On The Technical Aspects Of A Global Constructed Auxiliary Language
By Behemoth4
“If as one people speaking the same language they have begun to do this, then nothing they plan to do will be impossible for them." - Genesis 11:6
In the competition for the global lingua franca, English reigns supreme. It is the most widely spoken language of all time, spoken by one in seven people and used around the globe. The influence of English is still growing, and it is very unlikely it is going to be dethroned in the foreseeable future.
Still, it is obvious to anyone that English, with its notoriously unruly orthography and wide variety of irregularities, is not ideal for international communication. In fact, some argue that no natural language is worthy of the title, especially one associated with a dominant culture. Instead, we should construct a language for all humankind, a synthetic tongue free of the complications that naturally arise as language drifts. An international auxiliary language, or IAL.
Many utopists (very few of whom were linguists) have tried their hand at designing such a language. None have succeeded. What those utopists fail to appreciate is the power of incentives: people want to learn languages that are useful enough to be worth the effort, which means languages that provide access to a lot of speakers. The growth of IALs, which start from zero speakers, is fueled only by idealism, which soon runs out. Esperanto, the most successful IAL by a landslide, has 1.6 million speakers by a generous estimate, comparable to Latvian with 1.75 million.
Should an IAL ever rise to the dominant language of the world, it will not be any of those known today. Instead, it would likely be designed by a team of expert linguists commissioned and backed by some powerful international entity, like the UN, after the fall of English as the lingua franca, which could be even centuries after the fall of the US as a cultural superpower: Latin survived well into the Renaissance despite the fall of Western Rome a thousand years earlier. Any hobbyist IAL is pretty much doomed from the start.
This essay doesn't aim to change that fact. Instead, it treats designing auxlangs as an interesting technical problem, without regard for their practicality. My interest was sparked by Justin B. Rye's thorough teardown of Esperanto, where the king of IALs, and conlangs as a whole, is shown to be riddled with the same problems amateur conlangers are expected to outgrow in a few years. I realized I could do better with a few months of focused work.
Creating an auxlang is easy. Creating a good auxlang is not. It requires an understanding of linguistics, lots of research, constantly confronting one's own biases and a good tolerance for compromise. This essay is not a substitute for any of those, but it can help in getting one to the right mindset.
An IAL can be judged by many different metrics, and it must succeed in all of them to be considered worthwhile. This is a difficult task, as many of the goals are in direct conflict, and a language that shines in one aspect will necessarily be worse in some other.
Still, not all languages are made equal. It is possible for an IAL to be worse than some other contender in all categories. In this situation a concept from engineering called the Pareto frontier applies. The Pareto frontier in this context is the set of all possible languages which are better than any other in at least one aspect (not necessarily the same aspect for every contender), or all languages for which there isn't an alternative that is just better in every way. Finding the Pareto frontier is an efficient strategy for reducing the options one has to think about, as without any other information, all options on the Pareto frontier are equally good.
Here are the six metrics to consider, in no particular order:
An IAL should not arbitrarily privilege any language, area or people, especially not those who are already privileged. The language should be as easy whether your native language is Hindi, Arabic, Mandarin, English or Swahili.
An IAL should be practical for use, whether scientific, literary or everyday. It should provide the means to speak efficiently, precisely and expressively, and it should be easy to write and display both digitally and physically.
An IAL should align with what people are familiar with, and it should feel as intuitive and natural as possible. If most languages do something one way, it should follow the lead.
An IAL should be no more complicated than necessary. Exceptions to rules presented should be minimal and well-justified.
All of the rules of an IAL should be clearly laid out. There should be a well-defined standard way of speaking the language, and acceptable variation from the standard should be similarly well-defined.
An IAL should be appealing both visually and auditorily. It should have its own style and identity, and not simply feel like a pidgin or a broken version of some other language.
These are by no means definitive, but I would expect them to be mostly uncontroversial.
Elaborating Definiteness
Many auxlangs abhor strict rules. They'll say you can speak how you want, the most important thing is to be understood. This is true of all languages by default, but it doesn't make it any easier to understand when someone speaks differently from what you are used to.
