M384 · Logic · Week 4

Completeness for \(\mathcal A\), and the Classical Syllogistic \(\mathcal S\)

Monday: building models, the completeness proof for A, and a decision procedure. Wednesday and Friday: the classical syllogistic \(\mathcal S\). The four categorical sentence forms, their exact truth conditions, and the proof system.

Goals for this week

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

Day 1 (Monday, September 14): Canonical models and completeness for \(\mathcal A\)

Today we build a model out of the syntax of \(\mathcal A\), use it to prove completeness, and turn that proof into an algorithm that returns either a proof tree or a counter-model.

What we are using from before
  • The language \(\mathcal A\) has exactly one kind of sentence: All p are q, for nouns \(p,q\) from a set \(\mathsf P\).
  • A model is a pair \(\mathcal M=(M,[\![\ ]\!])\): a set \(M\) together with a subset \([\![p]\!]\subseteq M\) for each noun \(p\).
  • \(\mathcal M\vDash\) All \(p\) are \(q\)  iff  \([\![p]\!]\subseteq[\![q]\!]\).
  • \(\Gamma\vDash\varphi\) means every model of \(\Gamma\) is a model of \(\varphi\).
  • The proof system \(\mathbf A\) (same letter, different font) has two rules, and \(\Gamma\vdash\varphi\) means there is a proof tree with root \(\varphi\) whose non-axiom leaves belong to \(\Gamma\).
axiom
All \(p\) are \(p\)
All \(p\) are \(n\)All \(n\) are \(q\)
barbara
All \(p\) are \(q\)

We proved soundness: if \(\Gamma\vdash\varphi\) then \(\Gamma\vDash\varphi\). The open question is the other direction.

1. From proofs to a preorder

The completeness proof starts from a simple idea. The proof system tells us, for each pair of nouns, whether one is provably included in the other. That relation between nouns already looks like an ordering, so let us name it.

Definition 4.1 — the relation \(\le_\Gamma\)

Let \(\Gamma\) be a set of sentences. For nouns \(u,v\) write \[u\le_\Gamma v \quad\text{iff}\quad \Gamma\vdash \text{All } u \text{ are } v.\] We drop the subscript and write \(u\le v\) when \(\Gamma\) is clear.

For instance, if \(\Gamma=\{\text{All }a\text{ are }b,\ \text{All }b\text{ are }c\}\), then \(a\le b\) and \(b\le c\) by one-point trees, and \(a\le c\) by one application of (barbara).

Proposition 4.2

For every \(\Gamma\), the pair \((\mathsf P,\le_\Gamma)\) is a preorder.

Reflexivity. For any noun \(u\), the one-point tree justified by (axiom) proves All \(u\) are \(u\). So \(u\le u\). This is exactly what (axiom) is in the system for.

Transitivity. Suppose \(u\le v\) and \(v\le w\). Take a proof tree \(T_1\) with root All \(u\) are \(v\) and a proof tree \(T_2\) with root All \(v\) are \(w\). Put them side by side and apply (barbara). The result is a proof tree over \(\Gamma\) with root All \(u\) are \(w\), so \(u\le w\).

The two rules of the system are therefore not an arbitrary pair. They are precisely what is needed to make \(\le_\Gamma\) reflexive and transitive.

Definition 4.3 — down-set

In a preorder \(P\), the down-set of an element \(p\) is \[\downarrow\! p=\{q\in P: q\le p\}.\]

For instance, in a preorder with \(a\le b\le c\) and nothing else besides the reflexive pairs, \(\downarrow\! c=\{a,b,c\}\), \(\downarrow\! b=\{a,b\}\) and \(\downarrow\! a=\{a\}\).

2. The canonical model

We now build a model. The surprising part is where the material comes from: the points of the model are the nouns of the language, and the interpretation of a noun is read off from the proof system.

Definition 4.4 — the canonical model of \(\Gamma\)

Let \(\Gamma\) be a set of sentences of \(\mathcal A\). The canonical model \(\mathcal M=(M,[\![\ ]\!])\) is given by \[M=\mathsf P,\qquad [\![u]\!]=\ \downarrow\! u=\{v\in\mathsf P:\Gamma\vdash \text{All } v \text{ are } u\}.\]

In words: the universe is the set of nouns, and a noun \(u\) denotes the set of all nouns that \(\Gamma\) proves to be included in \(u\).

Example 4.5 — a canonical model computed by hand

Let \(\mathsf P=\{a,b,c,d\}\) and \[\Gamma=\{\text{All }a\text{ are }b,\ \ \text{All }b\text{ are }c,\ \ \text{All }d\text{ are }c\}.\]

Step 1: find what is provable. From All \(a\) are \(b\) and All \(b\) are \(c\), (barbara) gives All \(a\) are \(c\). Adding the three sentences of \(\Gamma\), the reflexive instances from (axiom), and nothing else, the provable sentences are

All \(a\) are \(a\),  All \(a\) are \(b\),  All \(a\) are \(c\),  All \(b\) are \(b\),  All \(b\) are \(c\),  All \(c\) are \(c\),  All \(d\) are \(c\),  All \(d\) are \(d\).

Step 2: read off the down-sets. \([\![u]\!]\) collects the nouns \(v\) with \(\Gamma\vdash\) All \(v\) are \(u\), so we scan the list above for sentences ending in \(u\):

\([\![a]\!]=\{a\}\qquad [\![b]\!]=\{a,b\}\qquad [\![c]\!]=\{a,b,c,d\}\qquad [\![d]\!]=\{d\}\)

Step 3: check the model against \(\Gamma\). All \(a\) are \(b\) needs \(\{a\}\subseteq\{a,b\}\): true. All \(b\) are \(c\) needs \(\{a,b\}\subseteq\{a,b,c,d\}\): true. All \(d\) are \(c\) needs \(\{d\}\subseteq\{a,b,c,d\}\): true. So this model satisfies \(\Gamma\).

