Chapter 7 — interactive

Reading Honest Signals

The symbiont tool: run the code from the book straight in your browser.

Run the four cells in order, change them, and watch how the result changes. You don't need to install anything. The full code is on GitHub.

Bee — the creative Ant — the doer Butterfly — the shaper Capybara — the harmoniser Leech — the taker

Pyodide is loading… (10–20 seconds the first time)

The chat: class trip to Lisbon

Five students plan their study trip. The same 33 messages as in the book are preloaded below as the file chat.txt — you may change them.

chat.txt

Mon 14:02 - Mira: I'd absolutely love Lisbon. What do you think of that?
Mon 14:03 - Mira: Idea: we could also travel back via Porto, it's supposed to be really beautiful.
Mon 14:05 - Tom: I checked the flights. Lisbon return from Duesseldorf about 180 euros in May. Here's the list with three providers.
Mon 14:11 - Lara: Ooh yes Lisbon! There's a viewing terrace above the rooftops, I could use it for our programme poster.
Mon 14:18 - Jonas: Both suggestions have something. Let's gather this calmly, no rush, before we decide.
Mon 18:42 - Felix: Idea: why not Lisbon AND Porto? Do both on one trip, that would really be something.
Tue 08:51 - Mira: Yes Felix, exactly like I already said yesterday. Porto would be great, Tom can you check train connections Lisbon-Porto?
Tue 09:14 - Mira: What do you think of building in a day of surfing? Cascais is supposed to have perfect waves for beginners.
Tue 09:30 - Tom: Hostel research is done. Three options, all under 30 euros a night, location checked. Table in the Drive.
Tue 09:42 - Lara: I'll take care of the programme booklet. Look, I've already designed a logo, does it fit Lisbon?
Tue 10:01 - Jonas: Lara, great work. Tom, thanks for the table. Mira, your surf day sounds good, does it cost extra?
Tue 14:20 - Mira: Imagine we also do a Fado concert in the evening. That'd be a whole different experience from just the beach.
Tue 14:35 - Tom: Specifically: surf day in Cascais costs 45 euros per person, Fado concert from 18 euros. I'll factor it into the budget.
Tue 22:11 - Felix: A watersports day on the Atlantic coast would be a highlight for me. Something like that just has to be part of Lisbon.
Wed 08:50 - Lara: Last night I made a mood board for the Insta post, orange sunset, white houses. Have a look!
Wed 09:02 - Mira: Lara that picture is beautiful! Idea: we could start the Insta account before the trip, as a bit of anticipation.
Wed 09:15 - Tom: Sign-up runs until Friday. Who's definitely in now? Please say yes or no here.
Wed 09:20 - Jonas: Yes from me.
Wed 09:21 - Lara: Yes of course! And I'd love to help design the sign-up email, shall I suggest something?
Wed 09:25 - Mira: Yes! And a new idea: each of us brings a song for our shared Spotify playlist.
Wed 13:55 - Felix: By the way, I showed my parents our whole Lisbon planning, the programme booklet, the Insta concept, everything. They were thrilled.
Thu 10:00 - Tom: Status: 5 sign-ups have come in. Felix, please check the hostel table again if you want to contribute anything, otherwise it goes out tomorrow morning.
Thu 10:14 - Jonas: Lara, the programme booklet turned out great, top work. Tom, thanks for booking the hostel.
Thu 11:30 - Lara: I've finished designing the programme booklet. Have a look, does the orange gradient fit the Lisbon vibe?
Thu 14:08 - Mira: Idea for the last evening: picnic by the Tejo with Pastel de Nata? Would be a lovely finish.
Thu 14:45 - Tom: Picnic: ok, costs about 8 euros per person for snacks. I'll take care of buying supplies on site.
Fri 08:30 - Felix: Frau Berger just spoke to me, I explained our travel planning to her. She was really impressed by what we pulled off.
Fri 08:45 - Tom: 5 of 5 confirmed. Sign-up is being submitted.
Fri 09:00 - Jonas: Great that it worked out. Lovely effort by everyone.
Fri 12:11 - Lara: Insta account is online: @11b_lisbon. First picture posted, have a look!
Fri 14:00 - Mira: What if we make a little travel-diary app? Each person writes three sentences in it each day.
Fri 14:20 - Lara: I like the idea, I'll create the layout for it.
Fri 14:35 - Jonas: Sounds good. Let's not make it too big, no stress, three sentences are enough.