Understanding a language fluently is an exercise in automatic pattern-recognition, and is easier if those patterns are simple, few in number and consistently applied. When we see something that doesn't fit the patterns, it breaks the flow, and switches the mind from effortless understanding to deciphering, even if it doesn't cause ambiguity. A good example of this is mixing up "your" and "you're": it is very jarring, even if there are very few situations where it isn't obvious which is meant.
Grammatical rules are not shackles limiting expression, but scaffolding to facilitate it. This is why prescriptive standard versions of languages exist: to enforce a standard set of patterns that is fluently understood by all speakers of the language, no matter their home dialect. It allows everyone to understand each other, instead of trying to decipher each other. The standard variety of an IAL should similarly be a prescriptive, fully defined language.
Sources
There are going to be some references to the top 10 languages by native speakers in this essay. Around 46% of people speak one of them as their native tongue. The languages are:
Mandarin, Spanish, English, Hindi, Arabic, Portuguese, Bengali, Russian, Japanese, Punjabi
Referring to these languages is a simple check to avoid centering too much on Europe, and handling a manageable number of languages makes it easier to research statistical information. These ten languages are a diverse sample of Eurasian languages (also covering the Americas and Australia, due to colonialism).
Still, referring back to these particular languages isn't without its own bias. The list contains three languages spoken in India (Hindi, Bengali and Punjabi), and excludes the myriad of medium-sized languages of Sub-Saharan Africa completely (the first one on the list would be Hausa, at #36). Also, the list includes both Spanish and Portuguese, which are rather similar.
Any numbers or statistics are Wikipedia, Phoible and WALS.
The main dilemma any IAL phonology has to face is between ease, how easy the phonology is to speak and understand for speakers of different backgrounds, and density, how much information a single syllable contains. Focus too much on ease, and the words become long and cumbersome. Focus too much on density, and the language requires extensive study and practice to pronounce correctly.
Other factors are how well the original pronunciation of the borrowed words can be preserved, which improves with a more permissive phonology, and how well the language fits into its chosen writing system, which is generally easier with smaller phonemic inventory.
The Phonemic Inventory
For any sound the human voice can produce, there exists a language that doesn’t use it, and therefore people who have to learn it. Therefore a phonetic inventory that absolutely anyone can pronounce without learning any new phonemes is impossible. Still, some sounds are more common than others: All of /m k i a j u p w n/ occur in more than 80% of languages.
I posit that every phonetic inventory on the Pareto frontier consists of some number of the most common phonemes cross-linguistically, with perhaps minor modifications and weightings. This leaves size as the main variable. Probably the minimal practical inventory is found in Toki Pona, with nine consonants (/m n p t k s w l j/) and five vowels (/a e i o u/). The global average is 24 (±2) consonants, but an IAL should probably have fewer.
An uncontroversial example inventory comprised of the most common sounds looks somewhat like this:
/m n p t k b d g/
/w l j s h/
/a i e o u/
Using the very common five-vowel system and thirteen consonants, this inventory is very similar to that of Japanese (not counting allophones, without /z/ and with /l/ instead of /r/).
Should one want to expand this inventory, the next most common phonemes in order are /ŋ ɲ f t͡ʃ ʔ ʃ r z/, with the least common sound (/z/) being found it 31% of languages.
IAL inventories normally tend to include the "whatever-rhotic" /r/, pronounced [r~ɾ~ɹ~ɻ~ʀ~ʁ~χ]. I would advise against this. While many languages have a "rhotic", there is very little in common between them, and what sounds like a rhotic to one person might read as something completely different for another. Some languages, such as Japanese and Korean, don't distinguish /l/ and a rhotic at all. The alveolar trill (the most common rhotic) is also a difficult sound to learn, even if it is in one's native language.
Excluding the rhotic (or preferably [r] in the standard) isn't that clear-cut. Leaving it out makes it impossible to use some highly recognizable roots, such as "tri", in the basic vocabulary. An option is to include ⟨r⟩ in the orthography, pronounced /l/, but that is rather inelegant.