Step 4: use it as a test. Is All \(c\) are \(b\) true here? That needs \(\{a,b,c,d\}\subseteq\{a,b\}\), which fails, since \(c\notin\{a,b\}\). So the canonical model is a model of \(\Gamma\) in which All \(c\) are \(b\) is false, and therefore \(\Gamma\nvDash\) All \(c\) are \(b\).

Step 3 of the example was not luck. It happens for every \(\Gamma\).

Lemma 4.6

The canonical model of \(\Gamma\) satisfies \(\Gamma\).

Let All \(p\) are \(q\) belong to \(\Gamma\). That is a one-point proof tree, so \(p\le q\). We must show \([\![p]\!]\subseteq[\![q]\!]\). Take \(v\in[\![p]\!]\). By Definition 4.4 this means \(v\le p\). Since \(\le\) is transitive (Proposition 4.2) and \(p\le q\), we get \(v\le q\), that is, \(v\in[\![q]\!]\). As \(v\) was arbitrary, \([\![p]\!]\subseteq[\![q]\!]\).

Theorem 4.7 — Completeness of \(\mathbf A\)

If \(\Gamma\vDash\) All \(p\) are \(q\), then \(\Gamma\vdash\) All \(p\) are \(q\).

Let \(\mathcal M\) be the canonical model of \(\Gamma\). By Lemma 4.6, \(\mathcal M\vDash\Gamma\). Our hypothesis says every model of \(\Gamma\) satisfies All \(p\) are \(q\), so in particular \(\mathcal M\vDash\) All \(p\) are \(q\). By the definition of truth, \([\![p]\!]\subseteq[\![q]\!]\).

Now we only need one element of \([\![p]\!]\), and there is an obvious candidate. Since \(p\le p\) by (axiom), we have \(p\in[\![p]\!]\). Hence \(p\in[\![q]\!]\), which by Definition 4.4 says \(\Gamma\vdash\) All \(p\) are \(q\).

Remark 4.8

The proof shows more than it claims. If \(\Gamma\nvdash\) All \(p\) are \(q\), then All \(p\) are \(q\) is already false in the canonical model. So to settle whether a sentence follows from \(\Gamma\), you never have to look at more than one model. A model with this property is called a characteristic model of \(\Gamma\).

Building a model out of syntactic objects can look like a category mistake at first. It is not. A model is a set with an interpretation function, and nothing stops the set from being a set of nouns. Completeness is a statement about the relation between syntax and semantics, so it is reasonable that one structure carrying both aspects settles it.

3. A decision procedure

Completeness is a theorem about all \(\Gamma\) and all \(\varphi\). It does not yet tell you what to do when someone hands you a specific finite \(\Gamma\) and a specific \(\varphi\). For that we reorganise the same information as a graph.

Definition 4.9 — graphs, paths, the all-graph

A graph is a pair \(\mathcal G=(G,\to)\) with \(G\) a set of nodes and \(\to\) a relation on \(G\).

A path from \(g\) to \(h\) is a finite sequence of nodes, each joined to the next by an edge. We write \(g\to^{*}h\) when such a path exists. Paths of length \(0\) are allowed, so \(g\to^{*}g\) always holds. The relation \(\to^{*}\) is the reflexive-transitive closure of \(\to\).

Given \(\Gamma\subseteq\mathcal A\) and a set \(S\) of nouns, the all-graph \(\mathcal G_{\Gamma,S}\) has node set \(S\) and \[x\to_\Gamma y \quad\text{iff}\quad \text{All } x \text{ are } y \text{ belongs to }\Gamma.\] The graph is read straight off \(\Gamma\): one edge per sentence, no reasoning involved.

For instance, with \(\Gamma=\{\text{All }a\text{ are }b,\ \text{All }b\text{ are }c\}\) and \(S=\{a,b,c\}\), the all-graph has the two edges \(a\to b\) and \(b\to c\). Then \(a\to^{*}c\) via the path \(a\to b\to c\), while \(c\to^{*}a\) fails because no edge leaves \(c\).

Theorem 4.10

Let \(\Gamma\) be a set of sentences of \(\mathcal A\), let \(S\) contain every noun occurring in \(\Gamma\), let \(\mathcal G\) be the all-graph of \(\Gamma\) on \(S\), and let \(p,q\in S\). The following are equivalent:

  1. \(p\to_\Gamma^{*}q\) in \(\mathcal G\);
  2. \(\Gamma\vdash\) All \(p\) are \(q\);
  3. \(\Gamma\vDash\) All \(p\) are \(q\).

(1)⇒(2). Induction on the length \(n\) of a path from \(p\) to \(q\). If \(n=0\) then \(p=q\) and (axiom) proves All \(p\) are \(p\). Suppose the claim holds for \(n\), and let a path of length \(n+1\) run from \(p\) to \(q\), with \(q'\) the node just before \(q\). The initial segment has length \(n\), so by induction hypothesis \(\Gamma\vdash\) All \(p\) are \(q'\). The last edge means All \(q'\) are \(q\) is in \(\Gamma\). One application of (barbara) at the bottom finishes it.

(2)⇒(3). This is soundness.

(3)⇒(1). We prove the contrapositive. Assume \(p\not\to_\Gamma^{*}q\), and let \(\mathcal N\) have universe \(S\) with \([\![u]\!]=\{v\in S: v\to^{*}_\Gamma u\}\). First, \(\mathcal N\vDash\Gamma\): if All \(x\) are \(y\) is in \(\Gamma\) and \(z\in[\![x]\!]\), then a path from \(z\) to \(x\) extended by the edge \(x\to y\) is a path from \(z\) to \(y\), so \(z\in[\![y]\!]\). Second, \(p\in[\![p]\!]\) because of the empty path, while \(p\notin[\![q]\!]\) by assumption. So \(\mathcal N\nvDash\) All \(p\) are \(q\).