The word lists

For each symbiont role a small list of typical words. Remember Chapter 6: these are roles, not labels for people.

Role	Typical words and phrases
Bee	idea, new, what do you think of, we could, imagine, what if
Ant	finished, done, i'll take care of, here's the list, specifically, status
Butterfly	beautiful, look, design, fit, mood, vibe, logo
Capybara	no rush, let's, no stress, thanks, sounds good, everyone, i understand, mediate
Leech	showed my parents, impressed by, we pulled off, a highlight for me

The workshop

Four steps, four cells with a shared memory. What you compute in Step 1 is still available in Step 2 — so run them from top to bottom.

Step 1Read the chat in as a file

Reads chat.txt and splits each line into time, sender and text.

messages = []
with open("chat.txt", encoding="utf-8") as f:
    for line in f:
        line = line.strip()
        if " - " not in line or ": " not in line:
            continue
        time, rest = line.split(" - ", 1)
        sender, text = rest.split(": ", 1)
        messages.append({"time": time, "sender": sender, "text": text})

print(f"Found {len(messages)} messages in total.")
print("Example:", messages[0])

Expected: “Found 33 messages in total.”

Step 2Write the word lists into the program

A dictionary (dict): the keys are the role names, the values the word lists.

wordlists = {
    "Bee":       ["idea", "new", "what do you think of", "we could",
                  "imagine", "what if"],
    "Ant":       ["finished", "done", "i'll take care of", "here's the list",
                  "specifically", "status"],
    "Butterfly": ["beautiful", "look", "design", "fit",
                  "mood", "vibe", "logo"],
    "Capybara":  ["no rush", "let's", "no stress", "thanks", "sounds good",
                  "everyone", "i understand", "mediate"],
    "Leech":     ["showed my parents", "impressed by",
                  "we pulled off", "a highlight for me"],
}
print("Word lists defined for", len(wordlists), "roles.")

Defines wordlists.

Step 3Count the words per person

The heart of it: for each person, count how often words from each list occur.

people = sorted({n["sender"] for n in messages})
counts = {p: {role: 0 for role in wordlists} for p in people}

for n in messages:
    text = n["text"].lower()
    for role, words in wordlists.items():
        for w in words:
            counts[n["sender"]][role] += text.count(w)

for p in people:
    hits = counts[p]
    top = max(hits, key=hits.get)
    print(f"{p:6s} -> {hits}   (most likely: {top})")

Needs Steps 1 and 2.

Step 4A little visualisation

A bar chart: five bars per person for the five roles. The first time, matplotlib loads briefly.

import matplotlib
matplotlib.use("agg")
import matplotlib.pyplot as plt, numpy as np, io, base64

roles = list(wordlists)
colour = {"Bee":"#E0A82E","Ant":"#6B7A8F","Butterfly":"#C77DA0",
          "Capybara":"#6F9E78","Leech":"#B5524C"}
people = sorted(counts)

x = np.arange(len(people)); w = 0.15
fig, ax = plt.subplots(figsize=(8.6, 4.4))
for i, role in enumerate(roles):
    vals = [counts[p][role] for p in people]
    ax.bar(x + (i - 2)*w, vals, w, label=role, color=colour[role],
           edgecolor="white", linewidth=0.4)

ax.set_xticks(x); ax.set_xticklabels(people)
ax.set_ylabel("Hits")
for s in ("top","right"): ax.spines[s].set_visible(False)
ax.grid(axis="y", color="#E5E5E5"); ax.set_axisbelow(True)
ax.legend(ncol=5, frameon=False, loc="upper center", bbox_to_anchor=(0.5, 1.14), fontsize=9)
ax.set_title("Who plays which role in the Lisbon chat?", pad=26, fontweight="bold")

buf = io.BytesIO()
plt.savefig(buf, format="png", dpi=150, bbox_inches="tight", facecolor="white")
plt.close(fig)
_chart_b64 = base64.b64encode(buf.getvalue()).decode()
print("Chart created.")

Needs Step 3.

Try it out

Change a word list in Step 2 — add the word suggest to the Bee, for instance — and run Steps 2, 3 and 4 again. How does the picture shift?

Or load your own, anonymised chat into chat.txt above, in the same format (Day HH:MM - Name: Text) and run it all again. Where are the limits of the method — and what does a plain word list miss that a language model would catch?