を vs が: With Potentials and 〜たい

You already know that a Japanese transitive verb marks its object with を: パンを食べる（たべる）"eat bread," 日本語（にほんご）を話す（はなす）"speak Japanese." Then you meet the potential form and the 〜たい desire ending, and suddenly the same object shows up wearing が: 日本語が話せる "can speak Japanese," 水（みず）が飲みたい "want to drink water." English never signals this switch — "speak" and "can speak" keep the same object either way — so it feels arbitrary. It is not. This page gives you a decision procedure for が vs を in exactly these two constructions, and explains why が is the traditional choice while を is quietly winning ground.

The core shift in one line

When a plain transitive verb becomes a potential (〜える／〜られる／できる) or takes 〜たい, the predicate stops describing an action and starts describing a state — an ability, or a want. In a state, the thing is no longer something you act on; it is the thing the state is about, i.e. its subject. And the subject of a state is the natural home of が.

The verb changed from 話す (do the action of speaking) to 話せる (be able to speak). The moment it did, 中国語 stopped being the thing acted on and became the thing the ability is about — so を relaxes into が. This is the same logic that puts が after 好き（すき）, 上手（じょうず）, and 分かる（わかる）; those predicates are stative by nature, and the potential and 〜たい drag ordinary action verbs into the same camp. The full stative-predicate family lives on the が with 好き, ほしい, できる page.

Potentials: object flips to が

Every verb has a potential form (読む→読める, 食べる→食べられる, する→できる, 来る→来られる). When you use it, the thing you are able to do naturally takes が.

Notice the mental jump the plain form primes you for the wrong particle: 納豆を食べる is fine ("eat nattō," a real action), so your reflex is to keep を when it goes potential. Resist it. 食べられる describes nattō as edible-to-you or not — a state — so が is the traditional target.

〜たい: wanting to do it

〜たい attaches to a verb stem to mean "want to do." Because a want is a state, the object of the wished-for action classically takes が.

Here ビール and 甘い物 are the craved things, front and center — exactly where が, the marker of the thing a state is about, belongs. Compare the plain action ビールを飲む "drink beer," where you are actively doing the drinking.

The honest part: が is traditional, を is rising

Now the truth textbooks often hide. With potentials and especially with 〜たい, を is extremely common in real modern Japanese and is not an error. 日本語を話せる and 水を飲みたい are said and written constantly. This is a genuine gradient — が is the traditional, prescriptively "correct," and never-wrong choice; を is the increasingly standard alternative, and grammarians have tracked its steady advance for decades. You will hear both from the same native speaker in one conversation.

So the real question is not "which is right" but "which sounds more natural here." Two forces tip the balance:

When the object is short and adjacent, が sounds crisp and natural:

When the object drifts away from the verb, or the phrase gets long and action-focused, を takes over and が can even start to sound stiff:

In those two, the object is folded into a longer, action-driven clause, and を is the smoother choice — が would over-spotlight the noun. The rule of thumb: short and adjacent → が feels right; long, separated, or action-focused → を feels right. Either is grammatical.

The hard boundary: が is flat wrong with a plain action verb

The flexibility above lives only inside the potential and 〜たい constructions. The instant the verb is a plain, non-potential action verb, the object must take を, and が is simply ungrammatical.

You cannot say ×パンが食べる for "eat bread." が there would mean "bread eats (something)" — bread as the actor. The stative reasoning that licensed が for 食べられる / 食べたい evaporates the moment the verb is a plain action, because there is no state for the noun to be the subject of. Keep the two worlds apart: potential / 〜たい → が or を; plain action → を only.

Why the drift, and why only here

Here is the distinguishing insight worth carrying away. Potentials and 〜たい sit on a fault line: they are built from transitive verbs (which demand を) but behave like stative predicates (which demand が). The noun is being pulled in two directions at once — its old life as a direct object (を) and its new life as the subject of a state (が) — and Japanese lets speakers resolve the tension either way. That is precisely why the を→が choice is gradient here and nowhere else. A pure stative like 好き or ほしい was never a transitive verb, so it has no を to fall back on: 車がほしい, never ×車を欲しい. And a pure action like 食べる was never stative, so it has no が to reach for. Only the hybrids — potential and 〜たい — straddle both, and only they give you the choice.

Common mistakes

Mistake 1 — Using が with a plain action verb. The が option belongs to potentials and 〜たい, not to plain 食べる／読む／飲む.

Mistake 2 — Thinking を is an error with 〜たい. It is not; it is very common. Do not "correct" a natural を into が by reflex.

Mistake 3 — Keeping を when the potential wants a crisp が. With a short, adjacent object, が is the more natural potential marking.

Mistake 4 — Over-generalizing the flexibility to 好き / できる / ほしい. The が-or-を freedom is unique to potentials and 〜たい. The pure statives take が only.

Key takeaways

• A plain transitive verb takes を; turn it into a potential or add 〜たい and the object may switch to が, because the predicate is now a state and the object is its subject.
• が is traditional and never wrong with potentials and 〜たい; を is common and fully accepted, especially with long, separated, or action-focused objects. This is a gradient, not a right/wrong split.
• Quick call: short + adjacent object → が; long / separated / action-focused → を.
• The freedom exists only for potentials and 〜たい (the hybrids of action and state). Plain action verbs take を only (×パンが食べる); pure statives 好き・できる・ほしい take が only (×映画を好き).