Notice that the model \(\mathcal N\) in the last part is the canonical model again, cut down to the finitely many nouns that actually occur. That is what makes the whole thing computable.

Algorithm 4.11 — deciding \(\Gamma\vDash\) All \(p\) are \(q\)
  1. Let \(S\) be the set of nouns occurring in \(\Gamma\) together with \(p\) and \(q\).
  2. Build the all-graph \(\mathcal G=(S,\to_\Gamma)\): one edge \(x\to y\) for each sentence All \(x\) are \(y\) in \(\Gamma\).
  3. Compute the reflexive-transitive closure \(\to^{*}_\Gamma\).
  4. If \(p\to^{*}_\Gamma q\): answer yes, and turn a path \(p\to x_1\to\cdots\to q\) into a proof tree by applying (barbara) once per edge.
  5. If not: answer no, and return the counter-model with universe \(S\) and \([\![u]\!]=\{v\in S:v\to^{*}_\Gamma u\}\).
Example 4.12 — turning a path into a proof tree

Let \(\Gamma=\{\text{All }a\text{ are }b,\ \text{All }b\text{ are }c,\ \text{All }c\text{ are }d\}\). The path \(a\to b\to c\to d\) has three edges, so the tree has three leaves from \(\Gamma\) and two applications of (barbara), each one consuming the next edge:

All \(a\) are \(b\)All \(b\) are \(c\)
barbara
All \(a\) are \(c\)
All \(c\) are \(d\)
barbara
All \(a\) are \(d\)

A path with \(n\) edges gives a tree with \(n-1\) applications of (barbara); a path with \(0\) edges gives the one-point tree justified by (axiom).

Example 4.13 — a full run of the algorithm

Let \(\mathsf P=\{j,k,l,m,n,p,q\}\) and let \(\Gamma\) consist of

All \(j\) are \(k\),  All \(j\) are \(l\),  All \(k\) are \(l\),  All \(l\) are \(k\),  All \(l\) are \(m\),  All \(k\) are \(n\),  All \(m\) are \(q\),  All \(p\) are \(q\),  All \(q\) are \(p\).

Question: does \(\Gamma\vDash\) All \(p\) are \(n\)?

Steps 1–2. \(S\) is all seven nouns. The all-graph has nine edges, one per sentence (Figure 1).

j k l m n q p
Figure 1. The all-graph of \(\Gamma\) on \(S=\{j,k,l,m,n,p,q\}\). Each sentence All \(x\) are \(y\) contributes exactly one edge \(x\to y\). Note the two-way edges between \(k\) and \(l\) and between \(p\) and \(q\).

Step 3. Follow the arrows from \(p\). We can reach \(q\), and from \(q\) we return to \(p\). There is no edge out of \(p\) or \(q\) to anywhere else, so the only nodes reachable from \(p\) are \(p\) and \(q\).

Step 4–5. In particular \(p\not\to^{*}n\), so the answer is no, and the algorithm hands back a counter-model. Its universe is \(S\), and each \([\![u]\!]\) is the set of nodes from which \(u\) can be reached:

\([\![j]\!]=\{j\}\),   \([\![k]\!]=[\![l]\!]=\{j,k,l\}\),   \([\![m]\!]=\{j,k,l,m\}\),   \([\![n]\!]=\{j,k,l,n\}\),   \([\![p]\!]=[\![q]\!]=\{j,k,l,m,p,q\}\).

Check the conclusion: \(p\in[\![p]\!]\) but \(p\notin[\![n]\!]\), so All \(p\) are \(n\) is false here, while all nine sentences of \(\Gamma\) are true. It is often easier to see this in the associated preorder, drawn as a Hasse diagram in Figure 2: nouns that reach each other are merged into one node, and \([\![u]\!]\) is everything at or below \(u\).

j k, l m n p, q
Figure 2. The preorder \((S,\to^{*})\) of Example 4.13, drawn upward. Reading \([\![u]\!]\) off this picture: \([\![n]\!]\) is \(n\) together with everything below it, namely \(\{j,k,l,n\}\). Since \(p\) sits in a different branch, \(p\notin[\![n]\!]\).
Exercise 1 — run the decision procedure

Self-check only. Nothing here counts towards your grade.

Let \(\mathsf P=\{a,b,c,d,e,f,g,h\}\) and let \(\Gamma\) consist of the ten sentences

All \(a\) are \(b\),  All \(b\) are \(c\),  All \(c\) are \(a\),  All \(c\) are \(d\),  All \(d\) are \(f\),  All \(f\) are \(e\),  All \(e\) are \(d\),  All \(e\) are \(g\),  All \(g\) are \(h\),  All \(b\) are \(h\).

Draw the all-graph on paper first. Then work through Algorithm 4.11 for each of the following claims.

(a) Does the entailment hold?

(b) For the claim \(\Gamma\vDash\) All \(b\) are \(e\), give a path in the all-graph that witnesses it. Type the nouns along the path in order, separated by commas, starting with \(b\) and ending with \(e\). Any correct path is accepted.

(c) For the claim \(\Gamma\vDash\) All \(f\) are \(c\), the algorithm returns the counter-model \(\mathcal N\) with universe \(S\). Write down \([\![c]\!]\) and \([\![f]\!]\) in \(\mathcal N\), as comma-separated lists of nouns.

\([\![c]\!]=\)

\([\![f]\!]=\)

The all-graph has two cycles, \(a\to b\to c\to a\) and \(d\to f\to e\to d\), joined by the edge \(c\to d\), and then a tail \(e\to g\to h\) with a shortcut \(b\to h\).

