAc, label = c("boca1235", "okol1242")) plot_glotto(tree_GAg) ``` `nonbranching_nodes()` identifies which nodes have only one child: ```{r nonbranching, message=FALSE, warning=FALSE} glottoTrees::nonbranching_nodes(tree_GAc) glottoTrees::nonbranching_nodes(tree_GAg) ``` `collapse_node()` can also alter subgrouping by converting a nested structure ((A,B),C) into a flat one (A,B,C): ```{r collapse-subgroup, out.width="70%", fig.align="center", message=FALSE, warning=FALSE} tree_GAh <- glottoTrees::collapse_node(tree_GA_abr, label = "okol1242") plot_glotto(tree_GAh) ``` ### How to add tips {-} `add_tip()` adds a new tip below a specified parent node: ```{r add-tip, out.width="55%", fig.align="center", message=FALSE, warning=FALSE} tree_GAi <- glottoTrees::add_tip(tree_GA_abr, label = "xxxx1234", parent_label = "sout2683") plot_glotto(tree_GAi) ``` ### How to clone tips {-} `clone_tip()` duplicates one or more tips — useful when glottolog provides one glottocode for multiple lects in the typologist's sample: ```{r clone-tip, out.width="70%", fig.align="center", message=FALSE, warning=FALSE} tree_GAj <- glottoTrees::clone_tip(tree_GA_abr, label = c("akar1243", "akak1252")) plot_glotto(tree_GAj) ``` Setting `subgroup = TRUE` places clones in a new subgroup node: ```{r clone-subgroup, out.width="70%", fig.align="center", message=FALSE, warning=FALSE} tree_GAk <- glottoTrees::clone_tip(tree_GA_abr, label = c("akar1243", "akak1252"), subgroup = TRUE) plot_glotto(tree_GAk) ``` Multiple clones can be created with the `n` argument: ```{r clone-n, out.width="70%", fig.align="center", message=FALSE, warning=FALSE} tree_GAl <- glottoTrees::clone_tip(tree_GA_abr, label = "akab1248", n = 3, subgroup = TRUE) plot_glotto(tree_GAl) ``` After cloning, tip labels may no longer be unique. `apply_duplicate_suffixes()` adds a hyphen-number suffix to duplicate labels: ```{r apply-suffixes, out.width="70%", fig.align="center", message=FALSE, warning=FALSE} tree_GAm <- glottoTrees::apply_duplicate_suffixes(tree_GAj) plot_glotto(tree_GAm) ``` ### How to move a tip {-} `move_tip()` moves a tip to a new position beneath a specified parent node: ```{r move-tip, out.width="70%", fig.align="center", message=FALSE, warning=FALSE} tree_GAn <- glottoTrees::move_tip(tree_GA_abr, label = "apuc1241", parent_label = "jeru1239") plot_glotto(tree_GAn) ``` ### How to move a node and its descendants {-} `move_node()` moves an internal node — along with all structure below it — to a new parent: ```{r move-node, out.width="70%", fig.align="center", message=FALSE, warning=FALSE} tree_GAo <- glottoTrees::move_node(tree_GA_abr, label = "jeru1239", parent_label = "okol1242") plot_glotto(tree_GAo) ``` ::: {.callout-note} ## Summary: toolkit for curating tree topology The functions `remove_tip()`, `keep_tip()`, `keep_as_tip()`, `convert_to_tip()`, `collapse_node()`, `add_tip()`, `clone_tip()`, `move_tip()` and `move_node()` provide a general-purpose toolkit for modifying a tree's set of tips and its subgrouping structure to match any typological sample. ::: ## How to add branch lengths {#add-lengths} Branch lengths convey information and most phylogenetic comparative methods — including genealogically-sensitive averages — are sensitive to relative branch lengths. Glottolog's trees contain equal branch lengths, which is almost certainly unrealistic. A good approximation to the most likely distribution of branch lengths under a variety of assumptions is **exponential** [@venditti2010phylogenies]: very long branches are rare; very short ones are frequent. `rescale_branches_exp()` implements this by setting the deepest branches to length 1/2, the next layer to 1/4, and so on: ```{r exp-ga, out.width="70%", fig.align="center", message=FALSE, warning=FALSE} tree_GAp <- glottoTrees::rescale_branches_exp(tree_GA_abr) plot_glotto(tree_GAp) ``` ```{r exp-ea, out.width="70%", fig.align="center", message=FALSE, warning=FALSE} tree_EA <- glottoTrees::get_glottolog_trees("Eskimo-Aleut") tree_EA_abr <- glottoTrees::abridge_labels(tree_EA) tree_EAa <- glottoTrees::rescale_branches_exp(tree_EA_abr) plot_glotto(tree_EAa, nodelabels = FALSE) ``` `ultrametricize()` stretches terminal branches so that all tips are equidistant from the root: ```{r ultrametric, out.width="70%", fig.align="center", message=FALSE, warning=FALSE} tree_EAb <- glottoTrees::ultrametricize(tree_EAa) plot_glotto(tree_EAb, nodelabels = FALSE) ``` `rescale_deepest_branches()` adjusts only the deepest layer of branches — useful when multiple family trees have been joined and you wish to control the implied distance between first-order branches. Here we triple the deepest branch length and then ultrametricise: ```{r deepest-branches, out.width="70%", fig.align="center", message=FALSE, warning=FALSE} tree_arnhem_a <- glottoTrees::rescale_branches_exp(tree_arnhem_abr) tree_arnhem_b <- glottoTrees::rescale_deepest_branches(tree_arnhem_a, 1.5) tree_arnhem_c <- glottoTrees::ultrametricize(tree_arnhem_b) plot_glotto(tree_arnhem_c, nodelabels = FALSE) ``` ## Exporting trees {#exporting} Trees can be saved in Newick format using `write.tree()` from `ape`, for use with other software such as FigTree: ```{r write-tree, eval=FALSE, message=FALSE, warning=FALSE} ape::write.tree(tree_arnhem_c, "my_arnhem_tree.tree") ``` `relabel_with_names()` replaces glottocodes with glottolog's full language names before plotting or exporting: ```{r relabel, out.width="70%", fig.align="center", message=FALSE, warning=FALSE} tree_arnhem_c_namelabels <- glottoTrees::relabel_with_names(tree_arnhem_c) plot_glotto(tree_arnhem_c_namelabels, nodelabels = FALSE) ``` --- # Worked Example: Yin (2020) {#example} ::: {.callout-note} ## Section Overview **What you will learn:** How to apply the full phylogenetic workflow — supertree construction, tip curation, branch length assignment, and genealogically-sensitive averaging — to a real published typological dataset. This section reproduces the phylogenetic analysis of @yin_violations_2020. ::: @yin_violations_2020 examined violations of the sonority sequencing principle in 496 languages and calculated the genealogically-sensitive proportions of languages in which various violations occurred. The language sample consisted of 496 languages from the CLICS2 database [@anderson2018clics2] and the AusPhon-Lexicon database [@round_ausphon-lexicon_2017]. ::: {.callout-warning} ## Computation Time The supertree construction in this section (`assemble_supertree()`) may take several minutes to run, depending on your hardware. This is normal. We recommend running the code chunks in this section interactively rather than knitting the full document. ::: ## The data {-} Yin's raw data consists of language names, glottocodes, and binary indicators of whether each language has consonant clusters with sonority reversals in word-initial onsets or word-final codas (coded 1 for yes, 0 for no). This dataset is provided with `phyloWeights` as the dataframe `yin_2020_data`, with columns `name`, `tip`, `has_onset_violation` and `has_coda_violation`. The first ten rows: ```{r yin-data, message=FALSE, warning=FALSE} head(yin_2020_data, n = 10) ``` ## Preparing the supertree {-} The tree for Yin's study was constructed from a glottolog supertree. The language sample covered relatively few families in the Americas (so North and South America were merged) and only one African language (Arabic), so Africa and Eurasia were also merged. We use glottolog v4.2, as in the original study, to ensure exact reproducibility: ```{r yin-supertree, message=FALSE, warning=FALSE} yin_macro <- list( c("South America", "North America"), c("Africa", "Eurasia"), "Papunesia", "Australia" ) supertree <- glottoTrees::assemble_supertree( macro_groups = yin_macro, glottolog_version = "4.2" ) supertree_abr <- glottoTrees::abridge_labels(supertree) ``` Five tips were cloned, in cases where Yin had data for two varieties corresponding to just one tip in the glottolog supertree: ```{r yin-clone, message=FALSE, warning=FALSE} supertree_a <- glottoTrees::clone_tip( supertree_abr, subgroup = TRUE, label = c("ayab1239", "basu1242", "biri1256", "ikar1243", "peri1265") ) supertree_b <- glottoTrees::apply_duplicate_suffixes(supertree_a) ``` Eight tips were added, in cases where for sister lects (A, B), glottolog placed A as a node above B. In such cases, a new tip A was placed below the existing glottolog node A: ```{r add-tips, message=FALSE, warning=FALSE} supertree_c <- supertree_b nodes_to_add_as_tips <- c("alor1249", "gami1243", "guri1247", "mand1415", "sins1241", "wang1291", "warl1254", "yand1253") for (node_i in nodes_to_add_as_tips) { supertree_c <- glottoTrees::add_tip(supertree_c, label = node_i, parent_label = node_i) } ``` From this supertree, only the 496 languages in Yin's dataset were kept. The internal node `mada1298` and all non-branching internal nodes were then collapsed: ```{r big-keep-tip, message=FALSE, warning=FALSE} supertree_d <- glottoTrees::keep_as_tip(supertree_c, label = yin_2020_data$tip) supertree_e <- glottoTrees::collapse_node(supertree_d, label = "mada1298") supertree_f <- glottoTrees::collapse_node( supertree_e, label = glottoTrees::nonbranching_nodes(supertree_e) ) ``` Finally, branch lengths were assigned. Branches were first assigned exponential lengths. Then, to diminish the importance of macro groups, the branches above them were shortened to 1/40. The effect is that the implied distance between families in different macro groups is only marginally greater than