The decline of local anchoring: A quantitative investigation

Author

Bettelou Los, Gea Dreschler, Ans van Kemenade, Erwin Komen, Stefano Coretta

Packages

library(tidyverse)
theme_set(theme_light())
library(mgcv)
library(tidygam)

Read data

eng_hist <- read_csv("data/eng_hist.csv")

New names:
Rows: 21558 Columns: 34
── Column specification
──────────────────────────────────────────────────────── Delimiter: "," chr
(25): TextId, SubType, Title, Author, Genre, Cat, Locs, Locw, ft_clsMain... dbl
(7): ResId, Date, Size, Words, ft_antdist, Include, TextList lgl (2): ...32,
...33
ℹ Use `spec()` to retrieve the full column specification for this data. ℹ
Specify the column types or set `show_col_types = FALSE` to quiet this message.
• `` -> `...32`
• `` -> `...33`
• `` -> `...34`

eng_filt <- eng_hist %>%
  filter(Include == 1) %>%
  mutate(
    period = case_when(
      EnglishPeriod == "OE" ~ 1,
      EnglishPeriod == "ME" ~ 2,
      EnglishPeriod == "eModE" ~ 3,
      EnglishPeriod == "lModE" ~ 4
    ),
    EnglishPeriod = factor(EnglishPeriod, levels = c("OE", "ME", "eModE", "lModE")),
    Pentaset = factor(Pentaset, levels = c("Identity", "Inferred", "Assumed", "Inert", "New")),
    TextId = as.factor(TextId),
    cent = ifelse(
      str_length(Date) == 3,
      str_sub(Date, 1, 1),
      str_sub(Date, 1, 2)
    ) %>% as.numeric() + 1,
    genre = str_to_lower(Genre) %>%
      str_sub(1, 3)
  )

eng_count <- eng_filt %>%
  group_by(EnglishPeriod, period, TextId, Pentaset, Words) %>%
  count()

eng_count_date <- eng_filt %>%
  group_by(cent, TextId, Pentaset, Words) %>%
  count()

eng_count_genre <- eng_filt %>%
  group_by(EnglishPeriod, period, TextId, Pentaset, Words, genre) %>%
  count()

preps <- c("after", "at", "about", "among", "before", "between", "binnan", "by", "for", "from", "mid", "of", "on", "through", "till", "to", "upon", "without")

eng_preps <- eng_filt %>%
  mutate(
    ft_p_text = case_when(
      ft_p_text == "ymbe" ~ "about",
      ft_p_text == "ymben" ~ "about",
      ft_p_text == "vmben" ~ "about",
      ft_p_text == "ere" ~ "before",
      ft_p_text == "in" ~ "binnan",
      ft_p_text == "within" ~ "binnan",
      ft_p_text == "with" ~ "mid",
      ft_p_text == "until" ~ "till",
      ft_p_text == "unto" ~ "to",
      TRUE ~ ft_p_text
    )
  ) %>%
  filter(
    ft_p_text %in% preps
  )

Data checks

table(is.na(eng_hist$TextId))


FALSE 
21558

table(is.na(eng_hist$Pentaset))


FALSE  TRUE 
21557     1

table(is.na(eng_filt$TextId))


FALSE 
18192

table(is.na(eng_filt$Pentaset))


FALSE 
18192

eng_filt %>%
  count(EnglishPeriod)

# A tibble: 4 × 2
  EnglishPeriod     n
  <fct>         <int>
1 OE             6185
2 ME             4068
3 eModE          4942
4 lModE          2997

eng_filt %>%
  count(Pentaset)

# A tibble: 5 × 2
  Pentaset     n
  <fct>    <int>
1 Identity  5532
2 Inferred  4452
3 Assumed    475
4 Inert     2515
5 New       5218

eng_filt %>%
  count(EnglishPeriod, Pentaset)