(a) All \(b\) are \(e\): yes, \(b\to c\to d\to f\to e\). All \(f\) are \(c\): no; from \(f\) you can only reach \(d,e,f,g,h\), and none of the edges into the first cycle starts there. All \(a\) are \(h\): yes, \(a\to b\to h\). All \(g\) are \(d\): no; \(g\) reaches only \(g\) and \(h\).

(b) \(b,c,d,f,e\). Going round a cycle first, as in \(b,c,a,b,c,d,f,e\), is also a path; it just gives a bigger proof tree.

(c) \([\![c]\!]=\{a,b,c\}\) and \([\![f]\!]=\{a,b,c,d,e,f\}\). Since \(f\in[\![f]\!]\) but \(f\notin[\![c]\!]\), All \(f\) are \(c\) fails in \(\mathcal N\), while every sentence of \(\Gamma\) holds there.

Exercise 2 — a counter-model from the algorithm

Self-check only. Nothing here counts towards your grade.

Let \(\mathsf P=\{m,n,p,q,r,s\}\) and let \(\Gamma\) consist of

All \(m\) are \(n\),  All \(n\) are \(p\),  All \(p\) are \(m\),  All \(q\) are \(p\),  All \(r\) are \(q\),  All \(q\) are \(s\),  All \(s\) are \(q\).

Let \(\varphi\) be All \(p\) are \(s\). Decide with the algorithm whether \(\Gamma\vDash\varphi\). If it does not, enter a model of \(\Gamma\) in which \(\varphi\) is false. The universe below is \(S=\mathsf P\) itself, so you can enter the model \(\mathcal N\) from step 5 of the algorithm directly, but any model that does the job is accepted; the check evaluates your model, it does not compare it with a stored one.

Rows are the nouns being interpreted; columns are the elements of the universe. Tick the cell in row \(u\), column \(v\) to put the element \(v\) into \([\![u]\!]\).

From \(p\) the only nodes reachable are \(p, m, n\) (the cycle \(m\to n\to p\to m\)); the edge \(q\to p\) points into that cycle, not out of it. So \(p\not\to^{*}s\) and \(\Gamma\nvDash\varphi\).

Step 5 of the algorithm gives \([\![u]\!]=\{v: v\to^{*}u\}\):

\([\![m]\!]=[\![n]\!]=[\![p]\!]=\{m,n,p,q,r,s\}\),   \([\![q]\!]=[\![s]\!]=\{q,r,s\}\),   \([\![r]\!]=\{r\}\).

Then \(p\in[\![p]\!]\) but \(p\notin[\![s]\!]\). A smaller model also works: put a single element into exactly \([\![m]\!],[\![n]\!],[\![p]\!]\), the nouns reachable from \(p\), and leave the other three sets empty.

Day 2 (Wednesday, September 16): Syntax and semantics of \(\mathcal S\)

Today we add Some and No to the language, give each of the four classical sentence forms an exact truth condition, and check which claims of the traditional square of opposition survive.

4. The language \(\mathcal S\)

We keep the nouns and the models of the previous chapter and add two new sentence formers.

Definition 4.14 — syntax and semantics of S

Starting from a set \(\mathsf P\) of nouns, the sentences of \(\mathcal S\) are

All \(p\) are \(q\),   Some \(p\) are \(q\),   No \(p\) are \(q\),   for \(p,q\in\mathsf P\).

A model is exactly what it was: a set \(M\) with a subset \([\![p]\!]\subseteq M\) for each noun. Truth in a model is given by

\[\begin{aligned} \mathcal M&\vDash \text{All } p \text{ are } q &&\text{iff}\quad [\![p]\!]\subseteq[\![q]\!]\\ \mathcal M&\vDash \text{Some } p \text{ are } q &&\text{iff}\quad [\![p]\!]\cap[\![q]\!]\neq\varnothing\\ \mathcal M&\vDash \text{No } p \text{ are } q &&\text{iff}\quad [\![p]\!]\cap[\![q]\!]=\varnothing \end{aligned}\]

The notions \(\Gamma\vDash\varphi\) and \(\Gamma\nvDash\varphi\) are defined from this exactly as before.

Example 4.15 — truth in a concrete model

Let \(M=\{1,2,3,4,5\}\) with \([\![\text{hawks}]\!]=\{2\}\), \([\![\text{birds}]\!]=\{1,2,5\}\), \([\![\text{turtles}]\!]=\{3,4\}\).

  • \(\mathcal M\nvDash\) Some hawks are turtles, because \(\{2\}\cap\{3,4\}=\varnothing\).
  • \(\mathcal M\vDash\) No hawks are turtles, for the same computation read the other way.
  • \(\mathcal M\nvDash\) No hawks are birds, because \(\{2\}\cap\{1,2,5\}=\{2\}\), which is non-empty.
  • \(\mathcal M\nvDash\) No birds are birds, because \([\![\text{birds}]\!]\cap[\![\text{birds}]\!] =\{1,2,5\}\neq\varnothing\). In general No \(p\) are \(p\) is true exactly when \([\![p]\!]=\varnothing\).

5. The four categorical forms

Traditional logic sorts categorical sentences along two axes: quantity (universal or particular) and quality (affirmative or negative). Two choices on each axis give four combinations, and these are the four sentence forms. They carry one-letter names that go back to medieval Latin: the affirmative ones are labelled with the vowels of affirmo, giving A and I, and the negative ones with the vowels of nego, giving E and O.