When we compare the inventory to the top ten languages by native speakers, some conflicts can be found:
A few of these can be answered by adding free variation allophones:
Arabic speakers just have to adapt. The phonemes it is missing are very common, and sacrificing them would significantly hurt both density and loanword recognizability. It isn't as bad as it sounds: Arabic already has them in loanwords, although they tend to be merged into their closest equivalents.
A dialectal pronunciation of /p t k b d g/ as [pʰ tʰ kʰ p t k] takes care of the languages that distinguish plosives only by aspiration, although distinguishing plosives by voicing is common enough (at least 57% of languages have most of /p t k b d g/), that simply not bothering is an option. Dialects are explored further in their own section.
Still, compromises such as these have to be made. An inventory both Mandarin and Spanish speakers can pronounce already looks rather minimalist: /m n p t k f s x l w j a e i o u/.
Diphthongs & Hiatus
Allowing sequences of vowels is very useful in borrowing words from all over the world. Diphthongs, however, are less common and usually a restricted set. An IAL should probably not have true diphthongs at all, instead opting for hiatus (pronouncing a vowel sequence as distinct syllables), which is pronounced as diphthongs in fast speech.
More than two consecutive vowels (as in Japanese aoi) and two of the same vowel in a row (as in Japanese ee) should probably be avoided.
Phonotactics
Native speakers of some languages (such as Hawaiian) have difficulty with any syllables more complex than (C)V. Taking this as the limit, the language would have in total only 70 different syllables, which is would mean long words and unrecognizable loanwords, so loosening the rules a bit could be acceptable.
69% of the languages on WALS allow at most (C)(w,j,l,r)V(C), so that is a good limit on how complex a syllable to allow in an IAL. Still, it would not hurt to cut that down a little: for example, Mandarin and Arabic don't allow consonant clusters in the onset (although the semivowels /w/ and /j/ are fair game). Japanese allows only /n/ in the coda, and native Spanish vocabulary allows only /n, l, r, s/.
So a possible syllable structure for this inventory could look like this:
(C)(w,j)V(N)
Where C is any consonant apart from /w, j/, V is a vowel, and N is one of /n, l, s/
Possible syllables would then look like /a/, /mu/, /in/, /sos/ or /kwel/. In total, there would be 720 possible syllables.
Other phonotactic constraints one might consider are forbidding gemination (two of the same consonant in a row) and/or forbidding /ji, wu/, which are difficult to distinguish from /i, u/ (which limits the possible syllables to 608).
Phonotactics are however rarely monolithic: common, native words fit them exactly, while recent borrowings and proper nouns are left unchanged, and a lot of vocabulary is found between the two extremes. Preferably, all of the rules used to cram loanwords into the phonotactics should be explicit and consistent, perhaps strictly divided into levels of nativization.
Allophony
The main function of allophony in an IAL is to make words easier to pronounce, generally through assimilation, where different sounds become more similar when close by.
A very common allophonic rule is nasal assimilation, where a preceding nasal assimilates to the place of articulation for a following sound, especially a stop:
/nk, ng/ > [ŋk, ŋg]
/np, nb, nm/ > [mp, mb, mː] (should probably be reflected in the orthography)
You are probably doing this without noticing it (try saying "handbag" quickly; it'll end up sounding like "ham-bag").
Not every language agrees on what is acceptable allophonic variation: Russian doesn't assimilate its nasals, and Japanese turns /h/ into [ɸ] (similar to f) before /u/. Still, any language requires a workable system of permitted allophony, especially for rapid speech. If such a system isn't provided, the early speakers will carry over the system from their native languages, and the later learners have to learn it by feel.
Suprasegmentals & Prosody
Phonemic tone is common in Africa and South-East Asia (found in 41.7% of languages on WALS), while phonemic stress is rare, only found in 7.6% of languages. Both are rather tricky to learn to someone unfamiliar to them, so the most logical option is to have neither.
In addition, while a slight majority of languages have variable stress, simple fixed stress is much easier. The big two in fixed stress are penultimate stress and initial stress, of which penultimate stress is slightly more common, while initial stress is useful in distinguishing word boundaries.