between families within a single macro group: ```{r branch-lengths, message=FALSE, warning=FALSE} supertree_g <- glottoTrees::rescale_branches_exp(supertree_f) yin_2020_tree <- glottoTrees::rescale_deepest_branches(supertree_g, 1/40) ``` The resulting tree, plotted as a fan with full language names: ```{r yin-tree, fig.dim=c(10, 10), fig.align="center", message=FALSE, warning=FALSE} full_names <- yin_2020_data$name[ match(yin_2020_tree$tip.label, yin_2020_data$tip) ] name_tree <- yin_2020_tree name_tree$tip.label <- full_names plot(ape::ladderize(name_tree, right = FALSE), type = "fan", cex = 0.3, label.offset = 0.002, edge.width = 0.5) ``` ## Preparing the dataframe {-} The dataframe `yin_2020_data` already has a `tip` column and two numerical columns (`has_onset_violation` and `has_coda_violation`), so it meets the requirements of `phylo_average()` directly. Non-numeric columns (here, `name`) are ignored automatically. ## Calculating genealogically-sensitive proportions {-} ```{r phylo-averaging, message=FALSE, warning=FALSE} yin_2020_results <- phyloWeights::phylo_average( phy = yin_2020_tree, data = yin_2020_data ) ``` Since we are using a single tree, the results are compact. The genealogically-sensitive proportions according to the ACL and BM methods: ```{r yin-avgs, message=FALSE, warning=FALSE} yin_2020_results$ACL_averages yin_2020_results$BM_averages ``` The first ten rows of phylogenetic weights from each method: ```{r yin-weights, message=FALSE, warning=FALSE} head(yin_2020_results$ACL_weights, n = 10) head(yin_2020_results$BM_weights, n = 10) ``` As a point of comparison, the raw (genealogically unweighted) proportions are the simple column means: ```{r raw-props, message=FALSE, warning=FALSE} mean(yin_2020_data$has_onset_violation) mean(yin_2020_data$has_coda_violation) ``` The difference between the raw proportions and the phylogenetically-weighted proportions reflects the correction for genealogical relatedness. Which direction the correction goes, and how large it is, depends on the distribution of the trait across the tree. --- # Using These Methods in Typological Research {#discussion} As we seek to analyse the empirical diversity of attested languages, there are fundamental reasons why genealogy must be part of the picture [see @macklin-cordes_challenges_nodate]. And since the genealogies of human languages are still incompletely known, it is imperative to make our phylogenetic hypotheses and assumptions as explicit and as testable as possible. Through `glottoTrees` and `phyloWeights`, linguistic trees and the code used to produce them can be published together with typological studies. This enables subsequent researchers to replicate findings and — crucially — to modify the phylogenetic assumptions and thereby test further hypotheses. ::: {.callout-note} ## Citing these packages If you find `glottoTrees` and `phyloWeights` useful in your research, please cite them as @glottoTrees and @phyloWeights respectively. To cite the notion of genealogically-sensitive averages and proportions, cite @macklin-cordes_challenges_nodate and/or the more specific references therein. If you encounter a bug or anomalous behaviour, please use the GitHub Issues pages or contact Erich Round directly at e.round\@uq.edu.au. ::: --- # Citation & Session Info {-} Round, Erich & Martin Schweinberger. 2026. *Practical Phylogenetic Methods for Linguistic Typology*. Brisbane: The Language Technology and Data Analysis Laboratory (LADAL). url: https://ladal.edu.au/tutorials/ladal_phylogentic_showcase/ladal_phylogentic_showcase.html (Version 2026.05.01). ``` @manual{round2026phylo, author = {Round, Erich and Schweinberger, Martin}, title = {Practical Phylogenetic Methods for Linguistic Typology}, note = {tutorials/ladal_phylogentic_showcase/ladal_phylogentic_showcase.html}, year = {2026}, organization = {The University of Queensland, Australia. School of Languages and Cultures}, address = {Brisbane}, edition = {2026.05.01} } ``` ::: {.callout-note} ## AI Transparency Statement This tutorial was adapted for LADAL by Martin Schweinberger with the assistance of **Claude** (claude.ai), a large language model created by Anthropic. The original tutorial text and all R code were authored by Erich Round and taken from the Supplementary Materials of @macklin-cordes_challenges_nodate. The adaptation involved converting the document to Quarto format, updating all `div` blocks to Quarto callouts, updating package installation code and glottolog version references, and adding LADAL-style section overviews. Erich Round and Martin Schweinberger take full responsibility for the accuracy of the content. ::: ```{r fin} sessionInfo() ``` --- [Back to top](#intro) [Back to LADAL home](/) --- # References {-}