LabelFormQuantity / qualityTrue in \(\mathcal M\) iff
AAll \(p\) are \(q\)universal affirmative \([\![p]\!]\subseteq[\![q]\!]\)
ENo \(p\) are \(q\)universal negative \([\![p]\!]\cap[\![q]\!]=\varnothing\)
ISome \(p\) are \(q\)particular affirmative \([\![p]\!]\cap[\![q]\!]\neq\varnothing\)
OSome \(p\) are not \(q\)particular negative \([\![p]\!]\setminus[\![q]\!]\neq\varnothing\)
The four categorical forms. Each row fixes a shape of sentence and the set-theoretic condition that makes it true.
A note on the language S

The official syntax of \(\mathcal S\) in Definition 4.14 contains only three sentence formers: All, Some and No. That covers A, E and I. It does not contain the O form Some \(p\) are not \(q\).

The O form needs a negation sitting on a noun: Some \(p\) are non-\(q\). Adding complemented nouns gives a larger language, called \(\mathcal S^{\dagger}\), and that is next week's topic. For this week we use the O form only when we talk about the four forms, as in the square below. Every reasoning problem you are asked to solve stays inside A, E and I.

qp A: All p are q pq E: No p are q pq I: Some p are q pq O: Some p are not q
Figure 3. The four forms. A and E say something about the whole of \([\![p]\!]\), so they are drawn with regions alone. I and O say that a particular region is inhabited, so they need a witness, drawn as a dot. The pictures are typical cases, not the only ones: for instance A also holds when \([\![p]\!]=\varnothing\), and that case has no picture with two visible circles.

6. The square of opposition, done properly

The traditional square arranges the four forms in a diagram and claims four kinds of relation between them. Now that the truth conditions are exact, we can simply check each claim.

What survives: the contradictories

Proposition 4.16

In every model \(\mathcal M\):

  1. \(\mathcal M\vDash\) All \(p\) are \(q\)  iff  \(\mathcal M\nvDash\) Some \(p\) are not \(q\);
  2. \(\mathcal M\vDash\) No \(p\) are \(q\)  iff  \(\mathcal M\nvDash\) Some \(p\) are \(q\).

(1) \([\![p]\!]\subseteq[\![q]\!]\) holds exactly when no element of \([\![p]\!]\) lies outside \([\![q]\!]\), that is, exactly when \([\![p]\!]\setminus[\![q]\!]=\varnothing\).

(2) The two truth conditions are \([\![p]\!]\cap[\![q]\!]=\varnothing\) and \([\![p]\!]\cap[\![q]\!]\neq\varnothing\), which are negations of one another by definition.

So A and O are exact contradictories, and so are E and I. These are the two diagonals of the square, and they hold with no side conditions at all.

What fails: everything on the sides

The traditional square also claims three further relations. Each one fails, and each one fails for the same reason: a model in which \([\![p]\!]=\varnothing\).

Example 4.17 — one model refutes three claims at once

Let \(M=\{1\}\), \([\![p]\!]=\varnothing\), \([\![q]\!]=\{1\}\). Then:

  • All \(p\) are \(q\) is true: \(\varnothing\subseteq\{1\}\).
  • No \(p\) are \(q\) is true: \(\varnothing\cap\{1\}=\varnothing\).
  • Some \(p\) are \(q\) is false: the intersection is empty.
  • Some \(p\) are not \(q\) is false: \(\varnothing\setminus\{1\}=\varnothing\).

Reading this off:

  • Subalternation (A implies I, and E implies O) fails: A is true and I is false here.
  • Contrariety (A and E cannot both be true) fails: both are true here.
  • Subcontrariety (I and O cannot both be false) fails: both are false here.
Why this is not a defect of our semantics

Aristotle and the medieval tradition read All \(p\) are \(q\) as carrying existential import: saying it commits you to there being some \(p\). Under that reading the whole square is correct. Our Definition 4.14 takes the modern reading, on which All \(p\) are \(q\) is true whenever \([\![p]\!]\) is empty.

Neither reading is wrong; they are different sentences. What matters is that once you fix the truth condition, which inferences are valid is settled, and you can no longer appeal to the picture. If you want the traditional square back, you have to add the assumption Some \(p\) are \(p\) to \(\Gamma\).

A E I O All p are q No p are q Some p are q Some p are not q contraries — fails subcontraries — fails sub- alternation fails sub- alternation fails contradictories (these hold)
Figure 4. The square of opposition under the semantics of Definition 4.14. Only the diagonals survive. The dashed relations all fail in the one-point model of Example 4.17.

7. Reasoning semantically in \(\mathcal S\)

Before we have a proof system, entailment claims have to be settled by hand: a positive claim needs an argument about arbitrary models, a negative claim needs one counter-model.

Example 4.18 — All \(p\) are \(q\), Some \(p\) are \(r\) \(\vDash\) Some \(q\) are \(r\)

Let \(\mathcal M\) be any model of the two hypotheses. So \([\![p]\!]\subseteq[\![q]\!]\) and \([\![p]\!]\cap[\![r]\!]\neq\varnothing\).

The second hypothesis hands us an element: pick \(x\in[\![p]\!]\cap[\![r]\!]\). Then \(x\in[\![p]\!]\), and the first hypothesis gives \(x\in[\![q]\!]\). Also \(x\in[\![r]\!]\). So \(x\in[\![q]\!]\cap[\![r]\!]\), which is therefore non-empty, that is, \(\mathcal M\vDash\) Some \(q\) are \(r\).

\(\mathcal M\) was arbitrary, so the entailment holds.

The shape of this argument is worth remembering: a Some hypothesis is used by naming a witness, and an All hypothesis is used by pushing that witness into a bigger set.

Example 4.19 — All \(p\) are \(v\), All \(q\) are \(w\), No \(v\) are \(w\) \(\vDash\) No \(p\) are \(q\)

Fix a model of the hypotheses: \([\![p]\!]\subseteq[\![v]\!]\), \([\![q]\!]\subseteq[\![w]\!]\) and \([\![v]\!]\cap[\![w]\!]=\varnothing\). Then \[[\![p]\!]\cap[\![q]\!]\ \subseteq\ [\![v]\!]\cap[\![w]\!]\ =\ \varnothing,\] because intersection is monotone in both arguments. A subset of the empty set is empty, so \([\![p]\!]\cap[\![q]\!]=\varnothing\), which is the conclusion.