Also, remember that intonation and prosody more generally is a thing that languages have. I can't really help with them.
Choice of script
On the world stage, no script can compete with the Latin alphabet. While only three of the top ten languages are written with it, all of the rest are written with different scripts. Alphabets are also superior in their ability to represent arbitrary sequences of sounds, which is very useful in transcribing words that might not exactly fit the phonotactics.
While the Latin alphabet is not perfect (⟨b d q p⟩ are a pain for dyslexics), and a purpose-built script might be better in principle, the switching costs of teaching everyone to read again and overturning years of technological tradition simply won't be worth it.
Capitalisation
While most scripts have only a single form for each grapheme, the Latin Alphabet has two: the uppercase form and the lowercase form. The simplest option would be to completely abandon one or the other, which would require learning only one form for each letter. As the capital letters are associated with SHOUTING, just using the lowercase letters would be preferable.
Here aesthetics come into play: no modern language that uses the Latin Alphabet actually does that, and not using capitalisation can make a language look childish and cutesy. While this would require all the learners to learn two forms of the letters and to learn new rules for capitalization (they are not universal: "Tuesday" and "English" are tiistai and englanti in Finnish), the aesthetic benefits might be worth it.
Spelling
Making a system of spelling for an IAL is relatively straightforward, compared to other aspects. Following a few guidelines will go a long way.
This is almost so obvious it doesn't need to be said. It should be possible to figure out the pronunciation from the written form of any nativized word, and vice versa, and a one-to-one correspondence is the most elegant way to achieve that. This rule can and should have exceptions, but they should all be well-founded and consistent.
Use the letters that most unambiguously represent the phonemes. Usually this means simply using the IPA letter for the sound (⟨k⟩ for /k/), but in the case of /j/, ⟨y⟩ is actually the more common grapheme with ⟨j⟩ more often representing /ʒ/ or /d͡ʒ/.
The modern world is based on technology, and if the language can't be written with the characters of a standard QWERTY keyboard, it has failed in practicality. Digraphs should be preferred over diacritics in all situations.
Stray away from apostrophes. Also, some added complexity is allowed in order not to make the language look better, for example ⟨ch⟩ instead of ⟨tsh⟩ when nativising /t͡ʃ/.
Writing /hidloelektlik/ as ⟨hydroelektrik⟩ makes it much easier to decipher, while adding a limited amount of complexity.
A thing to keep in mind is that you don't only need an orthography for your basic ("native") inventory, but also for all of the phonemes only occurring in loanwords.
An unique script
As outlined above, an unique script would be generally a bad idea. It is still interesting. Here are some things to consider:
Notice that a script doesn't necessarily have to be an alphabet. There are pretty much five types of scripts in use: logographies (one symbol per word), syllabaries (one symbol per syllable), abjads (one symbol per consonant, vowels unwritten), abugidas (one symbol per consonant, vowels marked by diacritics) and alphabets (one symbol per sound).
Of these, logographies with their tens of thousands of characters are immediately out, as are abjads, which divorce spelling from the pronunciation by their nature. The choice between the rest depends on the structure of the language: syllabaries for languages with generally very simple syllables, alphabets for those with complex ones, and abugidas for an in-between version.
Also, don't forget to design punctuation, and ways of emphasising words, among other typological tricks. This writing system will have to be generally better than the Latin Alphabet, which has been modded a lot over the millenium, to combat its inherent downsides.
Word order
The most common basic word order is Subject-Object-Verb, found in 41% of languages as the primary order. Of the ten largest languages, four are SOV; the three languages found on the Indian subcontinent and Japanese. Most of the rest are Subject-Verb-Object, the second most common order with 35% of languages, including the top three (Mandarin, Spanish, English). The Middle East is predominantly SOV, but Modern Standard Arabic is VSO, with a lot of its dialects tending to SVO. Both Europe and Africa are predominantly SVO.
The choice between the two dominant orders is one of priority: SOV is more common among smaller languages, while the heavyweights tend towards SVO. SVO is the more boring and justifiable option, although an IAL that could be criticised for privileging speakers of Hindi and Japanese is an entertaining thought.