# A tibble: 20 × 3
   EnglishPeriod Pentaset     n
   <fct>         <fct>    <int>
 1 OE            Identity  2367
 2 OE            Inferred  1139
 3 OE            Assumed    157
 4 OE            Inert      967
 5 OE            New       1555
 6 ME            Identity  1419
 7 ME            Inferred  1015
 8 ME            Assumed     72
 9 ME            Inert      351
10 ME            New       1211
11 eModE         Identity  1208
12 eModE         Inferred  1174
13 eModE         Assumed    118
14 eModE         Inert      421
15 eModE         New       2021
16 lModE         Identity   538
17 lModE         Inferred  1124
18 lModE         Assumed    128
19 lModE         Inert      776
20 lModE         New        431

eng_filt %>%
  ggplot(aes(EnglishPeriod)) +
  geom_bar()

eng_filt %>%
  ggplot(aes(Pentaset)) +
  geom_bar()

eng_filt %>%
  ggplot(aes(EnglishPeriod, fill = Pentaset)) +
  geom_bar() +
  scale_fill_brewer(palette = "Set1")

eng_filt %>%
  ggplot(aes(EnglishPeriod, fill = Pentaset)) +
  geom_bar(position = "fill") +
  scale_fill_brewer(palette = "Set1")

eng_filt %>%
  ggplot(aes(EnglishPeriod, fill = Pentaset)) +
  geom_bar(position = "dodge") +
  scale_fill_brewer(palette = "Set1")

eng_count %>%
  ggplot(aes(period, n, colour = Pentaset)) +
  geom_jitter(alpha = 0.2) +
  facet_wrap(~ Pentaset)

eng_count %>%
  ggplot(aes(period, n/Words, colour = Pentaset)) +
  geom_jitter(alpha = 0.2) +
  facet_wrap(~ Pentaset)

eng_filt %>%
  group_by(genre, Pentaset) %>%
  summarise(
    words = sum(Words),
    n = cumsum(n()),
    n_rel = n/words,
    .groups = "drop"
  ) %>%
  ggplot(aes(Pentaset, n_rel, fill = Pentaset)) +
  geom_bar(stat = "identity") +
  facet_wrap(genre ~.)

eng_filt %>%
  group_by(ft_objNP_type, Pentaset) %>%
  summarise(
    words = sum(Words),
    n = cumsum(n()),
    n_rel = n/words,
    .groups = "drop"
  ) %>%
  ggplot(aes(Pentaset, n_rel, fill = Pentaset)) +
  geom_bar(stat = "identity") +
  facet_wrap(ft_objNP_type ~.)

eng_preps %>%
  ggplot(aes(EnglishPeriod, fill = EnglishPeriod)) +
  geom_bar() +
  facet_grid(Pentaset ~ ft_p_text)

eng_preps %>%
  filter(
    Pentaset == "Identity",
    ft_p_text %in% c("after", "at", "between", "binnan", "by", "for", "from", "mid", "of", "on", "through", "to")
  ) %>%
  mutate(
    preposition = case_when(
      ft_p_text == "binnan" ~ "in",
      ft_p_text == "mid" ~ "with",
      TRUE ~ ft_p_text
    )
  ) %>%
  ggplot(aes(EnglishPeriod, fill = EnglishPeriod)) +
  geom_bar() +
  facet_wrap(~ preposition) +
  labs(
    x = "English period"
  ) +
  theme(
    axis.text.x = element_text(angle = 45, vjust = 0.5),
    legend.position = "none"
  )

ggsave("img/prep-period.png", width = 7, height = 5)