Example 4.20 — a non-entailment

Claim: Some \(p\) are \(q\), Some \(q\) are \(n\), All \(q\) are \(m\) \(\nvDash\) Some \(p\) are \(n\).

One counter-model suffices. Take \(M=\{1,2\}\) with \([\![p]\!]=\{1\}\), \([\![q]\!]=\{1,2\}\), \([\![n]\!]=\{2\}\), \([\![m]\!]=\{1,2\}\).

  • Some \(p\) are \(q\): \(\{1\}\cap\{1,2\}=\{1\}\neq\varnothing\). True.
  • Some \(q\) are \(n\): \(\{1,2\}\cap\{2\}=\{2\}\neq\varnothing\). True.
  • All \(q\) are \(m\): \(\{1,2\}\subseteq\{1,2\}\). True.
  • Some \(p\) are \(n\): \(\{1\}\cap\{2\}=\varnothing\). False.

The point is that two Some sentences sharing a noun need not share a witness. Here \(q\) is met by \(p\) at the point \(1\) and by \(n\) at the point \(2\), and nothing forces these to coincide.

A trick for building counter-models

When you have to falsify something, you need a universe and you need witnesses for the Some sentences. A convenient choice is to take the Some sentences of \(\Gamma\) themselves as the points of the model, and to put the sentence Some \(x\) are \(y\) into \([\![x]\!]\) and \([\![y]\!]\). Each Some sentence then witnesses itself and nothing else, which is exactly the minimum needed. In Example 4.20 this produces a two-point model that is the same as the one above after renaming.

This is the same move as in the canonical model of Day 1, where the points were nouns. Nothing requires the elements of a model to be numbers.

Exercise 3 — truth in a model

Self-check only. Nothing here counts towards your grade.

Work in the model \(M=\{1,2,3,4,5\}\) with \([\![\text{hawks}]\!]=\{2\}\), \([\![\text{birds}]\!]=\{1,2,5\}\), \([\![\text{turtles}]\!]=\{3,4\}\). Mark each sentence true or false.

  1. All hawks are birds — true: \(\{2\}\subseteq\{1,2,5\}\).
  2. Some birds are turtles — false: \(\{1,2,5\}\cap\{3,4\}=\varnothing\).
  3. No hawks are turtles — true: same intersection, empty.
  4. Some birds are not hawks — true: \(\{1,2,5\}\setminus\{2\}=\{1,5\}\).
  5. All turtles are hawks — false: \(3\in\{3,4\}\) but \(3\notin\{2\}\).
  6. Some hawks are hawks — true: \(\{2\}\cap\{2\}=\{2\}\neq\varnothing\). This sentence is true exactly when \([\![\text{hawks}]\!]\neq\varnothing\).
Exercise 4 — refute subalternation yourself

Self-check only. Nothing here counts towards your grade.

Build a model over \(M=\{1,2,3\}\) in which All \(p\) are \(q\) is true and Some \(p\) are \(q\) is false. Tick the elements you want in each set. Your model is tested directly, so any correct choice is accepted.

You have no choice about \([\![p]\!]\): it must be empty. If some \(x\) were in \([\![p]\!]\), then All \(p\) are \(q\) would put \(x\) into \([\![q]\!]\) as well, and then \(x\) would witness Some \(p\) are \(q\).

So take \([\![p]\!]=\varnothing\) and let \([\![q]\!]\) be anything at all, for instance \(\{1,2,3\}\) or \(\varnothing\). Both sentences then come out as required.

Day 3 (Friday, September 18): The proof system for \(\mathcal S\)

Today we set up the proof system \(\mathbf S\), check that each rule is sound, and build proof trees with it.

8. The rules

Definition 4.21 — the proof system \(\mathbf S\) for \(\mathcal S\)
axiom
All \(p\) are \(p\)
All \(p\) are \(n\)All \(n\) are \(q\)
barbara
All \(p\) are \(q\)
Some \(p\) are \(q\)
some1
Some \(p\) are \(p\)
Some \(p\) are \(q\)
some2
Some \(q\) are \(p\)
All \(q\) are \(n\)Some \(p\) are \(q\)
darii
Some \(p\) are \(n\)
All \(p\) are \(q\)No \(q\) are \(r\)
camestres
No \(r\) are \(p\)
No \(q\) are \(q\)
zero
All \(q\) are \(p\)
No \(p\) are \(p\)
no
No \(p\) are \(q\)
No \(p\) are \(q\)Some \(p\) are \(q\)
X
\(\varphi\)

In (X) the conclusion \(\varphi\) is an arbitrary sentence of \(\mathcal S\). The letters \(p,q,n,r\) range over all nouns, and different letters need not name different nouns.

9. Why each rule is sound

Soundness is again proved by induction on proof trees, exactly as for \(\mathbf A\). The induction itself is routine; all the content sits in checking one rule at a time. Fix a model \(\mathcal M\).

Soundness of (camestres)

Assume \([\![p]\!]\subseteq[\![q]\!]\) and \([\![q]\!]\cap[\![r]\!]=\varnothing\). Suppose towards a contradiction that \([\![r]\!]\cap[\![p]\!]\neq\varnothing\), and take \(m\) in it. From \(m\in[\![p]\!]\) and the first assumption, \(m\in[\![q]\!]\). Together with \(m\in[\![r]\!]\) this puts \(m\) in \([\![q]\!]\cap[\![r]\!]\), contradicting the second assumption. So \([\![r]\!]\cap[\![p]\!]=\varnothing\).

q p r p sits inside q, and r misses q entirely, so r cannot reach p
Figure 5. (camestres) in a picture. The premises place \([\![p]\!]\) inside \([\![q]\!]\) and push \([\![r]\!]\) off \([\![q]\!]\) altogether; the conclusion records that \([\![r]\!]\) and \([\![p]\!]\) therefore cannot meet.