Of course, other orders can also be considered: VSO might be a reasonable compromise between the two common orders, equally difficult for everyone while still being reasonably natural. What no IAL should have is a primary order with the object before the subject: the three possible orders combined are found in under 3% of languages.
One might ask, why not have free word order? Wouldn't everyone be happy that way? The problem is that free word order in a literal sense doesn't exist. Flexible word order does. In flexible word order, there is generally a neutral, dominant order, and other orders all have subtle but important differences in meaning, often related to information structure. Even in the few cases where two or more orders which are equally common, they are either determined by purely grammatical factors (such as the verb-second order of many Germanic languages) or by information structure. Any gaps of free choice, such as meaning-neutral variation in word order, will swiftly be filled by the habits of the early speakers.
This doesn't mean that there shouldn't be any way to shuffle around the constituents; there very much has to be to allow for topicalization and emphasis. But English, for example, works well with just a passive construction to move the subject and object from their canonical places. Adverbials can be moved more freely, although they too ought to have a standard neutral place and order.
The order of everything else in the language depends (but isn't determined by) the main order. For example, SVO languages have a slight tendency to be head-initial (modifiers follow nouns like cat black that, prepositions instead of postpositions, etc.), although many large SVO languages (including English and Mandarin) have the modifiers precede the noun, and yet others (such as Spanish) have some following and others preceding. SOV languages on the other hand tend to be heavily, sometimes strictly, head-final, which could be a very elegant option grammatically.
Even more important than how common certain noun-modifier orders are is how regularly they are applied: a learner shouldn't have to keep in mind which words come before which other words. Few exceptions can be made for categories such as numerals or relative clauses, but the fewer lists of words people have to learn that are treated differently the better.
Also, one small syntactic fact: Moving content question words to the front of the sentence is a mostly European phenomenon (although it is also common among minority languages in the Americas and Australia). Instead, the question words are generally left where the corresponding constituent would naturally fall (You ate what? You are coming when?).
Relative clauses
Relative pronouns that are marked for their role in the relative clause (such as English who/whom or preposition + whom/which) are almost completely a European phenomenon. Here are some of the most common strategies:
Relative pronoun strategy (English, Russian, 11%):
The man who saw me - The man whom I saw - The stick with which Jack hit the wolf
Gap strategy (Mandarin, Japanese, 49%):
The man that saw me - The man that I saw - The stick that Jack hit the wolf
Pronoun-retention strategy (Arabic, Hausa, 15%)
The man that (he) saw me - The man that I saw him - The stick that Jack hit the wolf with it
(Almost all pronoun-retention languages drop the pronoun in the subject, as it is redundant)
Correlation strategy (Hindi), subtype of the nonreduction strategy (11%)
Which man saw me, he… - I saw which man, he... - Jack hit the wolf with which stick, it…
The main competitors are the overwhelmingly common gap strategy, and the highly flexible pronoun-retention strategy. The pure gap strategy is dense and simple but ambiguous with anything further than the object, unless complicated somehow, while the pronoun-retention strategy adds a small amount of extra bulk for a lot of clarity. An option is to do what Mandarin does and allow either including or not including the prepositional phrase with the retained pronoun (this is called a resumptive pronoun).
There are usually limits on how far buried the relativized element can be, but the pronoun-retention strategy allows one to go absolutely nuts if they want to: The woman that everyone but I was slower than her and her friend. Obviously these kinds of sentences should in general be avoided, but they are nice to have if necessary.
Relative clauses have a strong tendency to follow their head noun.
Inflectional
In an ideal IAL, very few grammatical categories should be marked mandatorily. A speaker should be able to say "cat eat mouse" to mean "(A/the) cat(s) ate/eat(s)/will eat (a/the) mouse/mice", much like in Mandarin or Japanese. Of course, it is important to be able to mark these when necessary, but they are often superfluous.
