Grammatical Gender

A question that regularly pops up in conlang fora and groups is that of whether grammatical gender is good for anything.  Obviously, grammatical gender is not good enough to be universal. A few examples of languages that lack grammatical gender are

• English (with marginal gender - i.e. humans are still gendered, and you might find dialects where things that aren't neuter exist), Armenian, northern Ostrobothnian Swedish (at least Karleby - the situation is comparable to English), some Danish dialects
• Finnish, Estonian, Hungarian and related languages
• Turkish, Azeri and related languages
• Basque
• Georgian, Svan, Mingrelian

These are just the gender-less languages in or close to Europe. We find a whole slew of them elsewhere as well. Clearly a language can exist without gender, and as in the cases of Armenian and English, languages can lose gender as well. (Other examples of languages that have lost grammatical gender are Persian and some dialects of Swedish that have done so under influence from Finnish. The Danish dialects' loss of gender are more difficult to account for: the nearby region for those is quite well stacked with languages that have grammatical gender.) 

However, gender has been a remarkably stable feature of the Indo-European languages, the Semitic languages, and the Niger-Congo languages for millennia. There must be some benefits to having it. I do not posit that these benefits only can be gained from having gender, other strategies may also give the same benefits, but these benefits come bundled when you have gender - and that might be one reason why there's few signs of it disappearing from IE, Semitic and the Bantu branch of Niger-Congo, Ket, and a variety of other families.

I posit there are at least two major important advantages: added redundancy and distinguishing referents of anaphora. In addition, some minor advantages may also appear.

When we listen to, or even read a statement, most humans don't hear or register every phoneme or letter. In fact, we really just sample the utterance, and our brain fills out the remainder by some very good guesswork. However, we sometimes guess wrong, and if we are in a noisy environment, our sample may be too distorted by the background noise for us to reliably reconstruct what was uttered. (In texts, consider reading through dirty glasses, or a text that's been very worn).

In many languages with gender, gender congruence adds redundancy somewhat systematically, which helps the brain reconstruct the utterance more reliably. Errors still may happen, but the increase in redundancy reduces the risk for error. Other ways exist to reduce the risk ­- longer words, repetition of words, more 'context-sensitive' lexemes, etc, but gender is fairly efficient at it. Obviously, if no adjective or article is around to carry congruence, and there's no verb congruence, this added redundancy won't help - but chances are it'll appear often enough to make a difference often enough to be advantageous.

Another thing gender enables is a simple way of getting some of the benefits of proximate-obviate distinctions, i.e. distinguishing two (or more) third person referents. Of course gender is not the only way of resolving this: proximate-obviate systems may resolve which referent is proximate and which is obviate more reliably (i.e. there's always a way of figuring out which is which, unlike when two participants of the same gender co-occur), but this greater reliability generally requires a more complicated marking strategy, and thus the speaker must have planned-ahead far enough (unless obviative marking is mandatory on all non-proximate nouns). Obviously, gender will not enable such distinctions for every possible sentence, but there's a significant probability that two nouns you need to refer to will have different genders, and [you may sufficiently often derive enough benefit from it when they happen to be of different genders] that having grammatical genders becomes a good deal.

Conversely, gender may help distinguish homophonous nouns by means of the distinct congruence they trigger (thanks to Salmoneus for reminding me of that!). Congruence of course also is a powerful thing, see e.g. how the gender system in the Niger-Congo languages enables free word order yet no case marking is needed - the verb agrees with subject and with object, and that is generally sufficient. Some ways of achieving free word order without case and gender congruence exist (i.e. animacy hierarchies), but gender seems a rather convenient method.

These are the purely 'utilitarian' advantages of gender. Other ways of achieving the same advantages exist, but gender is a neat way of achieving several such benefits at once.