10. Proof trees in \(\mathbf S\)

Example 4.22 — conversion of No: No \(p\) are \(q\) \(\vdash\) No \(q\) are \(p\)

The system has no rule that simply swaps the two nouns in a No sentence. We get the effect out of (camestres) by feeding it a trivial left premise.

axiom
All \(p\) are \(p\)
No \(p\) are \(q\)
camestres
No \(q\) are \(p\)

Matching against the rule: its left premise All \(p\) are \(q\) is instantiated as All \(p\) are \(p\), so the rule's \(q\) is our \(p\); its right premise No \(q\) are \(r\) is instantiated as No \(p\) are \(q\), so the rule's \(r\) is our \(q\); the conclusion No \(r\) are \(p\) therefore reads No \(q\) are \(p\). This derived step is often called E-conversion, or (anti).

Example 4.23 — All \(p\) are \(v\), All \(q\) are \(w\), No \(v\) are \(w\) \(\vdash\) No \(p\) are \(q\)

This is Example 4.19 again, now as a proof rather than a semantic argument. Two applications of (camestres) are enough.

All \(q\) are \(w\)
All \(p\) are \(v\)No \(v\) are \(w\)
camestres
No \(w\) are \(p\)
camestres
No \(p\) are \(q\)

Upper step: the rule's \(p,q,r\) are our \(p,v,w\), giving No \(w\) are \(p\). Lower step: the rule's \(p,q,r\) are our \(q,w,p\), giving No \(p\) are \(q\). The characteristic feature of (camestres) is that it swaps the order of the nouns each time, so applying it twice restores the order you wanted.

Example 4.24 — All \(n\) are \(q\), All \(n\) are \(p\), Some \(n\) are \(n\) \(\vdash\) Some \(p\) are \(q\)

In words: if there is at least one \(n\), and every \(n\) is both a \(p\) and a \(q\), then some \(p\) is a \(q\). This is the inference that the traditional square treated as automatic; here it needs the extra premise Some \(n\) are \(n\).

All \(n\) are \(p\)Some \(n\) are \(n\)
darii
Some \(n\) are \(p\)
some2
Some \(p\) are \(n\)
All \(n\) are \(q\)
darii
Some \(p\) are \(q\)

The bottom step needs its All premise on the left and its Some premise on the right, in the order given in Definition 4.21; (some2) is what puts the nouns in the position that (darii) can use.

Example 4.25 — an inconsistent \(\Gamma\) proves everything

Let \(\Gamma=\{\text{Some }p\text{ are }q,\ \text{No }q\text{ are }p\}\) and let \(\varphi\) be any sentence at all, say All \(m\) are \(n\). We first convert the No sentence as in Example 4.22, and then the two contradictory sentences meet at (X).

axiom
All \(q\) are \(q\)
No \(q\) are \(p\)
camestres
No \(p\) are \(q\)
Some \(p\) are \(q\)
X
All \(m\) are \(n\)
Where this is going

Soundness of \(\mathbf S\) holds, by the induction sketched in section 9. Completeness is only partly available at this point. What can be proved now is the particular half: if \(\Gamma\vDash\) Some \(p\) are \(q\), then \(\Gamma\vdash\) Some \(p\) are \(q\). The proof builds a second canonical model whose points are the Some sentences of \(\Gamma\), in the style of the trick in section 7.

That model does not settle the All and No cases; it can satisfy an All sentence that is not provable. Repairing this is easier in the larger language \(\mathcal S^{\dagger}\), where negation is allowed on nouns and No \(p\) are \(q\) becomes a notational variant of All \(p\) are non-\(q\). That is next week.

Exercise 5 — is this a single legal step?

Self-check only. Nothing here counts towards your grade.

For each line, decide whether the conclusion follows from the premises by one application of one rule of Definition 4.21. Some lines are semantically correct but still need more than one step, and those count as "no".

  1. Yes, (barbara) with middle noun \(b\).
  2. Yes, (some2).
  3. No. Nothing licenses swapping the nouns of an All sentence, and it is not even semantically valid: take \([\![a]\!]=\varnothing\), \([\![b]\!]=\{1\}\).
  4. Yes, (darii): the rule's All premise is All \(b\) are \(c\) and its Some premise is Some \(a\) are \(b\).
  5. No. (darii) needs Some \(p\) are \(q\) with the same \(q\) as in the All premise. Here the All premise is about \(a\) and the Some premise reads Some \(a\) are \(c\), which has \(a\) on the wrong side. Apply (some2) first and it becomes legal, so this is valid but not in one step.
  6. Yes, (camestres) with the rule's \(p,q,r\) taken as \(a,b,c\).
  7. No. E-conversion is derivable (Example 4.22) but it is not one of the nine rules, so it is not a single step.
  8. Yes, (X): the premises are contradictory, so any conclusion is allowed.
Exercise 6 — complete the proof tree

Self-check only. Nothing here counts towards your grade.

Here is a derivation of Some \(c\) are \(d\) from \(\Gamma=\{\text{Some }a\text{ are }b,\ \text{All }a\text{ are }c,\ \text{All }b\text{ are }d\}\), with two nodes left blank. Fill them in. Write sentences in the form All a are b, Some a are b or No a are b.

All \(a\) are \(c\)
All \(b\) are \(d\)Some \(a\) are \(b\)
darii
some2
Some \(d\) are \(a\)
darii
some2
Some \(c\) are \(d\)

The check re-runs every step of the tree against the rules, so it tells you which step breaks if your answer is wrong.