This is easiest to see with sentences like "three cats visited yesterday", where marking for number and tense is simply redundant. Also, in sentences like "uranium is radioactive" a clearly timeless statement is given the present tense, and marking plurality necessitates a grammatical distinction between count and mass nouns (which all languages don't agree on), a complication that can be avoided by just not marking it.
Another factor to consider is how these optional grammatical factors are marked. Morphologies are generally split into isolating (categories marked with separate words), agglutinative (one category marked with each affix) and fusional (multiple categories merged into one affix). Of these, the fusional strategy with its opaque grids of affixes should clearly be avoided, unless the added density counteracts the added complexity. Agglutinative languages neatly pack categories into dense, easily analysable words, but stumble when forced to add affixes to irregular words or more complex expressions. The best bet for an IAL is likely an isolating morphology. It is also common for pidgins to be largely isolating, suggesting that it is in some sense more natural.
The simplest and most elegant way to treat for example tense and plurality markers is to treat them exactly like all other modifiers. Tense markers can be simply adverbs or auxiliary verbs, and plurals simply quantifiers, which are themselves not distinguished from
adjectives. One cat, two cat, multiple cat.
Although I have chattered about tense, remember that marking aspect is even more common, especially the imperfective (description) vs. perfective (narrative) distinction, that in English is tangled into the simple-progressive split. Also remember that all aspects (and tenses) should be able to be expressed by some construction, and it is a good idea to explicitly design those constructions, at least for the most common ones.
Derivational
Derivational morphology is distinguished from inflectional morphology by the fact that it produces wholly new lexemes (words), that have a meaning not wholly determined by their components, especially if semantic drift has occured. For an IAL, a robust derivational morphology is almost a given, because of its usefulness in expanding the vocabulary.
Sometimes a dedicated affix might not be necessary. Why have affixes for "-place" or "-person", when it would be simpler just to make a compound with the word for "place" or "person"? The less categories that are treated differently the better.
Also, compounds should either always be combined into one word (treehouse) or not (fire engine). Any alternation between the two will lead to jarring and persistent mistakes.
Structure
Some auxiliary languages (especially those designed as a hobby) aim to have very few words, which can then be combined to speak about everything. More formal IALs tend to simply mimic their source languages, and have a word for pretty much everything. The question is, why not both?
An IAL could be designed to have a "bootstrap vocabulary" of 800-1200 words, which if well designed could be a simple but functional language in itself, allowing the learner to start using the language as fast as possible. It would consist of grammatical words (and, of), semantic primitives (any, have) and basic content words (house, bird, red).
When choosing the definitions for the content words, widening the meaning is an easy way to reduce the number of words one might need; not distinguishing "house" and "building" would not hurt. Still, distinctions occur in languages because they are useful; if the language doesn't distinguish "mother" from "father", people will have to constantly say "female parent" and "male parent".
Then, outside the core vocabulary there are more precise categories, such as the names of specific animals, difficult-to-express near-synonyms and roots for common words with overlong compounds. These can all be defined (or at least explained) using the original core vocabulary. Outside this middle vocabulary is the technical and culturally specific vocabulary, not strictly necessary for a person to be fluent in the language, much like loanwords.
Common words
Some words are truly international, words like "kilogram", "helicopter" or "zombie". These words however tend to be very technical or culturally specific. For most of the common, everyday concepts, each language has their own words.
While taking their common words from Latin and the Romance languages has been the go to method for IALs, its results are underwhelming. While such languages are easier to learn for speakers of Spanish (#2) and Portuguese (#6) (and in part, English (#3)), speakers of Mandarin (#1), Hindi (#4) and Arabic (#5) get little benefit. Also, any language with common vocabulary taken from a single language family loses any pretense of true universality.
A few languages (such as Sambahsa) try to appeal to a wider demographic by taking their vocabulary from Proto-Indo-European. While Indo-European languages have a wide reach, they only cover 8 of the top 20 languages by native speakers, which is still not even close to universal. Also, using a language that is over 4500 years old as the basis of the vocabulary has the added drawback of making the vocabulary at most as recognizable for anyone as Hindi words are to English speakers.