GAMs

Period and Pentaset

if (file.exists("./data/cache/gam_1.rds")) {
  gam_1 <- readRDS("./data/cache/gam_1.rds")
} else {
  gam_1 <- bam(
    n ~
      Pentaset +
      s(period, by = Pentaset, k = 3) +
      s(TextId, bs = "re", m = 1) +
      offset(log(Words/100000)),
    data = eng_count,
    family = poisson,
    discrete = TRUE
  )
  saveRDS(gam_1, "./data/cache/gam_1.rds")
}

predict_gam(
  gam_1, series = "period",
  values = c("Words" = 1e+5),
  exclude_terms = "s(TextId)",
  length_out = 50, tran_fun = exp
) %>%
  plot(comparison = "Pentaset") +
  geom_vline(xintercept = c(1:4), linetype = "dotted") +
  scale_color_brewer(type = "qual", palette = "Set1") +
  scale_fill_brewer(type = "qual", palette = "Set1") +
  scale_x_continuous(labels = c("OE", "ME", "eModE", "lModE"), minor_breaks = NULL) +
  labs(
    y = "Number of PP per 100k words" 
  )

ggsave("img/period-penta.png", width = 7, height = 5)

We fitted a generalised additive mixed-effects model to the number of prepositional phrases, using a Poisson distribution. We included the following terms (in parentheses, an explanation of how the term contributes to the model): Pentaset (Identity, Inferred, Assumed, Inert, New), as a parametric term (average number of prepositional phrases according to Pentaset), a smooth over Period (OE, ME, eModE, lModE) by Pentaset (change in number of prepositional phrases over Period by Pentaset), and a by-text factor smooth over period (to account for variations between texts). An offset term was also included to account for the fact that length (in words) differed across texts. The reported estimates are the number of prepositional phrases assuming a text length of 100k words.

The number of Identity PPs decreases from Old English to Late Modern English, while Inferred PPs show a less pronounced increase, especially after Middle English. The number of Inert PPs seems to decrease from Old to Middle English, while it increases again by Late Modern English. The opposite pattern can be observed in the New PPs: after increasing from Old to Middle English, they decrease again by Late Modern English. Assumed PPs are overall less frequent and don’t show clear patterns of change through time, perhaps with the exception of a small increase between Early and Late Modern English.

Century and Pentaset

if (file.exists("./data/cache/gam_2.rds")) {
  gam_2 <- readRDS("./data/cache/gam_2.rds")
} else {
  gam_2 <- bam(
    n ~
      Pentaset +
      s(cent, by = Pentaset, k = 4) +
      s(TextId, bs = "re", m = 1) +
      offset(log(Words/100000)),
    data = eng_count_date,
    family = poisson,
    discrete = TRUE
  )
  saveRDS(gam_2, "./data/cache/gam_2.rds")
}

predict_gam(
  gam_2, series = "cent",
  values = c("Words" = 1e+5),
  exclude_terms = "s(TextId)",
  length_out = 50, tran_fun = exp
) %>%
  plot(comparison = "Pentaset") +
  scale_color_brewer(type = "qual", palette = "Set1") +
  scale_fill_brewer(type = "qual", palette = "Set1") +
  labs(
    y = "Number of PP per 100k words" 
  )

ggsave("img/cent-penta.png", width = 7, height = 5)

We fitted a generalised additive mixed-effects model to the number of prepositional phrases, using a Poisson distribution. We included the following terms (in parentheses, an explanation of how the term contributes to the model): Pentaset (Identity, Inferred, Assumed, Inert, New), as a parametric term (average number of prepositional phrases according to Pentaset), a smooth over Century by Pentaset (change in number of prepositional phrases over Period by Pentaset), and a by-text factor smooth over period (to account for variations between texts). An offset term was also included to account for the fact that length (in words) differed across texts. The reported estimates are the number of prepositional phrases assuming a text length of 100k words.

The results are virtually identical to the model fitted with English Period. The number of Identity PPs decreases from Old English to Late Modern English, while Inferred PPs show a less pronounced increase, especially after Middle English. The number of Inert PPs seems to decrease from Old to Middle English, while it increases again by Late Modern English. The opposite pattern can be observed in the New PPs: after increasing from Old to Middle English, they decrease again by Late Modern English. Assumed PPs are overall less frequent and don’t show clear patterns of change through time, perhaps with the exception of a small increase between Early and Late Modern English.