plotting trajectories

The grid library is necessary for arrow()

library(ggplot2)
library(grid)

library(dplyr)
library(knitr)

ay <- read.csv("http://jofrhwld.github.io/data/ay.csv")
head(ay) %>% kable

X	Word	Time	F1	F2	F3	VClass	N	RTime	LDur
1	child129	0.025925	739.771	1522.028	2309.263	ay	29	0.0000000	Long
2	child129	0.027925	742.721	1496.464	2309.456	ay	29	0.0109890	Long
3	child129	0.029925	741.999	1487.178	2313.133	ay	29	0.0219780	Long
4	child129	0.031925	738.795	1483.538	2314.338	ay	29	0.0329670	Long
5	child129	0.033925	735.457	1472.834	2305.342	ay	29	0.0439560	Long
6	child129	0.035925	734.296	1441.201	2290.184	ay	29	0.0549451	Long

ay columns:

• Word: the word
• Time: Time in s, relative to vowel onset
• RTime: proportional time, ((Time-min(Time))/diff(range(ay$Time))
• F1, F2, F3
• N: A unique id for each vowel.

Raw data Plot

The group = N argument is crucial for it to draw one path per vowel token.

ggplot(ay, aes(F2, F1))+
  geom_path(aes(group = N), arrow = arrow())+
  scale_y_reverse()+
  scale_x_reverse()+
  coord_fixed()

The group = N argument is crucial for it to draw one path per vowel token. Without it, this is what it would look like.

ggplot(ay, aes(F2, F1))+
  geom_path(arrow = arrow())+
  scale_y_reverse()+
  scale_x_reverse()+
  coord_fixed()+
  ggtitle("Chaos")

Smooting the raw data

These are simple loess smooths of F1 and F2 over the proportional time.

f1_mod <- loess(F1 ~ RTime, data = ay)
f2_mod <- loess(F2 ~ RTime, data = ay)

After fitting the model, you get the predicted values with predict()

pred <- data.frame(RTime = seq(0,1, length = 100))
pred$F1 <- predict(f1_mod, newdata = pred)
pred$F2 <- predict(f2_mod, newdata = pred)

Plotting again. There’s no group argument this time, since the pred data frame contains only one trajectory.

ggplot(pred, aes(F2, F1))+
  geom_path(arrow = arrow())+
  scale_y_reverse()+
  scale_x_reverse()+
  coord_fixed()

It’s still not quite optimal, since there’s no sense of how fast or slow the trajectory is. Here’s one approach

# evenly spaced indices between 1 and 99
indices <- seq(1,99, by = 5)

ggplot(pred, aes(F2, F1))+
  geom_path(arrow = arrow())+
  geom_point(data = pred[indices,])+
  scale_y_reverse()+
  scale_x_reverse()+
  coord_fixed()