Some basic vocabulary should be taken from common parts of Latin and Greek scientific terms (pre-, post-, neo-, psycho-), but the recognizability on those too drops quickly (coleoptera "beetle"), and including the less recognizable roots would just be using Romance vocabulary, with some Greek mixed in. Some common words are also international ("school" is a cognate in a bunch of languages).
The three strategies that could actually work are amalgam (using words from all over the planet equally), average (using words that are hybrids of the words from multiple languages) and a priori (just making the words up).
The a priori strategy (used by Kotava, for example) neatly sidesteps any problems with neutrality, making it exactly as difficult for everyone, as all of the words are arbitrary. It also allows the basic phonology to be simpler without interfering with recognizability, because there is nothing to recognize. However, this also makes words somewhat more difficult to learn, as the learner doesn't have any mnemonic hooks for them. Such mnemonic hooks could be added artificially, through a system of sound symbolism or some other explicit mnemonic structure.
The average strategy (used by Lojban) forms words by averaging and hybridizing the words from multiple languages, in an attempt to create a word that is somewhat familiar for everyone. This strategy holds promise, as it embeds mnemonic hooks for the speakers of the source languages in every word. Still, averaging words from multiple unrelated languages tends to produce something that could as well be made up: for example, Lojban pruxi, from Mandarin "鬼" (guǐ), English "spirit", Hindi "प्रेत" (pret), Spanish "espíritu", Russian "дух" (dukh) and Arabic "روح" (rūh). This strategy is also not very effective with too restrictive a phonology. While this can be done algorithmically, putting some craft into it might give a better result.
The amalgam strategy (used by Pandunia) is my current favourite of the three big options. Picking from the right source languages (preferably locally international ones), one might make a vocabulary of which most people recognize 10%-20%. This is an advantage over the other two strategies, as neither has any true recognizability. Also, a learner will learn some words of major world languages if they pay attention to the etymologies of words. Still, a majority of the vocabulary would be completely unfamiliar, and perhaps more difficult to pronounce than a priori words.
The choice between the last three options (or any hybrid of them) comes down to ideology, but just picking a strategy isn't enough to make a good basic vocabulary. It is an art and a science to make it both maximally easy to learn and maximally easy to use, no matter how one goes about it.
Rarer words
Outside the basic vocabulary, universality increases, and choosing forms becomes easier; "sugar" is almost universally some descendant of Sanskrit śárkarā. Many words aren't as fortunate, however, and compromises have to be made.
One must also not forget the possibilities of derivation and compounding: there isn't really an universal root for "week", but a compound like "sevenday" might easily be recognized. Should one want a shorter word, a generic derived noun from "seven" could still function as a mnemonic hook, more universal than any borrowed word. Names of the days of the week should also similarly be based on their order to avoid following any one cultural tradition.
On the other hand, a complex, practical language can't work with just a thousand roots. The language should have an expressive vocabulary, with enough precision in both technical and poetic terms. Nobody wants Newspeak as the global lingua franca.
Having any kind of dialects in an auxlang is a rarity, but is an effective way of making sure as many people as possible can use the language in some form. The dialects don't have to be limited only to the phonology: alternative words that are easier to pronounce or more familiar for people of certain linguistic background could easily be added, as well as minor variants in grammar rules.
While in general allowing "up to the speaker" variation tends to lead to the formation of unstated rules and complexities people have to learn by absorption, creating well-defined, mutually intelligible areal dialects could ease the burden of universality. However, the main purpose of an IAL is to facilitate communication across the world, for which it needs to be similar enough everywhere that people can understand each other without problem. Having too divergent dialects would make learning the dialect close to useless, as the learner would still have to learn the standard dialect to be understood internationally.
More divergent dialects could be a sensible option, if each spans a large area with many similar languages. They could function as local auxiliary languages, while being intelligible everywhere. Still, this is a difficult balance to get right, and the dialects would likely turn out to be much more trouble than they are worth.
IAL creators are almost by definition utopists, dreaming of a better world. So, what concrete things can they improve upon in the language? And more importantly, what can't they?