Node (a) is Some \(a\) are \(d\). The rule (darii) applied to All \(b\) are \(d\) and Some \(a\) are \(b\) has the rule's \(q=b\), \(n=d\), \(p=a\), so the conclusion is Some \(a\) are \(d\).

Node (b) is Some \(d\) are \(c\). The rule (darii) applied to All \(a\) are \(c\) and Some \(d\) are \(a\) has \(q=a\), \(n=c\), \(p=d\), so the conclusion is Some \(d\) are \(c\). One more (some2) turns it into Some \(c\) are \(d\).

Homework 4

Before you start

  • Fill in your name below. Do not write your student number; it is already attached to your Canvas upload.
  • Show your work. An answer of "yes" or "no" with nothing behind it earns no points. For a positive entailment claim, give an argument about arbitrary models or a proof tree; for a negative one, give a specific counter-model and check every sentence in it.
  • Two answer modes. Every question has a Type mode and a Write by hand mode. In Write by hand mode the text boxes are replaced by a ruled area you can write on with a pen, a finger or the mouse. Only the mode you have selected for a question goes into the PDF. Switching modes does not delete what you entered in the other mode.
  • How to submit. Press "Save homework as PDF" at the bottom, save the file, and upload it to Canvas.
  • Your answers are saved automatically in this browser. If you open the file on a different computer they will not be there.
  • Total: 100 points.

Question 1 — a canonical model15 points

Let \(\mathsf P=\{a,b,c,d\}\) and \[\Gamma=\{\text{All }a\text{ are }b,\ \ \text{All }b\text{ are }c,\ \ \text{All }c\text{ are }a,\ \ \text{All }d\text{ are }b\}.\]

(a) (4 pts) Describe the all-graph of \(\Gamma\) on \(\mathsf P\): list its nodes and its edges.
(b) (7 pts) Write down \([\![a]\!]\), \([\![b]\!]\), \([\![c]\!]\) and \([\![d]\!]\) in the canonical model of \(\Gamma\).
(c) (4 pts) For each of All \(d\) are \(a\), All \(a\) are \(d\) and All \(c\) are \(b\), say whether \(\Gamma\) proves it, and justify your answer using your model.

Question 2 — running the decision procedure20 points

Let \(\mathsf P=\{j,k,m,n,p,q\}\) and let \(\Gamma\) consist of

All \(j\) are \(k\),  All \(k\) are \(m\),  All \(m\) are \(k\),  All \(n\) are \(m\),  All \(p\) are \(n\),  All \(q\) are \(p\),  All \(p\) are \(q\).

(a) (8 pts) Decide whether \(\Gamma\vDash\) All \(q\) are \(k\). If it holds, give a proof tree in the system \(\mathbf A\), writing the rule name at each step. If it does not, give a counter-model and check it.
(b) (8 pts) Do the same for \(\Gamma\vDash\) All \(k\) are \(n\).
(c) (4 pts) Your two answers were obtained from the same graph. Explain in two or three sentences which step of Algorithm 4.11 produced each answer, and why the algorithm is guaranteed to give one of the two.

Question 3 — the four forms20 points

Work in the model \(M=\{1,2,3,4,5,6\}\) with \([\![a]\!]=\{1,2\}\), \([\![b]\!]=\{2,3,4\}\), \([\![c]\!]=\{5,6\}\), \([\![d]\!]=\varnothing\).

(a) (12 pts) Say whether each sentence is true or false in this model, and give the set computation behind each answer.
  1. All \(a\) are \(b\)
  2. Some \(a\) are \(b\)
  3. No \(b\) are \(c\)
  4. Some \(b\) are not \(a\)
  5. All \(d\) are \(c\)
  6. Some \(d\) are \(d\)
(b) (4 pts) Subalternation failed only because \([\![p]\!]\) was allowed to be empty. Show that one extra assumption repairs it: prove that in every model \(\mathcal M\), if \(\mathcal M\vDash\) All \(p\) are \(q\) and \(\mathcal M\vDash\) Some \(p\) are \(p\), then \(\mathcal M\vDash\) Some \(p\) are \(q\).
(c) (4 pts) Suppose that in some model both Some \(p\) are \(q\) and Some \(p\) are not \(q\) are false. What does that force \([\![p]\!]\) to be, and why? Is All \(p\) are \(q\) then true in that model?

Question 4 — semantic reasoning in S20 points

(a) (7 pts) Show that No \(p\) are \(q\), All \(r\) are \(p\) \(\vDash\) No \(r\) are \(q\). Argue about an arbitrary model; do not use the proof system.
(b) (7 pts) Show that All \(p\) are \(q\), Some \(q\) are \(r\) \(\nvDash\) Some \(p\) are \(r\). Give a counter-model and verify all three sentences in it.
(c) (6 pts) Decide whether All \(p\) are \(q\), No \(q\) are \(q\) \(\vDash\) No \(p\) are \(p\), and justify your answer.

Question 5 — proof trees in S25 points

Give a proof tree for each of the following, using only the rules of Definition 4.21. Write the name of the rule at every step. Leaves must be sentences of \(\Gamma\) or instances of (axiom).

(a) (5 pts) Some \(p\) are \(q\), No \(p\) are \(q\) \(\vdash\) All \(m\) are \(n\).
(b) (6 pts) Some \(p\) are \(q\), All \(q\) are \(r\) \(\vdash\) Some \(r\) are \(p\).
(c) (7 pts) All \(p\) are \(q\), No \(q\) are \(r\) \(\vdash\) No \(p\) are \(r\). Note the order of the nouns in the conclusion.
(d) (7 pts) All \(a\) are \(b\), All \(b\) are \(c\), All \(c\) are \(e\), No \(e\) are \(d\) \(\vdash\) No \(d\) are \(a\).