lighthouse  vs lighthouse keeper
The room is downstairs  vs a downstairs room
His work is first class vs his first class work
The water is knee-deep vs knee-deep in water
In the afternoon vs afternoon tea
overseas vs an overseas student

I can't seem to detect a change in the lighthouse example.  The main feature in all of these, however, is that stress patterns change slightly depending where in the phrase a given word occurs, the most stressed position most often being the end of a phrase.

This phenomenon is treated in excruciating depth in The Sound Pattern of English by Chomsky and Halle.  If anyone is interested in the gory details, note especially their "Stress Adjustment Rule" and their "Compound and Nuclear Stress Rules".

At any rate, according to the authors, the basic idea is that English has four levels of stress (1, 2, 3, 4, from most to least stressed).  When pronouncing a word in isolation, we (supposedly) use only levels 1, 3, and 4.  But when forming phrase groups, we reduce the primary stress of one or more words from 1 to 2 (but not the other, primary word of the phrase, often the last word), thus forming a phrase out of two or more words -- an entire phrase containing only one syllable with stress level 1.

The main reason I had difficulty following their arguments was probably because I don't always speak exactly in conformance to what their rules supposedly describe!   But besides that, the difference between stress levels 1 and 2 can be subtle to the point of going unnoticed by many.  Personally, I often have difficulty hearing the difference.

But, in any case, here's an example.  When a sentence ends with afternoon, stress level 1 is used for the last syllable (-noon).  When used in the phrase afternoon tea, however, stress level 1 is used again for the last syllable, but what is the last syllable now is a different syllable, namely the syllable tea.  So in the phrase afternoon tea, the stress level for -noon is reduced to level 2, while tea takes level 1.

af(3)ternoon(1), but af(3)ternoon(2)tea(1).

The same principle is followed in Japanese cooking, so it's not JAPanese COOKing, but japaNESE COOKing, -nese taking stress level 2 (where it would have had level 1 as an isolated word), and cook- taking level 1.

jap(3)anese(1), but jap(3)anese(2)cook(1)ing.
___________

The remarks above apply to AmE, so maybe in BrE, it is JAPanese COOKing and AFTernoon TEA.  I don't know.  (The Chomsky and Halle book treats AmE, not BrE.)

CJ

Many have expressed that stress is unpredictable. I have come across a nice explanation for why stress in English seems irregular. The answer to this question is in a book for teachers called "Teaching Pronunciation: A Reference for Teachers of English to Speakers of Other Languages" by the renowned author Marianne Celce-Murcia and two other ones.

Here is an exercpt of the book that bears the answer:

Far from being random, stress placement in English words derives from the reather colorful history of the language. Today, roughly thirty percent of the vocabulary of  English stems from its Old English origins and retains the native Germanic stress patterns. Many of the remaining words have been acquired through historical events, such as the Norman Conquest, which brought much French vocabulary into English, or through the influences of Christian religion and academia, which have done much to secure the position of words of Greek and Latin origin in the English langauge. In more recent times, English has derived a poriton of its vocabulary from the indigenous Native American languages... and from its Spanish-speaking neighbors...

...Although load words in English may sometimes retain the stress patterns of the language from which they derive, the are more often incorporated into the stresss patterns of English, which imposes on the a more indigenous or Germanic stress pattern by moving the stress to an earlier syllable, often the first. We can see this in borrowings such as GRAmar (from French graMAIRE) and CHOColate (from Spanich chocoLAte). In fact, the longer a borrowed word has been in the English langauge, the more likely it is theat this type of stress shift will occur.

Factors that influence stress placement include the historical origin of a word, affixation, and the word's grammatical function in an utterance.

After this long nice introduction, the author cites examples for the factors that affect stress placement.

Again, the book is Teaching Pronunciation coothered by Celece-Murcia.

Hope that has helped enrich the stress discussion.