Gender
People generally settle on a way to speak long before they stop being sexist. This results in language built on (and depending on your views of linguistic relativity, reinforcing) assumptions and prejudices that would better be abandoned. A great demonstration of this is the amazing "A Person Paper on Purity in Language" by Douglas Hofstadter, that imagines a language similarly divided not on gender, but on race.
An enlightened IAL should therefore be symmetric and indifferent with respect to gender: Males and females (and those not really fitting into either) should be treated equivalently, and gender should not have to be mentioned when it is irrelevant. Gendered pronouns are the most glaring example of this, especially without a natural-sounding neutral option. 67% of languages make do with just one pronoun for he/she.
Of course, this doesn't mean purging any and all gender distinctions from the language: having different words for mother and father, sister and brother or king and queen is very useful, as long as neutral words (parent, sibling, monarch) also exist. Also, preferably the words for different genders should be roughly equal in length.
Even if you don't personally think of this as an issue, it is rather simple to fix, and it matters a lot for many people.
Unambiguity
True ambiguity and misunderstandings derived from it are rather rare even in messy natural languages, because people are amazing at decoding context. That is why trying to stomp down on ambiguity should probably not be a high priority in an IAL.
Still, designing a vocabulary with no homonyms would not be that difficult, which immediately eliminates a lot of ambiguity, although there is no need to be overzealous. Also remember that words don't have to be identical to be confused, and that semantic distance matters: while having similar-sounding words for "purple" and "oak" is not a problem, the words for "purple" and "green" should definitely be distinct.
As a language that is perfectly syntactically unambiguous, Lojban is a good object of study. It however rates rather poorly on the simplicity and familiarity metrics (it is sometimes described as feeling more like spoken computer code than a human language), so a good auxlang probably won't emulate it to that degree.
A small thing that would be very useful is a distinction between first person plural inclusive (me and others including you) and exclusive (me and others, not including you), although only 32% of languages (only big one being Mandarin) make the distinction. This continues to show how there isn't really an evolutionary pressure for languages to be unambiguous.
What a language can't change
Our linguistic heritage carries around a lot of things that are very wrong in hindsight: guinea pigs are not from Guinea and are not pigs. This example is easy to fix, but some other terms are so ingrained in scientific vocabulary they can't be changed anymore: "oxygen", literally "acid-maker", stemming from an incomplete understanding of chemistry, is regrettably the most recognizable root for the element.
More ambitious changes would be even more laughable: don't even think about changing the calendar or the number system, or anything relating to mathematics for that matter. The presentation of the calendar, however, is dependent on the language; I suggest ISO 8601 (YYYY-MM-DD).
Language doesn't define the other cultural systems we use; it can only either support them or not.
A work of art is never finished, only abandoned. However, before abandoning their creation, an IAL designer should consider the often forgotten aspects of language. The list is by no means comprehensive, like any list cataloguing easily forgotten things.
The best way to capture the full richness of language is to translate different and varying things, as much as one can, with thought given to every aspect. This is general advice for anyone making any kind of language, especially a detailed one. Auxiliary languages tend to be a bit threadbare in their original design, left to be haphazardly expanded later by the community of speakers.
IALs are often maligned in conlanger circles, seen as boring and samey, lumped in the same category as standard fare Germanic and Romance languages. I largely agree. Still, I would say that the reputation auxlangs have is not due to any inherent problem with the premise, but with the kind of people that are drawn to them. Beginners.
In my opinion, the thing that makes a conlang good is the effort put into it. All types of languages can be made with little effort, or with a lot of effort, and it shows in the final product. Auxiliary languages are appealing because they seem both ambitious and easy for those just coming into the subject. As I hope you have seen, they are not easy by a long shot; nothing great is.
That difficulty should not drive people away from IALs: instead, I hope that it will inspire people to tackle the challenge with more dedication.
Changelog:
2018-08-07 18:20 UTC: Added the section about other choices of scripts
2018-08-08 17:00 UTC: Added the Bible quote
2018-08-09 08:30 UTC: Clarified the pronoun-retention strategy
2022-05-20: Spelling and grammar fixes